研究者詳細 - 嶋田　和孝

論文 - 嶋田　和孝

分割表示 >> ／全件表示146 件中 1 - 146 件目

Share What You Already Know: Cross-Language-Script Transfer and Alignment for Sentiment Detection in Code-Mixed Data 査読有り国際誌

Pahari N., Shimada K.

ACM Transactions on Asian and Low-Resource Language Information Processing 23 ( 7 ) 2024年07月

担当区分：最終著者記述言語：英語掲載種別：研究論文（学術雑誌）

Code-switching entails mixing multiple languages. It is an increasingly occurring phenomenon in social media texts. Usually, code-mixed texts are written in a single script, even though the languages involved have different scripts. Pre-trained multilingual models primarily utilize the data in the native script of the language. In existing studies, the code-switched texts are utilized as they are. However, using the native script for each language can generate better representations of the text owing to the pre-trained knowledge. Therefore, a cross-language-script knowledge-sharing architecture utilizing the cross-attention and alignment of the representations of text in individual language scripts was proposed in this study. Experimental results on two different datasets containing Nepali-English and Hindi-English code-switched texts, demonstrate the effectiveness of the proposed method. The interpretation of the model using the model explainability technique illustrates the sharing of language-specific knowledge between language-specific representations.

DOI： 10.1145/3661307

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85199966141&origin=inward

Towards Human-Level Evaluation: Assessing the Potential of GPT-4 in Automated Evaluation and Feedback Generation on Japanese Essays 査読有り

Sayaka Nakamoto, Yoshiaki Okamoto, Takashi Nakakouchi, Kazutaka Shimada

Proceedings of the 16th International Conference on Learning Technologies and Learning Environments 2024年07月

Automatic Few-shot Selection on In-Context Learning for Aspect Term Extraction 査読有り

Koki Imazato, Kazutaka Shimada

Proceedings of the 18th International Conference on E-Service and Knowledge Management, 2024年07月

Dataset Construction and Verification for Detecting Factual Inconsistency in Japanese Summarization 査読有り

Keisuke Iwamoto, Kazutaka Shimada

Proceedings of the 15th International Conference on Smart Computing and Artificial Intelligence 2024年07月

Share What You Already Know: Cross-Language-Script Transfer and Alignment for Sentiment Detection in Code-Mixed Data 査読有り

Niraj Pahari, Kazutaka Shimada

ACM Transactions on Asian and Low-Resource Language Information Processing 2024年07月

An utterance is enough to the gaze? Gaze detection from utterance information in multiparty discussion 査読有り

Kensho Wakita, Kazutaka Shimada

Proceedings of the 6th International Conference on Activity and Behavior Computing 2024年05月

Simile Identification with Pseudo Data Acquisition and Re-labeling 査読有り

Jimi Jintaro, Shimada Kazutaka

International Journal of Smart Computing and Artificial Intelligence ( 特定非営利活動法人国際応用情報学研究機構 ) 8 ( 1 ) 1 2024年05月

Layer Configurations of BERT for Multitask Learning and Data Augmentation 査読有り

Pahari N., Shimada K.

Journal of Advanced Computational Intelligence and Intelligent Informatics ( 富士技術出版株式会社 ) 28 ( 1 ) 29 - 40 2024年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（学術雑誌）

Multitask learning (MTL) and data augmentation are becoming increasingly popular in natural language processing (NLP). These techniques are particularly useful when data are scarce. In MTL, knowledge learned from one task is applied to another. To address data scarcity, data augmentation facilitates by providing additional synthetic data during model training. In NLP, the bidirectional encoder representations from transformers (BERT) model is the default candidate for various tasks. MTL and data augmentation using BERT have yielded promising results. However, a detailed study regarding the effect of using MTL in different layers of BERT and the benefit of data augmentation in these configurations has not been conducted. In this study, we investigate the use of MTL and data augmentation from generative models, specifically for category classification, sentiment classification, and aspect-opinion sequence-labeling using BERT. The layers of BERT are categorized into top, middle, and bottom layers, which are frozen, shared, or unshared. Experiments are conducted to identify the optimal layer configuration for improved performance compared with that of single-task learning. Generative models are used to generate augmented data, and experiments are performed to reveal their effectiveness. The results indicate the effectiveness of the MTL configuration compared with single-task learning as well as the effectiveness of data augmentation using generative models for classification tasks.

DOI： 10.20965/jaciii.2024.p0029

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85184850065&origin=inward

Dataset Construction and Evaluation for Aspect-Opinion Extraction in Bangla Fine-Grained Sentiment Analysis 査読有り

Al-Mahmud , Shimada K.

Lecture Notes in Networks and Systems 818 437 - 449 2024年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In the present age, increasingly more people and entities post their opinions on web-based platforms. With this increasing trend of web-based opinion content, it is quite impossible to digest it by hand. Consequently, a computerized system for opinion and sentiment analysis is required. Sentiment analysis builds systems that automatically extract useful information from massive online reviews. Conventional sentiment analysis simply classifies a document/text into positive/negative/neutral polarity on the basis of overall sentiment polarity. In this case, it is considered that one sentiment is carried for the whole text, which may not be the case in the practical scenario. It is possible to appear more than one aspect, opinion, and sentiment in a single text, especially when the text is long. This motivates the introduction of fine-grained sentiment analysis, in which one can identify what a person/entity is mentioning and how they feel about each aspect/target entity. Aspect-opinion term extraction is a crucial aspect-based sentiment analysis (ABSA). It is a kind of granular-level sentiment analysis. In this study, our target is aspect-opinion terms extraction in the Bangla language. However, no annotated Bangla dataset is available for this task. Hence, we obtained textual data from a publicly available Bangla restaurant dataset, annotated them, and prepared a novel dataset. We then performed the task as a sequence labeling technique by utilizing our dataset. For the experiment, we employed vanilla transformers-based models such as mBERT, BanglaBERT, and BanglishBERT with linear layers and conditional random field (CRF) on top of them. To enhance the models’ performance, we combined different feature embeddings of mBERT, BanglaBERT, and BanglishBERT. We then utilized the linear layers and CRF on top of them. The experimental results indicated that the combined feature embeddings technique significantly improves models’ performance, and (mBERT & BanglishBERT) with linear layer performed best: 0.5840 on the F1 score among all the tested models.

DOI： 10.1007/978-981-99-7862-5_33

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85187643421&origin=inward

Towards Human-Level Evaluation: Assessing the Potential of GPT-4 in Automated Evaluation and Feedback Generation on Japanese Essays 査読有り国際誌

Nakamoto S., Okamoto Y., Nakakouchi T., Shimada K.

Proceedings - 2024 16th IIAI International Congress on Advanced Applied Informatics, IIAI-AAI 2024 156 - 161 2024年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In recent years, Automated Writing Evaluation (AWE) has been extensively researched within the field of AI in education. This paper explores generative AI, such as GPT-4, which has garnered significant attention for its ability to score essays and provide feedback to students. We designed prompts for GPT-4 to assign scores and rationales based on a given rubric and to generate feedback beneficial for students' development. We compared the evaluations produced by GPT-4 with those made by human evaluators. The results demonstrate GPT-4's potential to assist in generating evaluations at a human level. In addition, we analysed the consistency of the scoring and the quality of the rationales and feedback generated by GPT-4. In this paper, we will share our analysis and also describe the points that need to be improved for implementation in practice.

DOI： 10.1109/IIAI-AAI63651.2024.00039

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85208091336&origin=inward

Dataset Construction and Verification for Detecting Factual Inconsistency in Japanese Summarization 査読有り国際誌

Iwamoto K., Shimada K.

Proceedings - 2024 16th IIAI International Congress on Advanced Applied Informatics, IIAI-AAI 2024 243 - 248 2024年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Abstractive document summarization is one of the most important tasks in natural language processing. Many approaches based on large language models have been proposed. However, it is known that the output of LLM often includes hallucinations, such as factual inconsistency. Therefore, detecting factual inconsistencies in a summary is one important task for summarization. One solution for the detection is to utilize machine learning techniques. In general, machine learning approaches require a large number of training data to generate a robust model. However, it is difficult automatically to collect article-summary pairs with factual inconsistency from the Web because hand-written summaries on the Web are usually correct. Moreover, some existing datasets are written in English. In this paper, we propose some approaches to construct Japanese datasets with factual inconsistency automatically. For this purpose, we utilize two approaches from previous studies: FactCC and SumFC. In addition, we propose a new approach to construct summaries with exaggerated expressions, as a variety of factual inconsistencies. We call the datasets JFactCC, JSumFC, and JExnoS. For JExnoS, we utilize a two-stage approach based on GPT-4 and BART for the generation of summaries with exaggerated expressions from correct article-summary pairs. We also verify the usefulness of each constructed dataset through an experiment about factual inconsistency detection with BERT.

DOI： 10.1109/IIAI-AAI63651.2024.00054

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85208094978&origin=inward

Automatic Few-Shot Selection on In-Context Learning for Aspect Term Extraction 査読有り国際誌

Imazato K., Shimada K.

Proceedings - 2024 16th IIAI International Congress on Advanced Applied Informatics, IIAI-AAI 2024 15 - 20 2024年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In this paper, we handle aspect term extraction with In-Context Learning (ICL) as the target task. ICL is a method for learning Large Language Models. Instead of updating the model's parameters, prompts are provided to guide it to perform a target task. While the strength of ICL is that the model does not need to undergo additional training, the prompt includes input-output instances that cause instability in accuracy. The instances are called few-shot. Hence, selecting appropriate instances has the most important role in ICL. For this purpose, we propose a selection method with active learning. Active learning is a method of selecting instances that are useful to the model for training from unlabeled data. We regard the active learning-based approach as a sub-task for the target task. We introduce two types of sub-tasks and evaluate the effectiveness of them in the target task.

DOI： 10.1109/IIAI-AAI63651.2024.00012

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85208144512&origin=inward

An Utterance is Enough to the gaze? Gaze Detection from Utterance Information in Multiparty Discussion 査読有り国際誌

Wakita K., Shimada K.

2024 International Conference on Activity and Behavior Computing, ABC 2024 2024年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Gaze information has an important role in conversation understanding. Generally, gaze detection is performed using a camera. However, such systems often require expensive equipment. To solve this problem, we propose a method of eye detection using an utterance during conversations. We first annotate an existing discussion corpus with gaze information and then analyze the annotated data. We apply machine learning to construct a gaze detection model. We utilize BERT, a large-scale pre-Trained language model, and CART, a traditional machine-learning approach based on a decision tree. Experimental results show that the CART-based model was more accurate than the BERT-based model and show the potential possibilities of gaze detection from utterances.

DOI： 10.1109/ABC61795.2024.10651736

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85203829893&origin=inward

Dataset Construction and Evaluation for Aspect-Opinion Extraction in Bangla Fine-grained Sentiment Analysis 査読有り国際誌

Al-Mahmud, Kazutaka Shimada

Proceedings of the 4th International Conference on Data Science and Applications 2023年07月

Disfluency detection with context information from real utterances and generative utterances 査読有り国際誌

Hiroto Nakashima, Kazutaka Shimada

Proceedings of 13th International Conference on Smart Computing and Artificial Intelligence 2023年07月

Detecting speech recognition errors using topic information and BERT 査読有り国際誌

Masaaki Yokoyama, Kazutaka Shimada

Proceedings of 13th International Conference on Smart Computing and Artificial Intelligence 2023年07月

Combination and Knowledge Extension of Pre-trained Language Model for Offensive Language Detection 査読有り国際誌

Zhenming Li, Kazutaka Shimada

Proceedings of 16th International Conference on E-Service and Knowledge Management 2023年07月

Handling Data Scarcity in Code-Mixing Using Learning-Based Approaches 査読有り国際誌

Niraj Pahari, Kazutaka Shimada

Proceedings of 16th International Conference on E-Service and Knowledge Management 2023年07月

Automated Scoring of Logical Consistency of Japanese Essays 査読有り国際誌

Nakamoto S., Shimada K.

Communications in Computer and Information Science 1831 CCIS 652 - 658 2023年07月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Automated essay scoring (AES) is to estimate the scores of essays automatically. Two types of AES models are commonly used: handcrafted feature-based and neural-based models. In this paper, we introduce AES systems based on the two types for evaluating the logical consistency of Japanese essays. In addition, to enhance the performance of models, we integrate the neural-based model with the handcrafted features: a hybrid AES system. In the experiment, we show the effectiveness of our hybrid AES system. Besides, most of our AES models obtained higher QWK scores than human evaluators.

DOI： 10.1007/978-3-031-36336-8_101

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85164955884&origin=inward

Demonstration of Effectiveness of Nativeness in Stepwise Learning by Performing Sentiment Analysis 査読有り国際誌

Al-Mahmud, Kazutaka Shimada

Proceedings of 2023 International Conference on Next-Generation Computing, IoT and Machine Learning (NCIM) 2023年06月

Dataset Construction and Opinion Holder Detection Using Pre-trained Models 査読有り国際誌

Mahmud Al-, Shimada Kazutaka

International Journal of Service and Knowledge Management ( 特定非営利活動法人国際応用情報学研究機構 ) 7 ( 2 ) 1 2023年01月

Rating Prediction of Multi-Aspect Reviews Using Simultaneous Learning 査読有り

Takeo M., Kawasaki S., Shimada K.

Proceedings of 2023 International Conference on Asian Language Processing, IALP 2023 358 - 363 2023年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

This paper proposes a method for rating prediction tasks in review documents. Existing models for predicting the rating scores assigned to each aspect were often developed separately. However, there is a relationship among aspects. Therefore, we apply a multi-task learning framework to our prediction models. We evaluate our model with a review dataset about a game software domain written in Japanese. Each review document contains six-level rating scores for seven aspects of the game software. We utilize BERT as the backbone model for the prediction. Our model simultaneously learns the parameters of seven BERTs for seven aspects. Hence, we name this method "Simultaneous Learning."It leads to the improvement of the performance of the prediction. In addition, we compare two types of input data for the rating prediction task: all sentences and selected sentences. Experimental results show the effectiveness of our simultaneous learning model in the multi-aspect rating prediction. Our simultaneous learning method obtained 0.713 on RMSE. It demonstrated a 0.046 improvement in RMSE, as compared with non-simultaneous learning. Further, we confirmed that our method based on simultaneous learning was effective in the case that the number of sentences related to an aspect is small.

DOI： 10.1109/IALP61005.2023.10337276

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85181775477&origin=inward

Multilingual Symptom Prediction by Simultaneous Learning Using BERT 査読有り

Zhu C., Pahari N., Shimada K.

Proceedings of 2023 International Conference on Asian Language Processing, IALP 2023 100 - 105 2023年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Natural language processing (NLP) techniques have become increasingly significant in the medical domain. However, the amount of relevant medical text data remains limited. In this work, we propose a BERT-based multilingual simultaneous learning (MSL) model for reducing the problem of scarce data. We evaluate the benefit of MSL on the NTCIR-13 MedWeb multi- label symptoms classification task. The results indicate that the MSL model performs slightly better than Single-Task Learning (STL) models. Additionally, it shows that the similarity between languages has an impact on the performance of the MSL model.

DOI： 10.1109/IALP61005.2023.10337285

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85181777849&origin=inward

Language Preference for Expression of Sentiment for Nepali-English Bilingual Speakers on Social Media 査読有り

Pahari N., Shimada K.

CALCS 2023 - Computational Approaches to Linguistic Code-Switching, Proceedings of the Workshop 23 - 32 2023年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Nepali-English code-switching (CS) has been a growing phenomenon in Nepalese society, especially in social media. The code-switching text can be leveraged to understand the sociolinguistic behaviours of the multilingual speakers. Existing studies have attempted to identify the language preference of the multilingual speakers for expressing different emotions using text in different language pairs. In this work, we aim to study the language preference of multilingual Nepali-English CS speakers while expressing sentiment in social media. We create a novel dataset for sentiment analysis using the public Nepali-English code-switched comments in YouTube. After performing the statistical study on the dataset, we find that the proportion of use of Nepali language is higher in negative comments when compared with positive comments, hence concluding the preference for using native language while expressing negative sentiment. Machine learning and transformer-based models are used as the baseline models for the dataset for sentiment classification. The dataset is released publicly.

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85185000192&origin=inward

Handling Data Scarcity in Code-Mixing Using Learning-Based Approaches 査読有り

Pahari N., Shimada K.

Proceedings - 2023 14th IIAI International Congress on Advanced Applied Informatics, IIAI-AAI 2023 7 - 12 2023年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

With the rise of multilingual populations, the use of structures and vocabularies from multiple languages in a single communication act is increasing. The act of communication can be written or verbal. This phenomenon of mixing components from multiple languages is called code-mixing or code-switching. The penetration of social networks has increased the occurrence of code-mixing. Even with the abundance of data, the scarcity of annotated data has limited the research on code-mixing in the natural language processing (NLP) community. Here, we aim to analyze learning-based approaches to overcome the scarcity of annotated data by fine-tuning various multilingual transformer-based pre-trained models by conducting experiments on Nepaliand Hindi-English language identification tasks. Multitask learning, which can learn from the related task and improve the main task, and self-supervised learning, which utilizes the unannotated data in the same domain, are studied in this paper. The results demonstrate that these approaches are beneficial in improving the performance of the transformer-based models for the scarcity of annotated data.

DOI： 10.1109/IIAI-AAI59060.2023.00012

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85183462866&origin=inward

Disfluency detection with context information from real utterances and generative utterances 査読有り

Nakashima H., Shimada K.

Proceedings - 2023 14th IIAI International Congress on Advanced Applied Informatics, IIAI-AAI 2023 462 - 467 2023年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Disfluency is one of the linguistic phenomenons that often occurs in speech. Recently, automatic speech recognition has improved the accuracy of its transcript. The perfect transcript contains filler, self-repairs, distluency, and so on, because they often occurs. We focus on distluency in speech recognition outputs. The distluency in the transcript can be noise for the downstream tasks that use the transcript. In this paper, we propose a BERT-based distluency detection model for dialogue utterance. BERT handles contextual information of a target utterance to detect distluency. However, dialogue utterances can be interrupted by other speakers in conversation. Hence, utterances in dialogue tend to be short and have less contextual information. Therefore, the context of the target utterances should be complemented. In this paper, we propose two types of complementing approaches. The complement is preprocessing for the input of BERT. The first complement approach utilizes real utterances around the target utterance. In this approach, we need to select sentences from real utterances, based on some conditions: e.g., the number of utterances and the same speaker or not. The second complement approach utilizes utterances generated by a language model. The language model is trained to generate continuous utterances from the input. The generated utterance is utilized for expanding the context of the input data. In experiments, utterances from the speaker of target utterances improve the distluency detection performance. However, generated context could not outperform the real context.

DOI： 10.1109/IIAI-AAI59060.2023.00095

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85183462855&origin=inward

Detecting speech recognition errors using topic information and BERT 査読有り

Yokoyama M., Shimada K.

Proceedings - 2023 14th IIAI International Congress on Advanced Applied Informatics, IIAI-AAI 2023 400 - 405 2023年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Detecting errors in speech recognition is important because these errors negatively affect the accuracy of downstream tasks. To detect recognition errors, we focus on context, label sequence, and topics. Words in a sentence contain the context. Context information is useful in detecting errors from speech recognition because the errors must be outside the scope of the context. Furthermore, recognition error detection is a sequential problem, namely a sequential labeling task. Therefore, handling a label sequence is useful for error detection. In addition, a target speech contains topics, such as a talk about music, sports, and so on. A word that does not match the topic is considered an error. Therefore, the topic information is also useful. In this paper, we propose a method based on BERT, CRF, and LDA from the viewpoints BERT is used for context recognition. Conditional Random Fields (CRF) are used for the sequence labeling task. Latent Dirichlet Allocation (LDA) is to capture topic information. In this study, we compared several combinations of them: BERT only, BERT-CRF, and BERT-LDA-CRF models. We evaluate the models with the Corpus of Spontaneous Japanese (CSJ) corpus. The experimental result shows the effectiveness of the integration of CRF and LDA into BERT.

DOI： 10.1109/IIAI-AAI59060.2023.00084

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85183463970&origin=inward

Combination and Knowledge Extension of Pre-trained Language Model for Offensive Language Detection 査読有り

Li Z., Shimada K.

Proceedings - 2023 14th IIAI International Congress on Advanced Applied Informatics, IIAI-AAI 2023 82 - 87 2023年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Nowadays, more and more offensive comments are posted on social media. Those offensive comments can seriously cause mental damage to other people. Therefore, those toxic and harmful comments should be detected timely and accurate. In this paper, we exploit the powerful pre-train language models (PLM) in detecting offensive language. We consider two PLMs in our paper: BERT and DeepMoji. We finetune their combination and evaluate the performance with two datasets: Ask FM and Curious Cat. We found that DeepMoji outperforms BERT. We analyze the result from two aspects and conclude that the task and data of pre-training are important to PLM. Seeing that BERT is less effective than DeepMoji, it is possible to improve the performance of BERT. We then propose a 'Knowledge Extension' method to improve the performance of the BERT model. We find that the performance of PLM with high-quality extensional knowledge can be improved significantly.

DOI： 10.1109/IIAI-AAI59060.2023.00026

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85183459713&origin=inward

Demonstration of Effectiveness of Nativeness in Stepwise Learning by Performing Sentiment Analysis 査読有り国際誌

Al-Mahmud , Shimada K.

2023 International Conference on Next-Generation Computing, IoT and Machine Learning, NCIM 2023 2023年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Nowadays, more people and entities express opinions on online platforms. With this increasing trend of online opinion mass information, it is impossible to analyze them manually. For this reason, an automatic approach to opinion and sentiment analysis is essential and useful in the real world. Several approaches based on machine learning techniques in sentiment analysis have been proposed. However, machine learning essentially needs massive data to generate a robust model. On the other hand, collecting the training data is costly, especially in low-resource language. To solve this problem, we introduce stepwise learning. Stepwise learning is a technique that improves the model capacity for a target task (i.e., main task) by exploiting auxiliary information from source tasks. It is effective when the data in the target tasks are inadequate. The target task of this paper is sentiment analysis for reviews in the Bangla language. Our proposed method consists of two stages: source and target tasks. For the source task, we train a transformers-based model, such as BanglaBERT, by a large dataset translated into the Bangla language; the original dataset is written in English. For the target task, we retrain and evaluate the model with a relatively small target dataset. We further investigate the effect of opinion expression style of writers between source and target data in stepwise learning. We call it 'nativeness.' The experimental result shows that nativeness has an important role in improving performance.

DOI： 10.1109/NCIM59001.2023.10212970

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85170528820&origin=inward

Combining Pre-Trained Language Models and Features for Offensive Language Detection 査読有り国際誌

Li Z., Shimada K.

Proceedings - 2022 13th International Congress on Advanced Applied Informatics Winter, IIAI-AAI-Winter 2022 5 - 10 2022年12月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Nowadays, people often express their abusive and offensive thoughts to others on social media easier. The abusive and toxic comments hurt others seriously. Therefore those abusive and toxic comments should be detected properly through natural language processing. In this paper, we focus on two types of features in offensive language: word-level and sentence-level fea-tures. We use lexicon-based and standard bag-of-words features as the word level. We introduce BERT-based and DeepMoji-based features as the sentence level. We apply the four features to a machine learning approach: support vector machines. We evaluate the method using the combinations of four features with a dataset, Curious Cat. The best F1 score was generated by the method with all features. This result shows the effectiveness of our proposed method. In addition, the experimental result indicates that DeepMoji generated from Twitter data is better than BERT which is generated from written language, for an offensive language detection task about social media data.

DOI： 10.1109/IIAI-AAI-Winter58034.2022.00012

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85160906880&origin=inward

Annotation and multi-modal methods for quality assessment of multi-party discussion 査読有り国際誌

Tsukasa Shiota, Kazutaka Shimada

Proceedings of the 36th Pacific Asia Conference on Language, Information and Computation 175 - 182 2022年10月

Multi-Task Learning Using BERT With Soft Parameter Sharing Between Layers 査読有り国際誌

Pahari N., Shimada K.

2022 Joint 12th International Conference on Soft Computing and Intelligent Systems and 23rd International Symposium on Advanced Intelligent Systems, SCIS and ISIS 2022 2022年10月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

The ability to learn from one task and pass the knowledge to another task is a remarkable job that is possible because of multi-task learning (MTL). It is helpful in the scenario where data is scarce. In this paper, we study the use of MTL in natural language processing (NLP) using BERT for sentiment classification, category classification, and aspect-opinion sequence classification. The Restaurant-ACOS dataset is split into two subsets, namely the sentiment-category data subset, for the first two tasks and the sequence classification data subset, for the third task. The BERT layers are grouped into top, middle, and bottom layers. These layers are either frozen, shared, or kept unshared for individual tasks. Experiments are conducted to find the best combination for the overall layers of BERT. The experiments demonstrate that MTL can improve the performance on these tasks when compared as a baseline method that uses single task learning. One of the best configurations was based on freezing on the bottom layers, sharing on the middle layers, and keeping on the top layers, respectively. In addition, we investigate the role of each layer via the size of layers. From the result, the individual layers capture the task-specific patterns whereas shared layers capture the domain knowledge.

DOI： 10.1109/SCISISIS55246.2022.10001943

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85146643824&origin=inward

Dataset Construction and Classification Based on Pre-trained Models for Opinion Holder Detection 査読有り

Al-Mahmud A., Shimada K.

Proceedings - 2022 12th International Congress on Advanced Applied Informatics, IIAI-AAI 2022 65 - 70 2022年07月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Nowadays, it is getting increased in the massive amount of internet users. People express subjective thinking (i.e., opinion) implicitly and explicitly on online platforms such as Facebook, Twitter, Amazon product reviews, etc. Opinion holders are the people or entities who express opinions implicitly and explicitly on the online platform. With the increasing trends of online opinion mass information, it is impossible to detect them manually. For this reason, an automatic approach for opinion holder detection is essential. Opinion holder detection is useful to detect specific person's/entity's concerns about a particular topic, product, or problem. Opinion holder detection consists of two steps: the presence of opinion holders in text and identification of opinion holders. In this paper, we focus on the first step, namely the presence of opinion holders in text. We handle this task as a binary classification problem: INSIDE or OUTSIDE. At first, we prepare a new English dataset for this task. Then, we apply two types of pre-trained models, BERT and DistilBERT, to the INSIDE/OUTSIDE classification task. BERT is a transformer-based pre-trained language model. DistilBERT is a small, fast, cheap, and light model based on knowledge distillation from the BERT architecture. As to the binary classification task, we employ a logistic regression model on the top layer of the pre-trained models. We compare the language models employed in our experiment in terms of the F1 score and accuracy. The experimental result shows that DistilBERT obtained superior performance among the models: 0.901 on the F1 score and 0.924 on the accuracy.

DOI： 10.1109/IIAIAAI55812.2022.00023

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85139566304&origin=inward

Pseudo data acquisition using machine translation and simile identification 査読有り

Jimi J., Shimada K.

Proceedings - 2022 12th International Congress on Advanced Applied Informatics, IIAI-AAI 2022 391 - 396 2022年07月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

The simile is a kind of figurative language. It expresses the target of the figurative language by using some typical phrases such as "like". It is important to distinguish whether the sentence is a simile or a literal for understanding a sentence. However, a large amount of data is required to generate a classifier by machine learning. Moreover, creating the dataset is costly. In this paper, we propose a pseudo dataset acquisition method for simile identification. We first construct a dataset of simile and literal sentences using machine translation. This process automatically generates pseudo simile and literal instances from three types of corpora. Then, we apply some machine learning approaches to the simile identification task. We compare Support Vector Machine, Naive Bayes, and BERT in the experiment. The experimental result shows the validity of the pseudo dataset as compared with a simple baseline. For the fine-tuning of BERT, our large pseudo training data were more effective than a small manual training data.

DOI： 10.1109/IIAIAAI55812.2022.00084

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85139568402&origin=inward

A Compression-Based Multiple Subword Segmentation for Neural Machine Translation 査読有り国際誌

Keita Nonaka, Kazutaka Yamanouchi, Tomohiro I, Tsuyoshi Okita, Kazutaka Shimada, Hiroshi Sakamoto

Electronics 11 ( 7 ) 2022年03月

記述言語：英語掲載種別：研究論文（学術雑誌）

In this study, we propose a simple and effective preprocessing method for subword segmentation based on a data compression algorithm. Compression-based subword segmentation has recently attracted significant attention as a preprocessing method for training data in neural machine translation. Among them, BPE/BPE-dropout is one of the fastest and most effective methods compared to conventional approaches; however, compression-based approaches have a drawback in that generating multiple segmentations is difficult due to the determinism. To overcome this difficulty, we focus on a stochastic string algorithm, called locally consistent parsing (LCP), that has been applied to achieve optimum compression. Employing the stochastic parsing mechanism of LCP, we propose LCP-dropout for multiple subword segmentation that improves BPE/BPE-dropout, and we show that it outperforms various baselines in learning from especially small training data.

DOI： 10.3390/electronics11071014

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85126910361&origin=inward

テキストアナリティクスとその先へ

嶋田和孝

情報・システムソサイエティ誌 ( 一般社団法人電子情報通信学会 ) 26 ( 4 ) 4 - 5 2022年02月

Recognizing a participant behavior in a multi-party conversation 査読有り

Shimada Kazutaka, Yonemitsu Shunsuke

Information Engineering Express ( 特定非営利活動法人国際応用情報学研究機構 ) 8 ( 1 ) 1 2022年01月

Corpus construction for topic-based summarization of multi-party conversation 査読有り国際誌

Yuri Nakayama, Tsukasa Shiota, Kazutaka Shimada

Proceedings of the International Conference on Asian Language Processing (IALP) 229 - 234 2021年12月

Sarcasm Detection with External Entity Information 査読有り

Xufei Xu, Kazutaka Shimada

Proceedings of the 10th International Conference on Smart Computing and Artificial Intelligence 2021年12月

Discussion Structure Prediction Based on Two-step Method 査読有り国際誌

Takumi Himeno, Kazutaka Shimada

Proceedings of Recent Advances in Natural Language Processing, RANLP 2021 543 - 551 2021年09月

Tell Me What You Read ? Automatic Expertise-Based Annotator Assignment for Text Annotation in Expert Domains 査読有り国際誌

Hiyori Yoshikawa, Tomoya Iwakura, Kimi Kaneko, Hiroaki Yoshida, Yasutaka Kumano, Kazutaka Shimada, Rafal Rzepka, Patrycja Swieczkowska

Proceedings of Recent Advances in Natural Language Processing, RANLP 2021 1579 - 1589 2021年09月

Relation Extraction using Multiple Pre-Training Models in Biomedical Domain 査読有り国際誌

Satoshi Hiai, Kazutaka Shimada, Taiki Watanabe, Akiva Miura, Tomoya Iwakura

Proceedings of Recent Advances in Natural Language Processing, RANLP 2021 535 - 542 2021年09月

Development and Application of Leader Identification Model Using Multimodal Information in Multi-Party Conversations 査読有り国際誌

Tsukasa Shiota, Kouki Honda, Kazutaka Shimada, Takeshi Saitoh

International Journal of Asian Language Processing 30 ( 04 ) 2021年05月

マルチモーダル情報を用いた複数人議論の品質評価

塩田宰, 嶋田和孝

人工知能学会研究会資料言語・音声理解と対話処理研究会 ( 一般社団法人人工知能学会 ) 91 ( 0 ) 22 2021年02月

Tell Me What You Read: Automatic Expertise-Based Annotator Assignment for Text Annotation in Expert Domains 査読有り

Yoshikawa H., Iwakura T., Kaneko K., Yoshida H., Kumano Y., Shimada K., Rzepka R., Swieczkowska P.

International Conference Recent Advances in Natural Language Processing, RANLP 1575 - 1585 2021年01月

記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

This paper investigates the effectiveness of automatic annotator assignment for text annotation in expert domains. In the task of creating high-quality annotated corpora, expert domains often cover multiple sub-domains (e.g. organic and inorganic chemistry in the chemistry domain) either explicitly or implicitly. Therefore, it is crucial to assign annotators to documents relevant with their fine-grained domain expertise. However, most of existing methods for crowdsoucing estimate reliability of each annotator or annotated instance only after the annotation process. To address the issue, we propose a method to estimate the domain expertise of each annotator before the annotation process using information easily available from the annotators beforehand. We propose two measures to estimate the annotator expertise: an explicit measure using the predefined categories of sub-domains, and an implicit measure using distributed representations of the documents. The experimental results on chemical name annotation tasks show that the annotation accuracy improves when both explicit and implicit measures for annotator assignment are combined.

DOI： 10.26615/978-954-452-072-4_177

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85123569558&origin=inward

Relation Extraction using Multiple Pre-Training Models in Biomedical Domain 査読有り

Hiai S., Shimada K., Watanabe T., Miura A., Iwakura T.

International Conference Recent Advances in Natural Language Processing, RANLP 530 - 537 2021年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

The number of biomedical documents is increasing rapidly. Accordingly, a demand for extracting knowledge from large-scale biomedical texts is also increasing. BERT-based models are known for their high performance in various tasks. However, it is often computationally expensive. A high-end GPU environment is not available in many situations. To attain both high accuracy and fast extraction speed, we propose combinations of simpler pre-trained models. Our method outperforms the latest state-of-the-art model and BERT-based models on the GAD corpus. In addition, our method shows approximately three times faster extraction speed than the BERT-based models on the ChemProt corpus and reduces the memory size to one sixth of the BERT ones.

DOI： 10.26615/978-954-452-072-4_060

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85123577323&origin=inward

Discussion Structure Prediction Based on a Two-step Method 査読有り

Himeno T., Shimada K.

International Conference Recent Advances in Natural Language Processing, RANLP 538 - 546 2021年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Conversations are often held in laboratories and companies. A summary is vital to grasp the content of a discussion for people who did not attend the discussion. If the summary is illustrated as an argument structure, it is helpful to grasp the discussion's essentials immediately. Our purpose in this paper is to predict a link structure between nodes that consist of utterances in a conversation: classification of each node pair into “linked” or “not-linked.” One approach to predict the structure is to utilize machine learning models. However, the result tends to over-generate links of nodes. To solve this problem, we introduce a two-step method to the structure prediction task. We utilize a machine learning-based approach as the first step: a link prediction task. Then, we apply a score-based approach as the second step: a link selection task. Our two-step methods dramatically improved the accuracy as compared with one-step methods based on SVM and BERT.

DOI： 10.26615/978-954-452-072-4_061

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85123585961&origin=inward

Corpus construction for topic-based summarization of multi-party conversation 査読有り

Nakayama Y., Shiota T., Shimada K.

2021 International Conference on Asian Language Processing, IALP 2021 229 - 234 2021年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In this paper, we report corpus construction and topic-based summarization methods for multi-party conversation. We have already constructed reference summaries and a list of important utterances in each discussion. However, fine-grained summaries about topics in a discussion often are desired in many situations. Therefore, we construct topic-based summaries and propose an important utterance extraction method and two summarization processes using the extracted utterances; extractive and abstractive methods. For the important utterance extraction, we use SVMs with 12 types of features. We use mBART, which is a neural network-based model, as the abstractive method. In the experiment, the extractive method was superior in terms of “accuracy as a summary (relevance),” while the readability of the abstractive method was superior.

DOI： 10.1109/IALP54817.2021.9675229

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85125168200&origin=inward

Sarcasm Detection Using Role Pairs: Bootstrapping Role Pairs and Weighting 査読有り

Hiai S., Shimada K.

2020 International Conference on Asian Language Processing, IALP 2020 235 - 240 2020年12月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© 2020 IEEE. Sarcasm detection has been treated as a task that classifies text as sarcastic or non-sarcastic. Sarcasm detection is a significant challenge for sentiment analysis because sarcasm involves a positive expression with negative meaning. Previous work focused on relation information between pairs of role expressions, such as 'boss and staff,' and proposed a sarcasm detection method based on surface and relation information. Although they showed the effectiveness of the relation information, due to the small scale of the list of role pairs, the effectiveness was limited to a small scale. In this paper, we attempt to improve the performance of the sarcasm detection method. We propose a role pair extraction method using a bootstrap method to obtain a larger role pair list than the previous work. In addition, we propose weighting methods for each role pair. Then, we evaluate the method using our role pair list and weighting methods by comparing the method using the list of the previous work. We find that we can obtain approximately threefold increase for role pairs. Although the effectiveness is limited, the experimental result shows a potential benefit because the method with topic similarity can reduce the influence of such noise pairs.

DOI： 10.1109/IALP51396.2020.9310499

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85099875523&origin=inward

Related Article Extraction Using Learning to Rank 査読有り

Shiota T., Shimada K., Nogami S., Fukuyama S.

2020 International Conference on Asian Language Processing, IALP 2020 50 - 55 2020年12月

担当区分：筆頭著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© 2020 IEEE. Related articles of a news article help readers to access other beneficial information efficiently. Thus, it is an important task for newspaper companies to link appropriate related articles to a news article on news websites. However, it costs to select related articles from enormous news articles that have been published in the past. To solve the issue, we introduce 'learning to rank' methods to the related article extraction task to support business operation. We propose some ranking methods and report the accuracy of each method with three evaluation metrics. We also show some input-output examples by the method that scores the highest accuracy in our experiments.

DOI： 10.1109/IALP51396.2020.9310463

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85099879320&origin=inward

Leader Identification Using Multimodal Information in Multi-party Conversations 査読有り

Shiota T., Honda K., Shimada K., Saitoh T.

2020 International Conference on Asian Language Processing, IALP 2020 7 - 12 2020年12月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© 2020 IEEE. It is one of the important tasks to predict a participant's role in a multi-party conversation. Many previous studies utilized only verbal or non-verbal features to construct models for the role recognition task. In this paper, we propose a model that combines verbal and non-verbal features for leader identification. We add non-verbal features and construct our prediction model with utterance, pose, facial, and prosodic features. In our experiments, we compare our model with a baseline model that is based on only utterance features. The results show the effectiveness of our multimodal approach. In addition, we improve the performance of the baseline model to add some new utterance features.

DOI： 10.1109/IALP51396.2020.9310465

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85099875287&origin=inward

Acquisition of Periodic Events with Person Attributes 査読有り

Yamamoto K., Shimada K.

2020 International Conference on Asian Language Processing, IALP 2020 229 - 234 2020年12月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© 2020 IEEE. In this paper, we aim to acquire 'periodic events' that represent significant actions that are happened by groups of people in particular seasons or timing. Recently, the importance of knowledge acquisition is increasing. Many studies about human action knowledge and temporal commonsense knowledge have been carrying out. We need a dataset to acquire periodic events. However, manually building the dataset with human attribute labels is costly. Therefore, we construct a human attribute classifier of Twitter users and create a large labeled tweets dataset automatically. Periodic events with specific human attributes are collected with our proposed method. Finally, we obtained commonsense event knowledge; e.g., 'Students often go to college at 1 P.M.' and 'Workers often work overtime on weekdays.'

DOI： 10.1109/IALP51396.2020.9310489

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85099887637&origin=inward

Detection of Lying Situations in Liar Corpus

Shohei Takabatake, Kazutaka Shimada, and Takeshi Saitoh

Proceedings of the 8th International Symposium on Applied Engineering and Sciences 2020年12月

Relation Identification Using Dialogical Features in Multi-Party Conversation

Taskumi Himeno, Kazutaka Shimada

Proceedings of the 8th International Symposium on Applied Engineering and Sciences 2020年12月

The Discussion Corpus toward Argumentation Quality Assessment in Multi-Party Conversation 査読有り国際誌

Tsukasa Shiota, Kazutaka Shimada

Proceedings of the 9th International Conference on Learning Technologies and Learning Environments 2020年09月

Don't beat a dead horse: Recognizing a person who returns to a done-deal in a multi-party conversation 査読有り国際誌

Shunsuke Yonemitsu, Kazutaka Shimada

Proceedings of the 8th International Conference on Smart Computing and Artificial Intelligence 2020年09月

Don't beat a dead horse: Recognizing a person who returns to a done-deal in a multi-party conversation 査読有り

Yonemitsu S., Shimada K.

Proceedings - 2020 9th International Congress on Advanced Applied Informatics, IIAI-AAI 2020 444 - 449 2020年09月

記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In this paper, we construct a corpus that contains a person who returns to a done-deal in a multi-party conversation. We call the person "BDH (Beat a Dead Horse) participant."Each conversation in the corpus consists of utterances about a topic with three participants. We prepare two topics for the corpus and one of three in participants is the BDH participant. The BDH participant mentions negative opinions about a done-deal in the previous conversation. For recognizing a BDH participant in a conversation, we generate a recognition model based on a machine learning approach from the corpus. We propose nine features and apply them to Support Vector Machines. We obtained 0.743 on the F-score by using the model with the best feature set. The bi-gram and tri-gram in them were the most effective features in the experiment.

DOI： 10.1109/IIAI-AAI50415.2020.00095

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85107216460&origin=inward

The Discussion Corpus toward Argumentation Quality Assessment in Multi-Party Conversation 査読有り

Shiota T., Shimada K.

Proceedings - 2020 9th International Congress on Advanced Applied Informatics, IIAI-AAI 2020 280 - 283 2020年09月

記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Recently, many studies in NLP, especially argument mining, have focused on revealing what a good argument is with written texts. However, there are a few studies to challenge argumentation quality assessment in face-to-face conversations due to a lack of corpora, as compared with written texts. Therefore, we construct a multimodal multi-party discussion corpus toward argumentation quality assessment in multi-party conversation. The corpus consists of 10 multi-party discussions (200 minutes) in total. Each participant is annotated for 6 scores about rhetoric, which is one of the classes of argumentation quality. In this paper, we explain the procedure of our corpus construction and report the results of the annotation.

DOI： 10.1109/IIAI-AAI50415.2020.00062

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85107229472&origin=inward

Timing Prediction of Facilitating Utterance in Multi-party Conversation 査読有り

Sembokuya T., Shimada K.

Communications in Computer and Information Science 1215 CCIS 267 - 279 2020年01月

担当区分：責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© 2020, Springer Nature Singapore Pte Ltd. Supporting consensus-building in multi-party conversations is a very important task in intelligent systems. To conduct smooth, active, and productive discussions, we need a facilitator who controls a discussion appropriately. However, it is impractical to assign a good facilitator to each group in the discussion environment. The goal of our study is to develop a digital facilitator system that supports high-quality discussions. One role of the digital facilitator is to generate facilitating utterances in the discussions. To realize the system, we need to predict the timing of facilitating utterances. To apply a machine learning technique to our model, we construct a data set from the AMI corpus, first. For the construction, we use some rules based on the annotation of the corpus. Then, we generate a prediction model with verbal and non-verbal features extracted from discussions. We obtained 0.75 on the F-measure. We compared our model with a baseline method. Our model outperformed the baseline (0.7 vs. 0.5 on the AUC value). In addition, we introduce additional features about the role of participants in the AMI corpus. By using the additional features, the F measure increased by 2 points. The experimental results show the effectiveness of our model.

DOI： 10.1007/978-981-15-6168-9_23

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85088497031&origin=inward

Acquisition of Knowledge with Time Information from Twitter 査読有り

Yamamoto K., Shimada K.

Proceedings of the 2019 International Conference on Asian Language Processing, IALP 2019 148 - 153 2019年11月

記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© 2019 IEEE. In this paper, we propose a knowledge acquisition method for non-Task-oriented dialogue systems. Such dialogue systems need a wide variety of knowledge for generating appropriate and sophisticated responses. However, constructing such knowledge is costly. To solve this problem, we focus on a relation about each tweet and the posted time. First, we extract event words, such as verbs, from tweets. Second, we generate frequency distribution for five different time divisions: e.g., a monthly basis. Then, we remove burst words on the basis of variance for obtaining refined distributions. We checked high ranked words in each time division. As a result, we obtained not only common sense things such as 'sleep' in night but also interesting activities such as 'recruit' in April and May (April is the beginning of the recruitment process for the new year in Japan.) and 'raise the spirits/plow into' around 9 AM for inspiring oneself at the beginning of his/her work of the day. In addition, the knowledge that our method extracts probably contributes to not only dialogue systems but also text mining and behavior analysis of data on social media and so on.

DOI： 10.1109/IALP48816.2019.9037659

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85083205940&origin=inward

Sarcasm detection using RnN with relation vector 査読有り

Hiai S., Shimada K.

International Journal of Data Warehousing and Mining 15 ( 4 ) 66 - 78 2019年10月

担当区分：最終著者記述言語：英語掲載種別：研究論文（学術雑誌）

Copyright © 2019, IGI Global. Sarcasm detection has been treated as a task that classifies text as sarcastic or non-sarcastic. Sarcasm detection is a significant challenge for sentiment analysis because sarcasm involves a positive expression with a negative meaning. Surface information in text is commonly used as a classification feature. However, the authors must consider both surface and non-surface features. In this article, the authors focus on relation information between pairs of role expressions, such as “boss and staff,” and propose a sarcasm detection method based on surface and relation information. First, the authors extract role pairs from a corpus. Then, the authors construct a relation vector generated from these role pairs and incorporate the relation vector into a recurrent neural network model. The authors evaluated the proposed method by comparing it to previously proposed methods. The results demonstrate the effectiveness of introducing the relation vector to sarcasm detection.

DOI： 10.4018/IJDWM.2019100104

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85072636831&origin=inward

Timing Prediction of Facilitating Utterance in Multi-party Conversation 査読有り

Tomonobu Sembokuya, Kazutaka Shimada

Proceedings of PACLING 2019 2019年10月

Construction of a liar corpus and detection of lying situations 査読有り

Takabatake S., Shimada K., Saitoh T.

Proceedings - 2018 Joint 10th International Conference on Soft Computing and Intelligent Systems and 19th International Symposium on Advanced Intelligent Systems, SCIS-ISIS 2018 971 - 976 2019年05月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© 2018 IEEE. Recognizing lies in human interaction is one of the most important and challenging tasks in artificial intelligence. For example, it leads to the development of a human-computer interaction system that detects a liar in a conversation. In another example, some systems in the future might occasionally need to tell a lie for not hurting user's feelings, namely a white lie for trouble-free conversations. In this paper, we construct a video corpus for developing a system that detects lies in conversations. The corpus consists of 540 question-answer pairs by 9 persons; 270 true answers and 270 lies. We manually analyze the images in the corpus for recognizing lies. In addition, we apply machine learning techniques, support vector machines with local binary pattern features and geometric features, to our corpus. We discuss the results of the analysis of the corpus and the machine learning approaches to our corpus.

DOI： 10.1109/SCIS-ISIS.2018.00161

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85067133291&origin=inward

Annotation and analysis of extractive summaries for the kyutech corpus 査読有り

Yamamura T., Shimada K.

LREC 2018 - 11th International Conference on Language Resources and Evaluation 3216 - 3220 2019年01月

記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© LREC 2018 - 11th International Conference on Language Resources and Evaluation. All rights reserved. Summarization of multi-party conversation is one of the important tasks in natural language processing. For conversation summarization tasks, corpora have an important role to analyze characteristics of conversations and to construct a method for summary generation. We are developing a freely available Japanese conversation corpus for a decision-making task. We call it the Kyutech corpus. The current version of the Kyutech corpus contains topic tags of each utterance and reference summaries of each conversation. In this paper, we explain an annotation task of extractive summaries. In the annotation task, we annotate an importance tag for each utterance and link utterances with sentences in reference summaries that already exist in the Kyutech corpus. By using the annotated extractive summaries, we can evaluate extractive summarization methods on the Kyutech corpus. In the experiment, we compare some methods based on machine learning techniques with some features.

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85059885448&origin=inward

Automatic Training Data Construction and Extractive Supervised Summarization for NTCIR-14 QA Lab-PoliInfo 査読有り

Hiai S., Otani Y., Yamamura T., Shimada K.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 11966 LNCS 97 - 109 2019年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© 2019, Springer Nature Switzerland AG. On the summarization task at NTCIR-14 QA Lab-PoliInfo, participants of the task need to generate a summary corresponding to an assembly member speech in assembly minutes within the limit length. Our method extracts important sentences to summarize an assembly member speech in the minutes. Our method applies a machine learning model to predict the important sentences. However, the given assembly minutes’ data do not contain information about the importance of the sentences. As a result, we cannot directly utilize machine learning techniques for the task. Therefore, we construct training data for the importance prediction model using a word similarity between sentences in a speech and those in the summary. In addition, we apply the sentence reduction process. In the process, we consider characteristics of summaries of assembly minutes to avoid removal of important words in extracted sentences. On the evaluation, all the scores by our supervised method with the constructed data outperformed unsupervised and supervised baseline methods. The result shows the effectiveness of our method.

DOI： 10.1007/978-3-030-36805-0_8

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85076978347&origin=inward

Trivia Score and Ranking Estimation Using Support Vector Regression and RankNet 査読有り

Kazuya Niina, Kazutaka Shimada

Proceedings of PACLIC32 2018年12月

Dialogue Act Annotation and Identification in a Japanese Multi-party Conversation Corpus 査読有り

Takashi Yamamura, Masato Hino, Kazutaka Shimada

Proceedings of Asia-Pacific Corpus Linguistics Conferenc (APCLC2018) 2018年09月

スポーツ要約生成におけるテンプレート型手法とニューラル型手法の提案と比較査読有り

田川裕輝, 嶋田和孝

自然言語処理 ( 一般社団法人　言語処理学会 ) 25 ( 4 ) 357 - 391 2018年09月

担当区分：最終著者,　責任著者記述言語：日本語掲載種別：研究論文（学術雑誌）

<p>本研究では，日本で人気のある野球に着目し，Play-by-play データからイニングの要約文の生成に取り組む．Web 上では多くの野球に関する速報が配信されている．戦評は試合終了後にのみ更新され，“待望の先制点を挙げる”のような試合の状況をユーザに伝えるフレーズ（本論文では Game-changing Phrase; GP と呼ぶ）が含まれているのが特徴であり，読み手は試合の状況を簡単に知ることができる．このような特徴を踏まえ，任意の打席に対して，GP を含む要約文を生成することは，試合終了後だけでなく，リアルタイムで試合の状況を知りたい場合などに非常に有益であるといえる．そこで，本研究では Play-by-play データから GP を含む要約文の生成に取り組む．また，要約生成手法としてテンプレート型文生成手法と Encoder-Decoder モデルを利用した手法の 2 つを提案する．</p>

DOI： 10.5715/jnlp.25.357

その他リンク： https://www.jstage.jst.go.jp/article/jnlp/25/4/25_357/_pdf

Image-based Attention Level Estimation of Interaction Scene by Head Pose and Gaze Information 査読有り

Komiya R., Saitoh T., Shimada K.

Proceedings - 17th IEEE/ACIS International Conference on Computer and Information Science, ICIS 2018 497 - 501 2018年09月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© 2018 IEEE. In many situations such as chat and discussion, we interact with other people, and interaction analysis is useful for improving communication skills and evaluating a learning effect of education. Although the audio information is possible to analyze the speaker behavior, it is difficult to analyze the non-speaker. On the other hand, the image information is possible to estimate the inside information, such as the concentration level and understanding level of the interaction of the non-speaker not only the speaker by using nodding, face direction, facial expression, and gaze. This paper proposes an image-based method to automatically estimate the attention level of participants for the interaction scene. We collected simulated interaction scenes by using the omnidirectional camera, and applied the proposed method. As a result, we confirmed the effectiveness of the proposed method.

DOI： 10.1109/ICIS.2018.8466462

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85055721201&origin=inward

Discussion map with an assistant function for decision-making: a tool supporting consensus-building 査読有り

Ryunosuke Kirikihira, Kazutaka Shimada

Proceedings of the tenth International Conference on Collaboration Technologies (CollabTech 2018) 2018年09月

Analysis of facilitators' behaviors in multi-party conversations for constructing a digital facilitator system 査読有り

Tsukasa Shiota, Takashi Yamamura,Kazutaka Shimada

Proceedings of the tenth International Conference on Collaboration Technologies (CollabTech 2018) 2018年09月

スポーツ要約生成におけるテンプレート型手法とニューラル型手法の提案と比較査読有り

田川裕輝, 嶋田和孝

自然言語処理 25 ( 4 ) 2018年09月

Image-based Attention Level Estimation of Interaction Scene by Head Pose and Gaze Information 査読有り

Rinko Komiya, Takeshi Saitoh, Kazutaka Shimada

Proceedings of the17th IEEE/ACIS International Conference on Computer and Information Science (ICIS 2018) 497 - 501 2018年06月

Simple and sophisticated inning summary generation based on encoder-decoder model and transfer learning 査読有り

Tagawa Y., Shimada K.

Proceedings of the 2017 International Conference on Asian Language Processing, IALP 2017 2018-January 252 - 255 2018年02月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© 2017 IEEE. This paper describes an inning summarization method for a baseball game by using an encoder-decoder model. Each inning in a baseball game contains some events, such as hits, strikeouts, homeruns and scoring. Simplified description of the events leads to the improvement of readability of the inning information. Our method learns a relation between play-by-play data in each inning and inning reports. We also incorporate sophisticated expressions acquired from game summaries with the model. We call them Game-changing Phrase, GP. One problem in our task is the size of training data for the learning. To solve this problem, we apply a transfer learning approach into our method. In the experiment, we evaluate the effectiveness of our method with the transfer learning.

DOI： 10.1109/IALP.2017.8300591

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85046629402&origin=inward

Stock market prediction using keywords from expert articles 査読有り

Ichinose K., Shimada K.

Advances in Intelligent Systems and Computing 700 409 - 417 2018年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© 2018, Springer International Publishing AG. The market analysis is one of the important tasks for text mining. In this situation, Web news has an important role to predict stock prices. In this paper, we propose a method to predict the Nikkei Stock Average, which is one of the most important stock market indexes. We extract viewpoints from experts’ articles for analyzing Web news. The extracted words are index words in the vector space of a machine learning technique. We also incorporate word embedding and bootstrap approaches into our method. It predicts “UP” or “DOWN” of the next day by using the articles of a day. We also evaluate our method with not only one-day prediction but also simulated trading. The experimental result shows that index words based on expert articles were effective for both one-day prediction and simulated trading.

DOI： 10.1007/978-3-319-72550-5_39

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85041546156&origin=inward

Sports game summarization based on sub-events and game-changing phrases 査読有り

Tagawa Y., Shimada K.

Studies in Computational Intelligence 742 65 - 80 2018年01月

記述言語：英語掲載種別：研究論文（学術雑誌）

© Springer International Publishing AG 2018. Microblogs are one of the most important resources for natural language processing. This paper describes a summarization task of sports events on Twitter. We focus on an abstractive approach based on sub-events in the sports event. Abstractive summaries usually are better than summaries generated by extractive approaches in terms of readability. Furthermore, our method can incorporate sophisticated phrases that explain the scene. First, our method detects burst situations in which many users post tweets when a sub-event in a game occurs. Tweets in the burst situations are the inputs of our method. Next, it extracts sub-event elements (SEEs) that contain actions in a game, such as “Player A made a pass to Player B” and “Player B made a shot on goal.” Then, it identifies the optimal order of the extracted SEEs by using a scoring method. Finally, it generates an abstractive summary on the basis of the ordered SEEs, such as “Playler B made a shot on goal from the Player A’s pass.” In addition, it adds game-changing phrases into the abstractive summary by some rules. In the experiment, we show the effectiveness of our method as compared with related work based on an extractive approach.

DOI： 10.1007/978-3-319-70636-8_5

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85041708730&origin=inward

Sarcasm detection using features based on indicator and roles 査読有り

Hiai S., Shimada K.

Advances in Intelligent Systems and Computing 700 418 - 428 2018年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© 2018, Springer International Publishing AG. Sarcasm is a non-literalistic expression and presents a negative meaning with positive expressions. Sarcasm detection is a significant challenge for sentiment analysis which is to analyze documents with opinions. In this study, we propose a method of sarcasm detection on Twitter. We focus on two kinds of feature words. One is words modified by the indicator “ ”. The other is words expressing a role. First, we extract these words from tweets. Next, our method uses the lists of these words for a machine learning approach to detect sarcastic tweets. The lists of extracted words are used as features in our method. In the experiment, we compare our method with a baseline based on the features in previous studies. The experimental result shows the effectiveness of our method.

DOI： 10.1007/978-3-319-72550-5_40

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85041536268&origin=inward

Trivia score and ranking estimation using support vector regression and ranknet 査読有り

Niina K., Shimada K.

Proceedings of the 32nd Pacific Asia Conference on Language, Information and Computation, PACLIC 2018 263 - 270 2018年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

© 2018 by the authors. Dialogue systems have been increasingly important these days. In particular, non-taskoriented dialogue systems are studied becauseof the success of neural network approachessuch as seq2seq models. However, these models tend to generate simple responses such as"yes" and "ok." To construct a dialogue system that holds users' attention continuously,we need to generate utterances that capturethe interest of the user. In this paper, wepropose a method to extract trivia sentencesfor the purpose. Trivia information perhapsadds a surprise to users. Therefore, capturing trivia information is beneficial for dialogue systems. We estimate a trivia scoreof a sentence by using machine learning approaches, Support Vector Regression (SVR)and RankNet. We obtained 0.79 and 0.78 onSVR for the nDCG@5 and RankNet for thenDCG@10, respectively. We focus on thesubject word in each sentence. The methodwith subject information outperformed thatwithout subject information; 0.79 with subjectinformation vs. 0.64 without subject information on the SVR for the nDCG@5.

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85083277805&origin=inward

Discussion map with an assistant function for decision-making: A tool for supporting consensus-building 査読有り国際誌

Kirikihira R., Shimada K.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 11000 LNCS 3 - 18 2018年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In this paper, we propose a tool for supporting consensus-building in conversations with multiple participants. We call it “Discussion Map with Assistant (DMA)”. It consists of nodes and links. We classify the nodes into two types; alternatives and criteria. Alternatives represent what the participants are choosing between. Criteria are used to judge the alternatives. Each criterion contains an importance value. Each link between nodes also contains an importance value. The system estimates a ranking list of alternatives among participants from each map. We introduce a forgetting function to the model. The system also supports the decision-making process by using discussion maps from participants. It generates sentences and charts that describe the current state of the discussion. We evaluate the effectiveness of the discussion map system with DMA in a decision-making task experimentally.

DOI： 10.1007/978-3-319-98743-9_1

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85053106635&origin=inward

Analysis of facilitators’ behaviors in multi-party conversations for constructing a digital facilitator system 査読有り国際誌

Shiota T., Yamamura T., Shimada K.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 11000 LNCS 145 - 158 2018年01月

担当区分：最終著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In this paper, we analyze characteristics of facilitators from multi-party conversations. The goal of our study is to construct a digital facilitator system that supports consensus-building and management of conversation for high-quality discussions. Therefore, we need facilitator’s knowledge, behavior, and patterns to realize a good digital facilitator. As the 1st step for the purpose, we focus on a macro viewpoint of facilitators’ behaviors on conversation corpora. First, we generate a model based on a decision tree that classifies each participant into a facilitator or a non-facilitator, from conversation corpora. The classification accuracies by the decision trees were 0.642 and 0.737 for two corpora, respectively. The main purpose of the decision tree generation is to extract patterns from imaginable characteristics, namely features for the classifier. Therefore, next, we discuss behaviors of facilitators by analyzing the decision tree manually. In the analysis, we focus on two types of corpora; one is that each participant has a role, such as a project manager, and another is that each participant has no role in the conversation. We investigate the influence of the difference of the setting through the analysis. From the manual analysis, we obtained some common behaviors and some different behaviors about facilitators from two corpora.

DOI： 10.1007/978-3-319-98743-9_12

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85053169078&origin=inward

Image based Attention Estimation for Interaction Scene 査読有り

Rinko Komiya, Takeshi Saitoh, Kazutaka Shimada

International Conference on ICT Robotics (ICT-ROBOT2017) 2017年11月

The Kyutech corpus and the topic segmentation using a combined method 査読有り

Takashi Yamamura, Kazutaka Shimada and Shintaro Kawahara

Proceedings of the 12th Workshop on Asian Language Resources 95 - 104 2016年12月

Generating abstractive summaries of sports games from Japanese tweets 査読有り

Yuuki Tagawa, Kazutaka Shimada

Proceedings of the 7th International Conference on E-Service and Knowledge Management (ESKM 2016) 2016年07月

Stock market prediction from news on the Web and a new evaluation approach in trading 査読有り

Ko Ichinose, Kazutaka Shimada

Proceedings of the 7th International Conference on E-Service and Knowledge Management (ESKM 2016) 2016年07月

A Sarcasm Extraction Method Based on Patterns of Evaluation Expressions 査読有り

Satoshi Hiai, Kazutaka Shimada

Proceedings of the 7th International Conference on E-Service and Knowledge Management (ESKM 2016) 2016年07月

タブレット端末を用いた議論の合意形成支援システムの開発

桐木平龍之介, 嶋田和孝, Giuseppe Carenini

人工知能学会研究会資料先進的学習科学と工学研究会 ( 一般社団法人人工知能学会 ) 75 ( 0 ) 08 2015年11月

担当区分：最終著者記述言語：日本語掲載種別：研究論文（研究会，シンポジウム資料等）

<p>Problem-based learning (PBL) is one of the most important learning approaches in recent years. In PBL, participants in a group discuss collaboratively to solve a problem and to make a decision through a conversation. In this paper, we propose a tool for supporting consensus-building on multi party conversations. We call it ``Discussion Map''. It consists of nodes and links between them. We divide the nodes into two types; alternatives and criteria. Each criterion contains the importance value. Each link between nodes also contains the importance value. We evaluate the effectiveness of the discussion map system experimentally.</p>

DOI： 10.11517/jsaialst.75.0_08

Multi-aspects Rating Prediction Using Aspect Words and Sentences 査読有り

Takuto Nakamuta, Kazutaka Shimada

Proceedings of the 29th Pacific Asia Conference on Language, Information and Computation (PACLIC29) 513 - 521 2015年10月

Trouble information extraction based on a bootstrap approach from Twitter 査読有り

Kohei Kurihara, Kazutaka Shimada

Proceedings of the 29th Pacific Asia Conference on Language, Information and Computation (PACLIC29) 471 - 479 2015年10月

Extractive Summarization Based on a Combined Method Using Several Features for Multi-Party Conversation 査読有り

Yo Tokunaga, Kazutaka Shimada

ACIS International Journal of Computer and Information Science 16 ( 2 ) 12 - 21 2015年10月

On-site likelihood identification of tweets using a two-stage method 査読有り

Kazutaka Shimada, Yurie Onitsuka, Shunsuke Inoue and Tsutomu Endo

Tourism Informatics: Towards novel knowledge based approaches 77 - 90 2015年07月

The effectiveness of time sequence information on a sightseeing spot recommender 査読有り

Kazutaka Shimada, Hisashi Uehara and Tsutomu Endo

Tourism Informatics: Towards novel knowledge based approaches 135 - 150 2015年07月

Trouble information extraction based on a bootstrap approach from Twitter 査読有り

Kurihara K., Shimada K.

29th Pacific Asia Conference on Language, Information and Computation, PACLIC 2015 471 - 479 2015年01月

記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In this paper, we propose a method for ex-Tracting trouble information from Twitter. One useful approach is based on machine learn-ing techniques such as SVMs. However, trou-ble information is a fraction of a percent of all tweets on Twitter. In general, imbalanced distribution is not suitable for machine learn-ing techniques to generate a classifier. An-other approach is to extract trouble informa-Tion by using handwritten rules. However, constructing high coverage rules by handwork is costly. First, we verify these problems in a preliminary experiment. Then, to solve these problems, we apply a bootstrapping method to our trouble information extraction task. We introduce three characteristics and a scoring method to the bootstrapping. As a result, the iteration process on the bootstrapping in-creased the number of tweets and patterns for trouble information dramatically.

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84967214979&origin=inward

Multi-Aspects rating prediction using aspectwords and sentences 査読有り

Nakamuta T., Shimada K.

29th Pacific Asia Conference on Language, Information and Computation, PACLIC 2015 513 - 521 2015年01月

記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In this paper we propose a method for a rat-ing prediction task. Each review consists of several ratings for a product, namely aspects. To predict the ratings of the aspects, we uti-lize not only aspect words, but also aspect sen-Tences. First, our method detects aspect sen-Tences by using a machine learning technique. Then, it incorporates words extracted from as-pect sentences with aspect word features. For estimating aspect likelihood of each word, we utilize the variance of words among aspects. Finally, it generates classifiers for each aspect by using the extracted features based on the aspect likelihood. Experimental result shows the effectiveness of features from aspect sen-Tences.

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84966908156&origin=inward

Understanding level estimation using discussion maps for supporting consensus-building 査読有り

Hironobu Takagi, Kazutaka Shimada

18th International Conference in Knowledge Based and Intelligent Information and Engineering Systems - KES2014 2014年09月

Multi-party conversation summarization based on sentence selection using verbal and nonverbal information 査読有り

Yo Tokunaga, Kazutaka Shimada

2nd International Conference on Smart Computing and Artificial Intelligence (ICSCAI 2014) 2014年08月

A comparative study of potential-of-interest days on a sightseeing spot recommender 査読有り

Kazutaka Shimada, Hisashi Uehara and Tsutomu Endo

International Workshop on Sustainable Tourism Innovations and Information Systems (STIIS2014) 2014年08月

集合知に基づく観光地推薦システムの構築査読有り

嶋田和孝,上原尚,遠藤勉

観光情報学会「観光と情報」 10 ( 1 ) 113 - 124 2014年06月

Cooperation level estimation of pair work using top-view image 査読有り

Sakaguchi K., Shimada K.

Advances in Intelligent Systems and Computing 272 77 - 87 2014年01月

記述言語：英語掲載種別：研究論文（学術雑誌）

© Springer International Publishing Switzerland 2014.To understand an interaction among persons is one of the most important tasks in artificial intelligence. In this paper, we propose a method for estimating a cooperation level in pair work. The task is a cooperation work that take place in front of a whiteboard by two persons. The goal of our study is to provide the cooperation level that is estimated by features extracted from images for teachers. The result of this study is useful for education support systems and problem based learning. We extract the standing location, operation ratio and head direction of each person from an overhead camera. We apply the features to two machine learning approaches: AdaBooost and multiple linear regression. We obtained 77.5% as the accuracy by the AdaBoost and 0.649 as the adjusted R2 by the regression.

DOI： 10.1007/978-3-319-05570-1_9

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84927761073&origin=inward

Person Identification Using Top-View Image with Depth Information 査読有り

Daichi Kouno, Kazutaka Shimada, Tsutomu Endo

International Journal of Software Innovation 1 ( 2 ) 67 - 79 2013年11月

Conversation summarization using machine learning and scoring method 査読有り

Kazutaka Shimada, Shinpei Toyodome, Tsutomu Endo

Proceedings of the 13th Conference of the Pacific Association for Computational Linguistics 2013年09月

Twitter Sentiment Analysis Based on Writing Style 査読有り

Hiroshi Maeda, Kazutaka Shimada and Tsutomu Endo

Proceedings of 8th International Conference on Natural Language Processing JapTAL 2012 2012年10月

On-site likelihood identification of tweets for tourism information analysis 査読有り

Kazutaka Shimada, Shunsuke Inoue, Tsutomu Endo

Proceedings of 3rd IIAI International Conference on e-Services and Knowledge Management (IIAI ESKM 2012) 2012年09月

A combined method based on SVM and online learning with HOG for hand shape recognition 査読有り

Shimada K., Muto R., Endo T.

Journal of Advanced Computational Intelligence and Intelligent Informatics 16 ( 6 ) 687 - 695 2012年09月

記述言語：英語掲載種別：研究論文（学術雑誌）

In this paper, we propose a combined method for hand shape recognition. It consists of Support Vector Machines (SVMs) and an online learning algorithm based on the perceptron. We apply HOG features to each method. First, our method estimates the hand shape of an input image by using SVMs. Here, an online learning method with the perceptron uses an input image as new training data if the image is effective in relearning in the recognition process. Next, we select a final hand shape from the outputs of SVMs and perceptrons by using the score from SVMs. The combined method with the online perceptron is robust against unknown users because it contains a relearning process for the current user. Therefore applying the online perceptron leads to an improvement in accuracy. We compare the combined method with a method that uses only SVMs. Experimental results show the effectiveness of the proposed method.

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84866617122&origin=inward

A person identification method using a top-view head image from an overhead Camera 査読有り

Nakatani R., Kouno D., Shimada K., Endo T.

Journal of Advanced Computational Intelligence and Intelligent Informatics 16 ( 6 ) 696 - 703 2012年09月

記述言語：英語掲載種別：研究論文（学術雑誌）

In this paper, we describe a novel image-based person identification task. Conventional face-based person identification methods have a low tolerance for occluded situations, such as overlapping of people in an image. We focus on an image from an overhead camera. Using the overhead camera reduces restrictions on the installation location of the camera and solves the problem of occluded images. First, our method identifies a person area in a captured image by using background subtraction. Then, it extracts four features from the area: (1) body size, (2) hair color, (3) hairstyle, and (4) hair whorl. We apply the four features to the AdaBoost algorithm. Experimental results show the effectiveness of our method.

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84866608953&origin=inward

Person Identification Using Top-view Image with Depth Information 査読有り

Daichi Kouno, Kazutaka Shimada, Tsutomu Endo

Proceedings of 13th ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD 2012) 140 - 145 2012年08月

Developing of Interactive e-Learning System with Dialogue 査読有り

Sayaka Minewaki, Kazutaka Shimada, Tsutomu Endo

Proceedings of ICIS '12 and the 2012 IEEE/ACIS 11th International Conference on Computer and Information Science 272 - 276 2012年05月

Analyzing Tourism Information on Twitter for a Local City 査読有り

Kazutaka Shimada, Shunsuke Inoue, Hiroshi Maeda and Tsutomu Endo

Proceedings of SSNE2011, International Workshop on Innovative Tourism 2011年12月

Multi-aspects review summarization with objective information 査読有り

Kazutaka Shimada, Ryosuke Tadano and Tsutomu Endo

Procedia - Social and Behavioral Sciences: Computational Linguistics and Related Fields 27 140 - 149 2011年12月

Aspect Identification of Sentiment Sentences Using A Clustering Algorithm 査読有り

Masashi Hadano, Kazutaka Shimada and Tsutomu Endo

Procedia - Social and Behavioral Sciences: Computational Linguistics and Related Fields 27 22 - 31 2011年12月

A person identification method using facial, clothing and time features 査読有り

Komatsu K., Shimada K., Endo T.

IWACIII 2011 - International Workshop on Advanced Computational Intelligence and Intelligent Informatics, Proceedings 2011年12月

記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In this paper, we propose a method for personal identification using facial features and context information. The method can overcome a problem of partially occluded images. In the proposal method, facial similarity is calculated by using the CLAFIC method from a face image and the parts in the face, such as the image of eyes. In addition, we focus on clothes of each target person and time information as the context. We apply these features into a scoring method and a machine learning method. In the experiments, we confirmed that proposal method is better than the method using only the face image similarity and the heuristic method based on a scoring.

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84874331219&origin=inward

A person identification method using a top-view head image from an overhead camera 査読有り

Nakatani R., Kouno D., Shimada K., Endo T.

IWACIII 2011 - International Workshop on Advanced Computational Intelligence and Intelligent Informatics, Proceedings 2011年12月

記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In this paper, we describe a novel image-based person identification task. Traditional face-based person identification methods have a low tolerance for occluded situation, such as overlapping of people in an image. We focus on an image from an overhead camera. Using the overhead camera reduces a restriction of the installation location of a camera and solves the problem of occluded images. First, our method identifies the person's area in a captured image by using background subtraction. Then, it extracts four features from the area; (1) body size, (2) hair color, (3) hairstyle and (4) hair whorl. We apply the four features into the AdaBoost algorithm. Experimental result shows the effectiveness of our method.

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84874305692&origin=inward

A combined method based on SVM and online learning with HOG for hand shape recognition 査読有り

Shimada K., Muto R., Endo T.

IWACIII 2011 - International Workshop on Advanced Computational Intelligence and Intelligent Informatics, Proceedings 2011年12月

記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In this paper, we propose a combined method for hand shape recognition. It consists of support vector machines (SVMs) and an online learning algorithm based on the perceptron. We apply HOG features to each method. First, our method estimates a hand shape of an input image by using SVMs. Here the online learning method with the perceptron uses the input image as new training data if the data is effective for relearning in the recognition process. Next, we select the final hand shape from the outputs of the SVMs and perceptron by using the score of SVMs. The combined method deals with a problem about decrease of the accuracy in the case that users change. Applying the online perceptron jointly leads to improvement of the accuracy. We compare the combined method with a method using only SVMs. The experimental result shows the effectiveness of the proposed method.

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84874341202&origin=inward

Multi-aspects Review Summarization with Objective Information 査読有り

Kazutaka Shimada, Ryosuke Tadano and Tsutomu Endo

Proceedings of Proceedings of the 12th Conference of the Pacific Association for Computational Linguistics 2011年07月

Aspect Identification of Sentiment Sentences Using a Clustering Algorithm 査読有り

Masashi Hadano, Kazutaka Shimada and Tsutomu Endo

Proceedings of Proceedings of the 12th Conference of the Pacific Association for Computational Linguistics 2011年07月

顔特徴とコンテキスト情報に基づく人物識別査読有り

山口純平, 嶋田和孝, 榎田修一, 江島俊朗, 遠藤勉

日本知能情報ファジィ学会誌, 知能と情報 23 ( 2 ) 13 - 21 2011年04月

Interactive Aspect Summarization Using Word-Aspect Relations for Review Documents 査読有り

Kazutaka Shimada ,Masashi Yamaumi,Ryosuke Tadano,Masashi Hadano,Tsutomu Endo

Proceedings of the 5th International Conference on Soft Computing and Intelligent Systems and 11th International Symposium on Advanced Intelligent Systems (SCIS & ISIS 2010) 183 - 188 2010年12月

Multi-aspects review summarization based on identification of important opinions and their similarity 査読有り

Tadano R., Shimada K., Endo T.

PACLIC 24 - Proceedings of the 24th Pacific Asia Conference on Language, Information and Computation 685 - 692 2010年12月

記述言語：英語掲載種別：研究論文（学術雑誌）

The development of theWeb services lets many users easily provide their opinions recently. Automatic summarization of enormous sentiments has been expected. Intuitively, we can summarize a review with traditional document summarization methods. However, such methods have not well-discussed "aspects". Basically, a review consists of sentiments with various aspects. We summarize reviews for each aspect so that the summary presents information without biasing to a specific topic. In this paper, we propose a method for multiaspects review summarization based on evaluative sentence extraction. We handle three features; ratings of aspects, the tf -idf value, and the number of mentions with a similar topic. For estimating the number of mentions, we apply a clustering algorithm. By integrating these features, we generate a more appropriate summary. The experiment results show the effectiveness of our method. © 2010 by Ryosuke Tadano, Kazutaka Shimada, and Tsutomu Endo.

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84863868523&origin=inward

Interactive aspect summarization using word-aspect relations for review documents 査読有り

Kazutaka S., Masashi Y., Ryosuke T., Masashi H., Tsutomu E.

SCIS and ISIS 2010 - Joint 5th International Conference on Soft Computing and Intelligent Systems and 11th International Symposium on Advanced Intelligent Systems 183 - 188 2010年12月

記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

As the World Wide Web rapidly grows, a huge number of online documents are easily accessible on the Web. We obtain a huge number of review documents that include user's opinions for products. To summarize the opinions is one of the hottest topics in natural language processing. We focus on aspects of a product in the summarization process. First, we identify a relation between aspects and each word in review documents. Our method employs an unsupervised approach for the identification process. The experimental result shows the effectiveness of the method. Next, we generate a summary by using the relations. In our system, users obtain the summary by an interactive approach.

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84866656025&origin=inward

Combination of 3 types of speech recognizers for anaphora resolution 査読有り

Shimada K., Tanamachi N., Endo T.

PACLIC 24 - Proceedings of the 24th Pacific Asia Conference on Language, Information and Computation 281 - 290 2010年12月

記述言語：英語掲載種別：研究論文（学術雑誌）

In this paper, we propose a method for anaphora resolution in speech understanding for a livelihood support robot. For robust speech recognition, we combine two types of speech recognizers; a large vocabulary continuous speech recognizer (LVCSR) and domain-specific speech recognizers (DSSR). One problem in the anaphora resolution is lack of the antecedent in the outputs. To solve the problem, we introduce 2 types of DSSRs; one medium-scale DSSR and several small DSSRs. In this paper, we describe the basic idea of our multiple speech recognizer first. The selection process in the recognizer is based on the similarity between the LVCSR and each DSSR. Then, by using the outputs from the LVCSR and the medium-scale DSSR, we resolve anaphoric expressions in the current output from a small-scale DSSR. The experimental result shows the effectiveness of our method. © 2010 by Kazutaka Shimada, Noriko Tanamachi, and Tsutomu Endo.

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84863869152&origin=inward

Combination of 3 types of speech recognizers for anaphora resolution 査読有り

Kazutaka Shimada,Noriko Tanamachi,Tsutomu Endo

Proceedings of the 24nd Pacific Asia Conference on Language, Information and Computation (PACLIC24) 281 - 290 2010年11月

Multi-aspects Review Summarization Based on Identification of Important Opinions and their Similarity 査読有り

Ryosuke Tadano,Kazutaka Shimada,Tsutomu Endo

Proceedings of the 24nd Pacific Asia Conference on Language, Information and Computation (PACLIC24) 685 - 692 2010年11月

A Hierarchical Multiple Recognizer for Robust Speech Understanding 査読有り

Takahiko Yokoyama,Kazutaka Shimada,Tsutomu Endo

Proceedings of The Pacific Rim International Conference on Artificial Intelligence (PRICAI) ( LNAI 6230 ) 706 - 712 2010年09月

Web image retrieval for abstract queries using text and image information 査読有り

Kazutaka Shimada,Suguru Ishikawa,Tsutomu Endo

Proceedings of The Fifth Asia Information Retrieval Symposium (AIRS 2009) 2009年10月

Speech Understanding in a Multiple Recognizer with an Anaphora Resolution Process 査読有り

Kazutaka Shimada,Akira Uzumaki,Mai Kitajima,Tsutomu Endo

Proceedings of the 11th Conference of the Pacific Association for Computational Linguistics (PACLING2009) 2009年09月

Effective construction and expansion of a sentiment corpus using an existing corpus and evaluative criteria estimation 査読有り

Ryosuke Tadano,Kazutaka Shimada,Tsutomu Endo

Proceedings of the 11th Conference of the Pacific Association for Computational Linguistics (PACLING2009) 2009年09月

A Graph-based Approach for Sentiment Sentence Extraction 査読有り

Kazutaka Shimada,Daigo Hashimoto,Tsutomu Endo

Lecture Notes in Artificial Intelligence, Vol. 5433, New Frontiers in Applied Data Mining, PAKDD 2008 International Workshops, Revised Selected Papers 5433 38 - 48 2009年02月

Sentiment Sentence Extraction Using a Hierarchical Directed Acyclic Graph Structure and a Bootstrap Approach 査読有り

Kazutaka Shimada,Daigo Hashimoto,Tsutomu Endo

Proceedings of the 22st Pacific Asia Conference on Language, Information and Computation (PACLIC22) 2008年11月

An Effective Speech Understanding Method with a Multiple Speech Recognizer based on Output Selection using Edit Distance 査読有り

Kazutaka Shimada,Satomi Horiguchi,Tsutomu Endo

Proceedings of the 22st Pacific Asia Conference on Language, Information and Computation (PACLIC22) 2008年11月

Seeing several stars: a rating inference task for a document containing several evaluation criteria 査読有り

Kazutaka Shimada,Tsutomu Endo

the 12th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 2008) 1006 - 1014 2008年04月

A Graph-based Approach for Sentiment Sentence Extraction 査読有り

Kazutaka Shimada,Daigo Hashimoto,Tsutomu Endo

First International Workshop on Algorithms for Large-Scale Information Processing in Knowledge Discovery (ALSIP 2008) 42 - 51 2008年04月

Movie Review Classification Based on a Multiple Classifier 査読有り

Kimitaka Tsutsumi,Kazutaka Shimada,Tsutomu Endo

Proceedings of the 21st Pacific Asia Conference on Language, Information and Computation 481 - 488 2007年11月

A case study of comparison of several methods for corpus-based speech intention identification 査読有り

Kazutaka Shimada,Kaoru Iwashita,Tsutomu Endo

Proceedings of the 10th Conference of the Pacific Association for Computational Linguistics (PACLING2007) 255 - 262 2007年09月

Acquisition of Synonyms Using Relations in Product Specification Tables

Tsunehito Seki,Kazutaka Shimada，Tsutomu Endo

Systems and Computers in Japan 38 ( 12 ) 25 - 36 2007年04月

Speech Understanding Based on Keyword Extraction and Relation Between Words 査読有り

Kazutaka Shimada,Tsutomu Endo,Sayaka MInewaki

Computational Intelligence 23 ( 1 ) 45 - 60 2007年02月

Table Recognition Using Attribute Likelihood Computed from Words 査読有り

Tsubasa Kitayama,Kazutaka Shimada,Tsutomu Endo

Fourth Workshop on Learning with Logics and Logics for Learning (LLLL2006) 49 - 52 2006年06月

表の属性と属性値の関係を利用した類義語抽出査読有り

関恒仁,嶋田和孝,遠藤勉

電子情報通信学会論文誌 89 ( 9 ) 2087 - 2100 2006年04月

Interpretation of Utterances based on Relevance Theory: Toward the Formalization of Implicature with the Maximum Relevance 査読有り

Sayaka Minewaki,Kazutaka Shimada,Tsutomu Endo

PACLING2005 214 - 222 2005年08月

Acquisition of new training data from unlabeled data for product specifications extraction 査読有り

Kazutaka Shimada,Tsutomu Endo

PACLING2005 272 - 277 2005年08月

Speech Understanding Using Confidence Measures and Dependency Relations 査読有り

Kazutaka Shimada,Yoshitaka Uchida,Shizue Sato,Sayaka Minewaki,Tsutomu Endo

PACLING2005 278 - 283 2005年08月

SVMおよびTransductive SVMを用いた製品スペック情報の抽出査読有り

嶋田和孝,林晃司,遠藤勉

自然言語処理 12 ( 3 ) 43 - 66 2005年07月

Product Specifications Summarization and Product Ranking System using User's Requests

Kazutaka Shimada,Tsutomu Endo

Information Modelling and Knowledge Bases XV 315 - 331 2004年03月

Multiform Summarization from Product Specifications 査読有り

Kazutaka Shimada,Tetsuro Ito,Tsutomu Endo

Proc. of PACLING2003 83 - 92 2003年08月

Keyword and Weighting for Product Specifications Extraction 査読有り

Kazutaka Shimada,Koji Hayashi,Tsutomu Endo

Proc. of PACLING2003 285 - 293 2003年08月

Information Extraction from Personal Computer Specifications on the Web Using a User's Request 査読有り

Kazutaka SHIMADA,Atsushi FUKUMOTO,Tsutomu ENDO

IEICE Transactions on Information and Systems E86-D ( 8 ) 1386 - 1395 2003年08月

Product Specifications Summarization and Product Ranking System using User's Requests 査読有り

Kazutaka Shimada,Tsutomu Endo

Pre-Proceedings of the 13th European-Japanese Conference on Information Modeling and Knowledge Bases 333 - 349 2003年06月

Classification of Images using Their Neighboring Sentences 査読有り

嶋田　和孝

Proceedings of PACLING 2001 2001年04月

　前のページ-次のページ　

1