研究者詳細 - 嶋田　和孝

論文 - 嶋田　和孝

分割表示146 件中 41 - 60 件目／全件表示 >>

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

Scopus

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

Scopus

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

Scopus

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

Scopus

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

Kyutacar

Scopus

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

Scopus

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

Kyutacar

Scopus

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

Kyutacar

Scopus

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

Scopus

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

Scopus

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

Scopus

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

Scopus

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

　前のページ - 次のページ　

<嶋田 和孝>

論文 - 嶋田 和孝

<嶋田　和孝>

論文 - 嶋田　和孝