講演名 | 2022-07-09 Retrieval of Similar Questions from QAbot Data based on Transformer Language Model 陳 梓豪(近畿大), 半田 久志(近畿大), 白浜 公章(近畿大), |
---|---|
PDFダウンロードページ | PDFダウンロードページへ |
抄録(和) | It has recently become possible to collect a large amount of question-answer pairs that arose in certain educational courses. However, one of the biggest difficulties to utilize such data is the lack of a retrieval system that helps a student efficiently find answers to his/her own question. This paper presents our system that takes advantage of a transformer language model which keeps breaking the state-of-the-art in various natural language processing tasks. In particular, our system uses Sentence-BERT to extract a useful feature of a question, so that questions semantically similar to the user’s one can be accurately identified through their feature matching. In addition, an important prerequisite for objective performance evaluation is a large-scale dataset containing many question-answer pairs. For this, a bot is developed on Slack to collect 1142 question-answer pairs that are individually annotated with the relevance to each of 20 queries. The experimental results on this dataset validate the effectiveness of our system based on Sentence-BERT. |
抄録(英) | It has recently become possible to collect a large amount of question-answer pairs that arose in certain educational courses. However, one of the biggest difficulties to utilize such data is the lack of a retrieval system that helps a student efficiently find answers to his/her own question. This paper presents our system that takes advantage of a transformer language model which keeps breaking the state-of-the-art in various natural language processing tasks. In particular, our system uses Sentence-BERT to extract a useful feature of a question, so that questions semantically similar to the user’s one can be accurately identified through their feature matching. In addition, an important prerequisite for objective performance evaluation is a large-scale dataset containing many question-answer pairs. For this, a bot is developed on Slack to collect 1142 question-answer pairs that are individually annotated with the relevance to each of 20 queries. The experimental results on this dataset validate the effectiveness of our system based on Sentence-BERT. |
キーワード(和) | Question answering / Information retrieval / Transformer / Sentence-BERT / QAbot |
キーワード(英) | Question answering / Information retrieval / Transformer / Sentence-BERT / QAbot |
資料番号 | ET2022-9 |
発行日 | 2022-07-02 (ET) |
研究会情報 | |
研究会 | ET |
---|---|
開催期間 | 2022/7/9(から1日開催) |
開催地(和) | オンライン開催 |
開催地(英) | Online |
テーマ(和) | 学習履歴データの蓄積・分析と実践応用/一般 |
テーマ(英) | Data Accumulation, Analysis and Practice Application of Learning History Data, etc. |
委員長氏名(和) | 渡辺 健次(広島大) |
委員長氏名(英) | Kenji Watanabe(Hiroshimai Univ.) |
副委員長氏名(和) | 國宗 永佳(千葉工大) |
副委員長氏名(英) | Hisayoshi Kunimune(Chiba Inst. of Tech.) |
幹事氏名(和) | 西尾 典洋(目白大) / 立岩 佑一郎(名工大) |
幹事氏名(英) | Norihiro Nisio(Mejiro Univ.) / Yuichiro Tateiwa(Nagoya Institute of Technology.) |
幹事補佐氏名(和) | 吉原 和明(近畿大) / 加納 徹(東京理科大) |
幹事補佐氏名(英) | Kazuaki Yoshihara(Kinki Univ.) / Toru Kano(Tokyo University of Science) |
講演論文情報詳細 | |
申込み研究会 | Technical Committee on Educational Technology |
---|---|
本文の言語 | ENG |
タイトル(和) | |
サブタイトル(和) | |
タイトル(英) | Retrieval of Similar Questions from QAbot Data based on Transformer Language Model |
サブタイトル(和) | |
キーワード(1)(和/英) | Question answering / Question answering |
キーワード(2)(和/英) | Information retrieval / Information retrieval |
キーワード(3)(和/英) | Transformer / Transformer |
キーワード(4)(和/英) | Sentence-BERT / Sentence-BERT |
キーワード(5)(和/英) | QAbot / QAbot |
第 1 著者 氏名(和/英) | 陳 梓豪 / Zihao Chen |
第 1 著者 所属(和/英) | 近畿大学(略称:近畿大) Kindai University(略称:Kindai Univ.) |
第 2 著者 氏名(和/英) | 半田 久志 / Hisashi Handa |
第 2 著者 所属(和/英) | 近畿大学(略称:近畿大) Kindai University(略称:Kindai Univ.) |
第 3 著者 氏名(和/英) | 白浜 公章 / Kimiaki Shirahama |
第 3 著者 所属(和/英) | 近畿大学(略称:近畿大) Kindai University(略称:Kindai Univ.) |
発表年月日 | 2022-07-09 |
資料番号 | ET2022-9 |
巻番号(vol) | vol.122 |
号番号(no) | ET-102 |
ページ範囲 | pp.5-11(ET), |
ページ数 | 7 |
発行日 | 2022-07-02 (ET) |