| Item type |
SIG Technical Reports(1) |
| 公開日 |
2020-12-14 |
| タイトル |
|
|
タイトル |
Topic- and Context-Focused Extractive Summarization of Wikipedia sentences by Fine-tuning BERT and Similarity Calculation |
| タイトル |
|
|
言語 |
en |
|
タイトル |
Topic- and Context-Focused Extractive Summarization of Wikipedia sentences by Fine-tuning BERT and Similarity Calculation |
| 言語 |
|
|
言語 |
eng |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
テキスト処理 |
| 資源タイプ |
|
|
資源タイプ識別子 |
http://purl.org/coar/resource_type/c_18gh |
|
資源タイプ |
technical report |
| 著者所属 |
|
|
|
Graduate School of Information, Production and Systems, Waseda University |
| 著者所属 |
|
|
|
Graduate School of Information, Production and Systems, Waseda University |
| 著者所属(英) |
|
|
|
en |
|
|
Graduate School of Information, Production and Systems, Waseda University |
| 著者所属(英) |
|
|
|
en |
|
|
Graduate School of Information, Production and Systems, Waseda University |
| 著者名 |
Yiqi, Shao
Mizuho, Iwaihara
|
| 著者名(英) |
Yiqi, Shao
Mizuho, Iwaihara
|
| 論文抄録 |
|
|
内容記述タイプ |
Other |
|
内容記述 |
Wikipedia has been one of the most famous online encyclopedias. Articles' contents in Wikipedia are constantly changing, where new content should be added into an article when a related new event happens. However, news articles about the event are usually too long to be inserted into the article directly, so summarization is necessary. When generating a summary, the title and subtitles that represent the topic of the article, and the context up to the inserting location should be considered. In this paper, we focus on topic- and context-focused extractive summarization which extracts valuable sentences from a news article according to a given topic and context, to form a summary to be inserted. As one of the most popular pretrained language models, BERT has been widely used in various natural language processing tasks, and BERT has been proved to greatly improve the performance of single-document extractive summarization. In this paper, we propose a two-step BERT-based model that can encode the topic and context into their representations for guiding generation of a summary. We conduct evaluations of our model on the benchmark dataset WikiCite and achieved the current state-of-the-art performance. |
| 論文抄録(英) |
|
|
内容記述タイプ |
Other |
|
内容記述 |
Wikipedia has been one of the most famous online encyclopedias. Articles' contents in Wikipedia are constantly changing, where new content should be added into an article when a related new event happens. However, news articles about the event are usually too long to be inserted into the article directly, so summarization is necessary. When generating a summary, the title and subtitles that represent the topic of the article, and the context up to the inserting location should be considered. In this paper, we focus on topic- and context-focused extractive summarization which extracts valuable sentences from a news article according to a given topic and context, to form a summary to be inserted. As one of the most popular pretrained language models, BERT has been widely used in various natural language processing tasks, and BERT has been proved to greatly improve the performance of single-document extractive summarization. In this paper, we propose a two-step BERT-based model that can encode the topic and context into their representations for guiding generation of a summary. We conduct evaluations of our model on the benchmark dataset WikiCite and achieved the current state-of-the-art performance. |
| 書誌レコードID |
|
|
収録物識別子タイプ |
NCID |
|
収録物識別子 |
AN10112482 |
| 書誌情報 |
研究報告データベースシステム（DBS）
巻 2020-DBS-172,
号 7,
p. 1-6,
発行日 2020-12-14
|
| ISSN |
|
|
収録物識別子タイプ |
ISSN |
|
収録物識別子 |
2188-871X |
| Notice |
|
|
|
SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc. |
| 出版者 |
|
|
言語 |
ja |
|
出版者 |
情報処理学会 |
IPSJ-DBS20172007.pdf (1.2 MB)
