2021 Fiscal Year Final Research Report
Structural extension of knolwedge graph utilizing temporal and semantic analysis of social media
| Project/Area Number |
19K11983
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 60080:Database-related
|
|---|
| Research Institution | Waseda University |
|---|
Principal Investigator |
Iwaihara Mizuho 早稲田大学, 理工学術院(情報生産システム研究科・センター), 教授 (40253538)
|
|---|
| Project Period (FY) |
2019-04-01 – 2022-03-31
|
| Keywords | データマイニング / テキストマイニング / 情報抽出 / 知識グラフ / 時系列分析 |
|---|
| Outline of Final Research Achievements |
Wikipedia is known as the largest social media collecting knowledge, from which knowledge graphs are extracted as computer-readable structured knowledge models. Knowledge graphs are utilized for search result enrichment and various natural language tasks. For developing high-quality knowledge graphs from Wikipedia, structured data such as lists and categories need to be utilized.
In this research, we developed new methods for predicting Wikipedia article pairs that should be merged, and pairs that should have links. For extracting keyphrases from article texts, we developed a method utilizing pretrained language models, improving known records on this task. We also proposed new methods for authorship attribution on tweets, utilizing text sentiment.
|
| Free Research Field |
メディア情報学
|
| Academic Significance and Societal Importance of the Research Achievements |
ウェブからの有用な情報の抽出は，日々生成される膨大なデータを整理分類する基礎的段階を含む．テキスト分類は伝統的に多くの手法が提案されてきたが，新たな形態のテキストとして，Wikipediaの記事の階層的構造や，ツィートのハッシュタグ，さらにこれらの時系列的要素などの課題が出現している．一方，訓練済み学習モデルと呼ばれる深層学習を元にした手法が，従来手法を一変させつつある．本研究では，キーフレーズ抽出，リンク予測，階層的分類等の問題および知識グラフの応用について幅広く研究を行い，いくつかの問題では従来を上回る性能を示すことができた．
|
