You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Mar 27, 2022. It is now read-only.
Graph-to-Sequenceアプローチによる学術論文のアブストラクト生成
近年の要約タスクでは、文書と要約文のペアをsequence-to-sequenceをベースとしたモデルによって学習させる方法が主流となっている。しかしながら、文書中の全文を入力として与えるのは計算資源や計算時間の制約から困難なため、文書の前半の一定文字数のみを用いたり、重要文抽出を行う等の前処理を行う必要があった。本研究では、文書からbigramを抽出し、文書を「各単語をノードにとるグラフ構造」に変換した上でgraph2seqの入力として与える。結果として、BiLSTMベースの要約モデルに比べ性能改善できることを示した。
文献情報
The text was updated successfully, but these errors were encountered: