Abstractive Neural Multi-document Summarization Considering Cross Document Structure

2023 Fiscal Year Final Research Report

• Project/Area Number: 21H03495
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Single-year Grants
• Section: 一般
• Review Section: Basic Section 61030:Intelligent informatics-related
• Research Institution: Tokyo Institute of Technology
• Principal Investigator: Okumura Manabu 東京工業大学, 科学技術創成研究院, 教授 (60214079)
• Co-Investigator(Kenkyū-buntansha): 上垣外 英剛 奈良先端科学技術大学院大学, 先端科学技術研究科, 准教授 (40817649)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: 自然言語処理 / 複数テキスト要約 / ニューラルモデル / 生成型要約 / 文書横断文間関係

Outline of Final Research Achievements

In document structure analysis, that analyzes the relationships between sentences, by utilizing large language models (LLMs), we proposed a method to imitate shift-reduce operations through prompts. As a result of evaluation experiments, the proposed method achieved the state-of-the art performance. In text summarization, we proposed a neural model that utilizes the results of this document structure analysis. We confirmed that this contributes to improving
the performance of the summarization. We further proposed a method for enabling the model to understand the summarization-specific information by predicting the summary length in the encoder and generating a summary of the predicted length in the decoder in fine-tuning. We confirmed that this also contributes to improving the performance.

Free Research Field

知能情報学

Academic Significance and Societal Importance of the Research Achievements

文間の関係を解析する文書構造解析器は，我々のグループが世界最高性能を達成していたが，引き続き研究開発を継続し，新しい手法を提案することで，現在も世界最高性能を維持している．テキスト要約において要約長を予測するというアイデアはこれまでに提唱されておらず，そういう意味で斬新なアイデアに基づいており，しかも，要約長を予測するよう要約モデルを学習することで性能向上に寄与することを示しており，学術的な意義は大きい．