Query Based Document Summarization

Abstract

The main goal of this research is to explore a better approach to text summarization of the meeting transcripts. Summarization of meetings transcripts is a very interesting research problem from the perspective of meetings occurring on daily basis around the world in almost every organization. Most of the existing document summarization techniques are based on an extractive type of summarization and some of them only deal with single documents. In this research, We propose an effective approach for the abstractive type of summarization of the meetings based on user queries. The user queries can be Generic or Specific in nature. Our proposed approach, first of all, identifies and extracts the relevant content from the meeting transcripts based on the queries from the multi-document as well as from the single document and then generates the summaries of the extracted spans using a transformer and deep learning-based models. Our proposed approach has been validated on three different meeting transcripts datasets. This includes academics, product, and politically related meeting transcripts datasets. Experiments on the QMSum dataset report notable improvement in the quality of the generated summaries as compared to the state-of-the-art