A Graph-Based Approach to Abstractive Summarization of Highly Redundant Comparative Opinions

This paper studies the problem of identifying and summarizing comparative opinions in text documents. Identifying comparative sentences is useful in practice because direct comparisons are perhaps one of the most convincing ways of evaluation, which may even be more important than opinions on each individual object. This paper proposes to study the comparative sentence identification and summarization problem. It first categorizes comparative sentences into different types, and then uses a pattern discovery approach to identifying comparative sentences from text documents. Then it use a graph based approach to fusion related opinions and select major opinions to create summary. Evaluation results on summarizing user reviews show that our method summaries have better agreement with human summaries compared to the baseline extractive method. The summaries are readable, reasonably well-formed and informative enough to convey the major comparative opinions. We present a comparative opinions summarization based on a graph-based summarization framework. It identifies comparative opinions from highly redundant comparative opinions and generates concise abstractive summaries of them. Evaluation results on summarizing user reviews show that our system summaries have better agreement with human summaries compared to the baseline extractive method. The summaries are readable, reasonably well-formed and informative enough to convey the major opinions, but they have problems like incomprehensible, grammar and conflicting information

Keywords:

Comparative Opinions Summarization , Abstractive Summarization , Textual Graph