Opinion Summarization with Integer Linear Programming Formulation for Sentence Extraction and Ordering

摘要

In this paper we propose a novel algorithm for opinion summarization that takes ac-count of content and coherence, simulta-neously. We consider a summary as a se-quence of sentences and directly acquire the optimum sequence from multiple re-view documents by extracting and order-ing the sentences. We achieve this with a novel Integer Linear Programming (ILP) formulation. Our proposed formulation is a powerful mixture of the Maximum Cov-erage Problem and the Traveling Sales-man Problem, and is widely applicable to text generation and summarization tasks. We score each candidate sequence accord-ing to its content and coherence. Since our research goal is to summarize reviews, the content score is defined by opinions and the coherence score is developed in training against the review document cor-pus. We evaluate our method using the reviews of commodities and restaurants. Our method outperforms existing opinion summarizers as indicated by its ROUGE score. We also report the results of human readability experiments.