Automatic sentiment-oriented summarization of multi-documents using soft computing

Abstract

This paper presents an automatic sentiment-oriented summarization of multi-documents using soft computing (called ASMUS). It integrates two main phases: sentiment analysis and sentiment summarization. Sentiment analysis phase includes multiple strategies to tackle the following drawbacks: (1) word coverage limit of an individual lexicon; (2) contextual polarity; (3) sentence types, while the sentiment summarization phase is a graph-based ranking model that integrates the sentiment information, statistical and linguistic methods to improve the sentence ranking result. We found that the current methods are suffering from the following problems: (1) they do not consider the semantic and syntactic information in comparison between two sentences when they share the similar bag-of-words (capturing meaning); (2) vocabulary mismatch problem (lexical gaps). Furthermore, ASMUS also considers content coverage and redundancy. We conduct the experiments on the Document Understanding Conference datasets. The results present the excellent outcomes of the ASMUS in sentiment-oriented summarization.