A unified graph model for Chinese product review summarization using richer information

  • Authors:

  • He Huang;Chunping Li

  • Affiliations:

  • Tsinghua University, Beijing, China;Tsinghua University, Beijing, China

  • Venue:
  • Proceedings of the First International Workshop on Issues of Sentiment Discovery and Opinion Mining
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

With e-commerce growing rapidly, online product reviews open amounts of studies of extracting useful information from numerous reviews. How to generate informative and concise summaries from reviews automatically has become a critical issue. In this paper, we present a novel unified graph model, composited information graph (CIG), to represent reviews with lexical, topic and together with sentiment information. Based on the model, we propose an automatic approach to address this issue. We use probabilistic methods to model the lexical, topic and sentiment information separately, associate with the discovered information in the CIG model, and generate summaries with a HITS-like algorithm called Mix-HITS considering both the Representativeness and Proportion Approximation. The experiments demonstrate that our method has improved performance over LexRank and ClusterHITS with Chinese and English datasets. Experimental results show that the proposed approach helps to build an effective way towards both the overall and contrastive summarization.