Summarizing Contrastive Viewpoints in Opinionated Text

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 66–76, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics Summarizing Contrastive Viewpoints in Opinionated Text Michael J. Paul∗ University of Illinois Urbana, IL 61801, USA mjpaul2@illinois.edu ChengXiang Zhai University of Illinois Urbana, IL 61801, USA czhai@cs.uiuc.edu Roxana Girju University of Illinois Urbana, IL 61801, USA girju@illinois.edu Abstract This paper presents a two-stage approach to summarizing multiple contrastive viewpoints in opinionated text. In the first stage, we use an unsupervised probabilistic approach to model and extract multiple viewpoints in text. We experiment with a variety of lexical and syntactic features, yielding significant perfor- mance gains over bag-of-words feature sets. In the second stage, we introduce Compara- tive LexRank, a novel random walk formula- tion to score sentences and pairs of sentences from opposite viewpoints based on both their representativeness of the collection as well as their contrastiveness with each other. Exper- imental results show that the proposed ap- proach can generate informative summaries of viewpoints in opinionated text. 1 Introduction The amount of opinionated text available online has been growing rapidly, increasing the need for sys- tems that can summarize opinions expressed in such text so that a user can easily digest them. In this pa- per, we study how to summarize opinionated text in a such a way that highlights contrast between multi- ple viewpoints, which is a little-studied task. Usually, online opinionated text is generated by multiple people, and thus often contains multi- ple viewpoints about an issue or topic. A view- point/perspective refers to “a mental position from which things are viewed” (cf. WordNet). An opin- ion is usually expressed in association with a partic- ular viewpoint, even though the viewpoint is usually ∗Now at Johns Hopkins University (mpaul@cs.jhu.edu). not explicitly given; for example, a blogger that is in favor of a policy would likely look at the positive aspects of the policy (i.e., positive viewpoint), while someone against the policy would likely empha- size the negative aspects (i.e., negative viewpoint). Moreover, in an opinionated text with diverse opin- ions, the multiple viewpoints taken by opinion hold- ers are often “contrastive”, leading to opposite po- larities. Indeed, such contrast in opinions may be a main driving force behind many online discussions. Futhermore, opinions regarding news events and other short-term issues may quickly emerge and dis- appear. Such opinions may reflect many differ- ent types of viewpoints which cannot be modeled by current systems. For this reason, we believe that a viewpoint summarization system would ben- efit from the ability to extract unlabeled viewpoints without supervision. Even if such clustering has in- accuracies, it could still be a useful starting point for human editors to select representative excerpts. Thus, given a set of opinionated documents about a topic, we aim at automatically extracting and sum- marizing the multiple contrastive viewpoints implic- itly expressed in the opinionated text to facilitate digestion and comparison of different viewpoints. Specifically, we will generate two types of multi- …

Content truncated to 500 words