Automatic generation of short informative sentiment summaries

Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics, pages 276–285, Avignon, France, April 23 - 27 2012. c⃝2012 Association for Computational Linguistics Automatic generation of short informative sentiment summaries Andrea Glaser and Hinrich Sch¨utze Institute for Natural Language Processing University of Stuttgart, Germany glaseraa@ims.uni-stuttgart.de Abstract In this paper, we define a new type of summary for sentiment analysis: a single- sentence summary that consists of a sup- porting sentence that conveys the overall sentiment of a review as well as a convinc- ing reason for this sentiment. We present a system for extracting supporting sentences from online product reviews, based on a simple and unsupervised method. We de- sign a novel comparative evaluation method for summarization, using a crowdsourcing service. The evaluation shows that our sentence extraction method performs better than a baseline of taking the sentence with the strongest sentiment. 1 Introduction Given the success of work on sentiment analy- sis in NLP, increasing attention is being focused on how to present the results of sentiment analy- sis to the user. In this paper, we address an im- portant use case that has so far been neglected: quick scanning of short summaries of a body of reviews with the purpose of finding a subset of reviews that can be studied in more detail. This use case occurs in companies that want to quickly assess, perhaps on a daily basis, what consumers think about a particular product. One-sentence summaries can be quickly scanned – similar to the summaries that search engines give for search results – and the reviews that contain interesting and new information can then be easily identified. Consumers who want to quickly scan review sum- maries to pick out a few reviews that are helpful for a purchasing decision are a similar use case. For a one-sentence summary to be useful in this context, it must satisfy two different “information needs”: it must convey the sentiment of the re- view, but it must also provide a specific reason for that sentiment, so that the user can make an informed decision as to whether reading the en- tire review is likely to be worth the user’s time – again similar to the purpose of the summary of a web page in search engine results. We call a sentence that satisfies these two crite- ria a supporting sentence. A supporting sentence contains information on the sentiment as well as a specific reason for why the author arrived at this sentiment. Examples for supporting sentences are “The picture quality is very good” or “The bat- tery life is 2 hours”. Non-supporting sentences contain opinions without such reasons such as “I like the camera” or “This camera is not worth the money”. To address use cases of sentiment analysis that involve quick scanning and selective reading of large numbers of reviews, we present a simple un- supervised system in this paper that extracts one supporting sentence per document and show that it is superior …

Content truncated to 500 words