Product Review Summarization by Exploiting Phrase Properties

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1113–1124, Osaka, Japan, December 11-17 2016. Product Review Summarization by Exploiting Phrase Properties Naitong Yu, Minlie Huang, Yuanyuan Shi*, Xiaoyan Zhu State Key Laboratory of Intelligent Technology and Systems Tsinghua National Laboratory for Information Science and Technology Dept. of Computer Science and Technology, Tsinghua University, Beijing 100084, PR China *Samsung R&D Institute China - Beijing ynt12@mails.tsinghua.edu.cn, aihuang@tsinghua.edu.cn yy.shi@samsung.com, zxy-dcs@tsinghua.edu.cn Abstract We propose a phrase-based approach for generating product review summaries. The main idea of our method is to leverage phrase properties to choose a subset of optimal phrases for generating the final summary. Specifically, we exploit two phrase properties, popularity and specificity. Popularity describes how popular the phrase is in the original reviews. Specificity describes how descriptive a phrase is in comparison to generic comments. We formalize the phrase selection procedure as an optimization problem and solve it using integer linear programming (ILP). An aspect-based bigram language model is used for generating the final summary with the selected phrases. Experiments show that our summarizer outperforms the other baselines. 1 Introduction With the growth of the Internet over the decades, e-commerce is becoming more and more popular. Product reviews are helpful for both merchants and customers. Merchants analyze the reviews to get feedback to improve their products. Customers make use of the reviews to get a better understanding of the product. The opinions in the reviews can help them make the final decision. However, the vast availability of such reviews becomes overwhelming to users when there is just too much to digest. Product review summarization is the task to address this problem. It summarizes the large number of reviews and generates a short readable summary which contains the overall rating of the opinions in the reviews. Traditional extractive summarization has been studied for a long time, such as (Hovy and Lin, 1999; Kupiec et al., 1995; Paice, 1990). Recently, there are also a number of studies on abstractive summariza- tion, such as (Banerjee et al., 2015; Bing et al., 2015; Liu et al., 2015). However, applying traditional summarization methods directly on product reviews doesn’t yield satisfying results. This is due to that product review summarization is quite different from traditional extractive summarization. From the per- spective of data size, the number of reviews of a product is often much larger than that of traditional data such as news articles. Another important difference is that sentences in product reviews are usually collo- quial and contain lots of noises. Directly extractive summaries may contain a large number of undesired information. A number of researchers have studied the task of review summarization. (Ganesan et al., 2010) pro- posed a graph-based method for generating ultra concise opinion summaries of products. They used predefined rules for finding valid sub-paths in the graph and converted those sub-paths into sentences. Since the sentence generation was rule-based, their method didn’t provide a well-formed grammatical summary. (Gerani et al., 2014) generated product review summaries by using …

Content truncated to 500 words