Review Sentiment Scoring via a Parse-and-Paraphrase Paradigm

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 161–169, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Review Sentiment Scoring via a Parse-and-Paraphrase Paradigm Jingjing Liu, Stephanie Seneff MIT Computer Science & Artificial Intelligence Laboratory 32 Vassar Street, Cambridge, MA 02139 {jingl, seneff}@csail.mit.edu Abstract This paper presents a parse-and-paraphrase pa- radigm to assess the degrees of sentiment for product reviews. Sentiment identification has been well studied; however, most previous work provides binary polarities only (positive and negative), and the polarity of sentiment is simply reversed when a negation is detected. The extraction of lexical features such as uni- gram/bigram also complicates the sentiment classification task, as linguistic structure such as implicit long-distance dependency is often disregarded. In this paper, we propose an ap- proach to extracting adverb-adjective-noun phrases based on clause structure obtained by parsing sentences into a hierarchical represen- tation. We also propose a robust general solu- tion for modeling the contribution of adver- bials and negation to the score for degree of sentiment. In an application involving extract- ing aspect-based pros and cons from restaurant reviews, we obtained a 45% relative improve- ment in recall through the use of parsing me- thods, while also improving precision. 1 Introduction Online product reviews have provided an exten- sive collection of free-style texts as well as prod- uct ratings prepared by general users, which in return provide grassroots contributions to users interested in a particular product or service as assistance. Yet, valuable as they are, free-style reviews contain much noisy data and are tedious to read through in order to reach an overall con- clusion. Thus, we conducted this study to auto- matically process and evaluate product reviews in order to generate both numerical evaluation and textual summaries of users’ opinions, with the ultimate goal of adding value to real systems such as a restaurant-guide dialogue system. Sentiment summarization has been well stu- died in the past decade (Turney, 2002; Pang et al., 2002; Dave et al., 2003; Hu and Liu, 2004a, 2004b; Carenini et al., 2006; Liu et al., 2007). The polarity of users’ sentiments in each seg- ment of review texts is extracted, and the polari- ties of individual sentiments are aggregated among all the sentences/segments of texts to give a numerical scaling on sentiment orientation. Most of the work done for sentiment analysis so far has employed shallow parsing features such as part-of-speech tagging. Frequent adjec- tives and nouns/noun phrases are extracted as opinion words and representative product fea- tures. However, the linguistic structure of the sentence is usually not taken into consideration. High level linguistic features, if well utilized and accurately extracted, can provide much insight into the semantic meaning of user opinions and contribute to the task of sentiment identification. Furthermore, in addition to adjectives and nouns, adverbials and negation also play an im- portant role in determining the degree of the orientation level. For example, “very good” and “good” certainly express different degrees of positive sentiment. Also, in previous studies, …

Content truncated to 500 words