Author-aware Aspect Topic Sentiment Model to Retrieve Supporting Opinions from Reviews

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 472–481 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Author-aware Aspect Topic Sentiment Model to Retrieve Supporting Opinions from Reviews Lahari Poddar, Wynne Hsu, Mong Li Lee School of Computing National University of Singapore {lahari, whsu, leeml}@comp.nus.edu.sg Abstract User generated content about products and services in the form of reviews are of- ten diverse and even contradictory. This makes it difficult for users to know if an opinion in a review is prevalent or bi- ased. We study the problem of search- ing for supporting opinions in the con- text of reviews. We propose a framework called SURF, that first identifies opinions expressed in a review, and then finds sim- ilar opinions from other reviews. We de- sign a novel probabilistic graphical model that captures opinions as a combination of aspect, topic and sentiment dimensions, takes into account the preferences of indi- vidual authors, as well as the quality of the entity under review, and encodes the flow of thoughts in a review by constraining the aspect distribution dynamically among successive review segments. We derive a similarity measure that considers both lex- ical and semantic similarity to find sup- porting opinions. Experiments on TripAd- visor hotel reviews and Yelp restaurant re- views show that our model outperforms existing methods for modeling opinions, and the proposed framework is effective in finding supporting opinions. 1 Introduction In order to make an informed decision when book- ing a hotel online, a user will often read through its reviews looking for specific feedbacks. For ex- ample, if he or she plans to do an early check-in and comes across a review that mentions a hassle- free early check-in as shown in Figure 1, it will be helpful to know whether other guests had sim- ilar experiences. If a review complains about bed Figure 1: A sample hotel review bugs or noise from construction nearby, then it is important to know if that was an occasional prob- lem based on a single user’s experience or happens frequently. However, it is impossible for an indi- vidual to go through the large volume of reviews to verify whether an opinion is prevalent. In this work, we study the problem of finding supporting sentences from reviews that corrobo- rate the opinions expressed in a target review sen- tence. This is useful as it enables users to easily look for appropriate comments on the specific is- sues they are interested in. A review is a collection of sentences where each sentence may have multiple segments separated by punctuations or conjunctions. Each segment ex- presses an opinion that can be represented as a combination of aspect, topic and sentiment. An aspect refers to the overall theme of a segment, a topic is the specific subject or issue discussed and the sentiment for each topic can be neutral, posi- tive or negative. Table 1 shows the segments and the possible latent aspect, topics and sentiment for …

Content truncated to 500 words