Modeling Social Norms Evolution for Personalized Sentiment Classification

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 855–865, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Modeling Social Norms Evolution for Personalized Sentiment Classification Lin Gong1, Mohammad Al Boni2, Hongning Wang1 1Department of Computer Science, 2Department of System and Information Engineering University of Virginia, Charlottesville VA, 22904 USA {lg5bt, ma2sm, hw5x}@virginia.edu Abstract Motivated by the findings in social sci- ence that people’s opinions are diverse and variable while together they are shaped by evolving social norms, we perform person- alized sentiment classification via shared model adaptation over time. In our pro- posed solution, a global sentiment model is constantly updated to capture the ho- mogeneity in which users express opin- ions, while personalized models are simul- taneously adapted from the global model to recognize the heterogeneity of opin- ions from individuals. Global model shar- ing alleviates data sparsity issue, and in- dividualized model adaptation enables ef- ficient online model learning. Extensive experimentations are performed on two large review collections from Amazon and Yelp, and encouraging performance gain is achieved against several state-of-the-art transfer learning and multi-task learning based sentiment classification solutions. 1 Introduction Sentiment is personal; the same sentiment can be expressed in various ways and the same expres- sion might carry distinct polarities across different individuals (Wiebe et al., 2005). Current main- stream solutions of sentiment analysis overlook this fact by focusing on population-level models (Liu, 2012; Pang and Lee, 2008). But the id- iosyncratic and variable ways in which individ- uals communicate their opinions make a global sentiment classifier incompetent and consequently lead to suboptimal opinion mining results. For in- stance, a shared statistical classifier can hardly rec- ognize that in restaurant reviews, the word “expen- sive” may indicate some users’ satisfaction with a restaurant’s quality, although it is generally asso- ciated with negative attitudes. Hence, a person- alized sentiment classification solution is required to achieve fine-grained understanding of individu- als’ distinctive and dynamic opinions and benefit downstream opinion mining applications. Sparse observations of individuals’ opinionated data (Max, 2014) prevent straightforward solu- tions from building personalized sentiment clas- sification models, such as estimating supervised classifiers on a per-user basis. Semi-supervised methods are developed to address the data spar- sity issue. For example, leveraging auxiliary in- formation from user-user and user-document re- lations in transductive learning (Hu et al., 2013; Tan et al., 2011). However, only one global model is estimated there, and the details of how individ- ual users express diverse opinions cannot be cap- tured. More importantly, existing solutions build static sentiment models on historic data; but the means in which a user expresses his/her opinion is changing over time. To capture temporal dynam- ics in a user’s opinions with existing solutions, re- peated model reconstruction is unavoidable, albeit it is prohibitively expensive. As a result, personal- ized sentiment analysis requires effective exploita- tion of users’ own opinionated data and efficient execution of model updates across all users. To address these challenges, we propose to build personalized sentiment classification …

Content truncated to 500 words