Aspect Based Sentiment Analysis using Sentiment Flow with Local and Non-local Neighbor Information

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2635–2646, Osaka, Japan, December 11-17 2016. Aspect Based Sentiment Analysis using Sentiment Flow with Local and Non-local Neighbor Information Shubham Pateria Samsung R&D Institute Bangalore, India s.pateria@samsung.com Abstract Aspect-level analysis of sentiments contained in a review text is important to reveal a detailed picture of consumer opinions. While a plethora of methods have been traditionally employed for this task, majority focus has been on analyzing only aspect-centered local information. How- ever, incorporating information from non-local neighbor aspects may capture richer context and enhance sentiment prediction. This may especially be helpful to resolve poor prediction due to ambiguities in review text. The context around an aspect can be incorporated using semantic relations within text and inter-label dependencies in the output. On the output side, this becomes a structured prediction task. However, non-local label correlations are computationally heavy and intractable to infer for structured prediction models like Conditional Random Fields (CRF). Moreover, some prior intuition is required to incorporate non-local context. Thus, inspired by previous research on multi-stage prediction1, we propose a two-level model for aspect-based analysis. The proposed model uses predicted probability estimates from first level to incorporate neighbor information in the second level. The model is evaluated on data taken from SemEval Workshops and Bing Liu’s review collection. It shows comparatively better performance against few existing methods. Overall, we get prediction accuracy in a range of 83-88% and almost 3-4 point increment against baseline (first level only) scores. 1 Introduction The voice of consumer is growing stronger. With numerous platforms now available for providing reviews, consumers find it easy to share their opinions and sentiments about a product, service or other subjects. Thus, it becomes essential to analyze such reviews in order to identify consumers preferences and grievances. Sentiment analysis for consumer reviews (or general text) is a prominent research area. Such analysis can be done on various levels - global (collection of text), sentence-level (where sentiment is assigned to one full sentence) or aspect-based. In Aspect Based Sentiment Analysis (ABSA), the problem of interest is to estimate sentiment associated with a specific aspect within a review. An example is given below, Example 1. The movie had a brilliant+ story. The location was awesome+ and I must highly praise+ the camera-work. However, I have to differ regarding the acting. It is hard to comprehend XYZ’s style in such a role. Her on-screen presence is not the usual; I have to say, her act left me in a very bad− mood. The rest of the cast was oko...averageo at best. Here, the text in bold marks aspect and italic text marks sentiment-indicators. Also, (+,-,o) underscore notations indicate positive, negative and neutral sentiments, respectively. Usually, there can be one or several aspects within a single sentence. The sentiment associated with any aspect can mostly be inferred by checking the terms associated with it (e.g., awesome - location). However, this may not be very 1 (Krishnan and Manning, 2006; …

Content truncated to 500 words