Exploiting Coherence for the Simultaneous Discovery of Latent Facets and associated Sentiments

Abstract

Facet-based sentiment analysis involves discovering the latent facets, sentiments and their associations. Traditional facet-based sentiment analysis algorithms typically perform the various tasks in sequence, and fail to take advantage of the mutual reinforcement of the tasks. Additionally, inferring sentiment levels typically requires domain knowledge or human intervention. In this paper, we propose a series of probabilistic models that jointly discover latent facets and sentiment topics, and also order the sentiment topics with respect to a multi-point scale, in a language and domain independent manner. This is achieved by simultaneously capturing both short-range syntactic structure and long range semantic dependencies between the sentiment and facet words. The models further incorporate coherence in reviews, where reviewers dwell on one facet or sentiment level before moving on, for more accurate facet and sentiment discovery. For reviews which are supplemented with ratings, our models automatically order the latent sentiment topics, without requiring seed-words or domain-knowledge. To the best of our knowledge, our work is the first attempt to combine the notions of syntactic and semantic dependencies in the domain of review mining. Further, the concept of facet and sentiment coherence has not been explored earlier either. Extensive experimental results on real world review data show that the proposed models outperform various state of the art baselines for facet-based sentiment analysis.

Feature Specific Sentiment Analysis for Product Reviews

“…al [4] propose a joint sentiment topic model to probabilistically model the set of features and sentiment topics using HMM-LDA. It is an unsupervised system which models the distribution of features and opinions in a review and is thus a generative model.…”

“…Dataset 1 consisted of 500 reviews extracted from the dataset used by Lakkaraju et. al [4]. The extracted data came from 3 domains laptops, camera and printers.…”

“…We also performed comparisons with another state-of-the-art system namely, CFACTS developed by Lakkaraju et. al [4]. Unlike the CFACTS system, our system has a much less data requirement as it does not train on domain-specific data.…”

“…We achieved a high accuracy across all domains over the baseline. We compare our approach with state-of-the-art systems [4] where we achieve a comparable accuracy despite data limitations. The system performance improved greatly not only over the naïve baseline but also over the chosen improved baseline [5].…”

Sentiment Topic Model with Decomposed Prior

“…As a result, it is of great value to automatically analyze the reviews to extract topics, sentiments and the associations between them. This problem received surging attention in both academic and industry recently [15,19,13,5,12,10,16,14,17]. The problem is different from traditional sentiment classification [18], where only the overall sentiment of a review (i.e., document-level sentiment) is cared about.…”