A Hybrid Co-occurrence and Ranking-based Approach for Detection of Implicit Aspects in Aspect-Based Sentiment Analysis

Abstract

People are increasingly using Web sites and Web services to express their opinions since the inception of Internet. Sentiment analysis is an active research domain that aims at the extraction of sentiments or opinion from the user text, thereby getting the associated sentiment orientation. Although sentiments can be analyzed at document level and phrase level, unfortunately these options are not sufficient for fine-grained analysis of sentiments. Therefore, it makes sense to focus on aspect-based sentiment analysis which is very promising in terms of providing accurate predictions on user sentiments. There is a lot of scope for the research community to provide solutions for various challenges involved in performing sentiment analysis at the aspect level. The primary goal of this work is to extract the implicit aspects from opinionated document using the co-occurrence of aspects with feature indicators and ranking the pair based on their frequency of co-occurrence. As a first step toward achieving this objective, a novel algorithm is proposed to detect the implicit aspects through co-occurrence and ranking. The proposed algorithm is reliable as it uses the association between explicit aspects and sentiment words to detect implicit aspects.