Multi-level knowledge-based approach for implicit aspect identification

摘要

Sentiment analysis or opinion mining is the area of research in Natural Language Processing (NLP) and text mining which deals with the systematic identification of subjective information from user generated text. At a more fine-grained level, aspect-based sentiment analysis focuses on the targets of users’ opinions and determining sentiment orientation of these opinions. Among different tasks of aspect-based sentiment analysis, aspect extraction is the key task which includes extraction of both explicit and implicit aspects. Due to the complexity of implicit aspects, not much effort has been put forward to solve the problem while explicit aspects have been studied extensively in the recent past. Existing approaches for implicit aspect extraction have focused on specific type of aspects and have neglected the actual problem. Therefore, in this paper, we have proposed a multi-level approach which identifies implicit aspects using co-occurrence and similarity-based techniques. This research focuses on the extraction of clues for implicit targets of users’ opinions and identification of true targets of users’ opinions with the help of implicit aspect clues. The proposed approach is divided into two phases: first, several rules are crafted to identify clues for implicit aspects in a review sentence. Secondly, aspects are assigned on the basis of extracted clues using proposed multi-level approach. The proposed model can extract not only implicit aspect clues associated with opinion words but also allocate clues to opinion words where no association is identified. This helps to identify implicit aspects with or without co-occurrences of opinion words with explicit aspects. Experimental evaluation elaborates the importance of implicit aspect clues in the identification of targets of users’ opinions. The proposed approach shows better results as compared with the state-of-the-art approaches for the identification of implicit targets of users’ opinions on a dataset of different product reviews.