COMBINED APPROACH FOR ASPECT TERMEXTRACTION IN ASPECT-BASED SENTIMENTANALYSIS

In the world of online news and social media, there is an ever increasing amount of textual data on the internet, and a large portion of it expresses subjective opinions. Sentiment Analysis (SA) termed as Opinion Mining, which is a method of automatically identify and extracting the subjective sentiments of a particular product or topic. Aspect Based Sentiment Analysis (ABSA) is a sub-field of SA which is used to extract more exact and refined opinions by splitting the text into aspects. The aim of this work is analyzing and implementing the method of Aspect Term Extraction (ATE) from user reviews on laptops and restaurants. The proposed method used Conditional Random Fields (CRF) which is able to optimize the use of features for classification. It is identified that the existing methods for SA failed to extract some implicit aspects in some cases. In this work, we proposed a new set of features to capture more semantic information from text and to improve the representation of a text. In aspect term extraction, these set of features are considered as additional positional features for giving training to the model of CRF. To extract missing implicit aspect terms, a set of rules are proposed and combined with the CRF approach to increase the efficiency of the aspect extraction. Detailed empirical evaluations are performed by experimenting with custom made features along with proposed features using a CRF supervised algorithm to accomplish the task of Aspect Term Extraction in terms of Recall, Precision and F-measure evaluation metrics adopted in the field. The improved performance in aspect term identification is observed by combining the proposed set of rules with the CRF algorithm. The performance of CRF algorithm is observed by comparing with other algorithms such as Decision Tree (DT), Naive Bayes (NB) and KNearest Neighbor (KNN) classifiers. The overall improvement in F1-Score of our proposed model shows an increment of 3% from that of the state-of-the-art methods.