ASPECT CATEGORY POLARITY DETECTION USING MULTI CLASS SUPPORT VECTOR MACHINE WITH LEXICONS BASED FEATURES AND VECTOR BASED FEATURES

Abstract

Sentiment Analysis (SA) is the classification and interpretation of the opinions or emotions within the text data using text analysis methods. Aspect Level Sentiment Analysis (ALSA) is a type of SA technique which uses different types of algorithms to extract the aspects of the entities from the text data and determining the sentiment of the aspects. ALSA is divided into two subtasks such as Aspect Category Detection (ACD) and Aspect Category Polarity Detection (ACPD). ACD determines the aspect categories discussed in each sentence from a set of review sentences and a given predefined set of aspect categories. Aspect categories need not occur as terms in the sentences and are characteristically coarser than the aspect terms. For every review sentence the aspect categories are provided to identify the polarity of aspect category. The objective of ACPD is to identify the polarity such as negative, positive, neutral or conflict of every aspect category mentioned in each sentence. In this paper, the task aspect category polarity detection is addressed. The three types of features such as linguistic features, Lexicons features and Vector Space features are extracted from the reviews. The classifier is trained with these features. The learned model is used to detect the polarities of the categories that are present in the test review sentences. The combination of proposed features with the existing features improves the accuracy of the system. The accuracy of the proposed system has enhanced by 1.3 % compared with best performing system.