Electronic Product Feature-Based Sentiment Analysis Using Nu-SVM Method

Abstract

Sentiment in a product online review is useful and influence decision-making a person may take in buying any product as well as that of organization in determining the number of product to produce. In an opinion, reviewer may provide positive and negative reviews at the same time that can be ambiguous. This is because opinion targets are often not the product as a whole; instead they are only part of a product called as feature, which have advantages and disadvantages based on the reviewers point of view. In this paper, the goal is to produce sentiment of a mobile phone opinion based on its feature. Opinion data used in this thesis are in English taken from www.cnet.com. Feature extraction is conducted by searching for phrases that match the dependency relation template, which is followed by feature filtering. The sentiment identification, positive and negative probability value, as well as target class label of the data preparation become the Nu SVM classifier input parameters. In the study of NU SVM, some data are treated as unlabeled data. The evaluation towards sentiment identification obtained from the study shows F1 Measure of 86.25% for positive class and 77.71% for negative class. The accuracy for feature identification, however, is 82%.