Opinion Spam Detection based on Supervised Sentiment Analysis Approach

Abstract

Reading other user's experience on different products and services becomes part of customer's behavior for purchase decision making process nowadays. For this reason, such online resources grow into a target for review spammers with the aim of either boosting their desired products or destroying the reputation of their competitors. Distinguishing between spam and true expressed sentiments is highly challenging due to the fact that this process demands linguistic and grammatical knowledge. Because of the language dependency nature of opinion analysis, natural language processing and opinion mining fields are utilized to overcome the challenges in each language. In this paper, we focus on building up a feature set to be employed with different classifiers as a trustworthy input for opinion spam detection in Persian language. Research on review spam detection in English uncovered some meaningful features so far that we utilized some of them, modify the meaning and usage of some of them to be adapted on Persian language and also add some innovative features accordingly. Our experiments reveal that first Decision Tree and then AdaBoost with the accuracy percentage of 98.67 and 98.00 are the best classifiers for Persian opinion spam detection. Also, the robustness of our extended feature set has been checked in comparison to other feature sets in the task of opinion spam detection.