Opinion Spam Classification on Steam Review using Support Vector Machine with Lexicon-Based Features

Abstract

Steam is a video game digital distribution platform developed by Valve Software. Steam provides a user review feature, where users can write about criticism or comments on games that can contain positive or negative sentiments. Based on the questionnaire that the author conducted to Steam users from all over Indonesia, the user review feature provided by Steam was not sufficient. This is because there are fake reviews that allow biased opinions from certain parties so that a phenomenon called review bombing often occurs where users review only to drop or raise the image of a product, not to review it sincerely. From these problems, a solution design is needed that can classify fake reviews on the Steam service. The Support Vector Machine (SVM) classification method was chosen as the model in combination with lexicon-based feature retrieval and Term Frequency – Inverse Document Frequency (TF-IDF) weighting. Of the 236 classification test data conducted by SVM, it produced 105 reviews which were categorized as Valid Reviews. Meanwhile, those categorized as Opinion Spam by SVM are 131 reviews. The accuracy level of the data classification model using Support Vector Machine method is of 81% by dividing training data by 70% and test data by 30% with a random state level of 109. A dashboard in the form of a web application has also been made that contains the classification model to be used for buying reference for Steam user.