Exploiting Function Words Feature in Classifying Deceptive and Truthful Reviews

Abstract

In this paper, we exploit function words as a feature for classifying deceptive and truthful opinions in the online reviews. Initially, we employ function words as a single feature to classify deceptive and truthful reviews. Then, we combine those function words with widely used text features in this task, namely, word and character n-grams, as our proposed combination feature. The experimental results exhibit that our proposed feature could perform well in this deceptive and truthful reviews classification. In particular, the obtained results of our proposed combination feature could outperform those using the baseline feature, i.e., word and character n-grams combinations. In addition, we apply the feature attribute selection, that is, Information Gain in conjunction with Ranker of WEKA, on our proposed combination feature. This treatment is to examine further the robustness of our proposed combination feature when dealing with this task. The latter experiment shows that after applying the feature attribute selection, our proposed combination feature was able to increase the classification results.