Deep context identification of deceptive reviews using word vectors

Abstract

This paper proposes deep context by word vectors for deceptive review identification. The basic idea is that since deceptive reviews and truthful reviews are composed by writers without and with real experience, respectively, there should be different contexts of words used by them. Unlike previous work using the whole text collection to learn the word vectors, we produce two numerical vectors for each word by embedding contexts of words in deceptive and truthful reviews separately. Specifically, we propose a representation method called DCWord (Deep Context representation by Word vectors) to use average word vectors derived from deceptive and truthful contexts, respectively, to represent reviews for further classification. Then, we investigate three classifiers as support vector machine (SVM), simple logistic regression (LR) and back propagation neural network (BPNN) to identify the deceptive reviews. Experimental results on the Spam dataset demonstrate that by using the DCWord representation, SVMand LR have produced comparable performance and they outperform BPNN in deceptive review identification. The outcome of this study provides potential implications for online business intelligence in identifying deceptive reviews.