Semi-Supervised Recursive Autoencoders for Social Review Spam Detection

Baohua Wang, Junlian Huang, Haihong Zheng, Hui Wu

Abstract: As spam hampers the productivity and performance of social media and causes erosion in the user base and thus associated financial loss, a semi-supervised recursive autoencoders model is applied to social review spams detection problem in this paper. The model is based on semi supervised recursive autoencoders, which learns vector representations of phrases and full sentences as well as their hierarchical structure from the text. This model exploits hierarchical structure and uses compositional semantics to understand meanings, without requiring any language-specific lexica, parsers or knowledge base. Experiments conducted on real dataset show that the approach can effectively detect the social review spams.