Product features extraction of online reviews based on LDA model

Aiming at the problems that low accuracy of product feature extraction, much human participation and difficult to handle the colloquial expression, a new product feature extraction method was proposed based on Latent Dirichlet Allocation (LDA). The online product reviews were parsed and labeled by using Chinese lexical analysis tool to generate the initial nouns set of product feature. The set of candidate product feature words was selected by LDA text training model, and the final product feature set was obtained through synonym lexicon expansion and feature filtering rules. The evaluate data of camera and mobile phone from JD. com was taken as the example to verify the effectiveness of the proposed method.