Abstract
Nowadays, deep learning has been widely used. In natural language learning,the analysis of complex semantics has been achieved because of its high degreeof flexibility. The deceptive opinions detection is an important applicationarea in deep learning model, and related mechanisms have been given attentionand researched. On-line opinions are quite short, varied types and content. Inorder to effectively identify deceptive opinions, we need to comprehensivelystudy the characteristics of deceptive opinions, and explore novelcharacteristics besides the textual semantics and emotional polarity that havebeen widely used in text analysis. The detection mechanism based on deeplearning has better self-adaptability and can effectively identify all kinds ofdeceptive opinions. In this paper, we optimize the convolution neural networkmodel by embedding the word order characteristics in its convolution layer andpooling layer, which makes convolution neural network more suitable for varioustext classification and deceptive opinions detection. The TensorFlow-basedexperiments demonstrate that the detection mechanism proposed in this paperachieve more accurate deceptive opinion detection results.