Text Classification Model Based on Multi-level Topic Feature Extraction
Yongchao Yan, Kai Zheng · 2020
The max pooling strategy of Recurrent Convolutional Neural Networks in text classification can only extract the influence of local features on text semantics, ignore global features. In order to solve that problem, this paper proposes a multiple level topic feature extraction model, which is based on Recurrent Convolutional Neural Networks and Latent Dirichlet Allocation. The Latent Dirichlet Allocation topic model can be used to model text, which can fully consider the global characteristics of the text and obtain the semantic information of different levels of text. Experimental results show that the proposed model is better than the classical Recurrent Convolutional Neural Networks and other models.