A Hybrid Algorithm for Text Classification Based on CNN-BLSTM with Attention

Abstract

We propose an effective text classification framework, which is the hybrid of different weights of character-level and word-level features through concatenation based on Convolutional Neural Network-bidirectional long short-term memory with attention (BACNN). The first step is word segmentation or character segmentation in the process of Chinese natural language processing. However, due to the different semantic relations in Chinese, Chinese sentences usually have several ways of word segmentation, which leads to the problem of word segmentation ambiguity. Although Chinese character segmentation is not ambiguity, its meaning is not accurate and rich enough. And in different situations, the character and word are different in importance. Therefore, to overcome the above problems, we propose the method of hybrid different weights of word-level and character-level features to let them make up the respective shortcomings. The experiment results indicate that our proposed method is better than the simple word or character level feature in classification performance.