Text Classification Model Based on Dependency Attention Graph Convolution

We propose a text classification model based on dependency graph attention convolution, which improves the model’s accurate recognition of the text category by mining the semantic features exhibited in the structure of dependency relationships, parts of speech, and word roots among multiple words in the text. Firstly, we identify domain keywords in the text by combining LDA and manual screening. Domain keywords are specific words or phrases that describe or define the domain or category to which the text belongs, and these keywords play significant role in text classification. Secondly, we use NLTK to extract the dependency relationships, parts of speech, and word root features of each word in the text, constructing multidimensional word features by identifying different relationships between multiple words. Then, we employ GCN to mine the structural information of these features in the semantic network. Finally, these semantic information are used to determine their text category by discriminator, completing the text classification task. Experimental results indicate that the model we propose has good performance.