Abstract:
Text is the most widely used information carrier in human society, and it is of great practical significance to classify it accurately. The existing methods have achieved some results in the problem of multi-label text classification, but there are still problems of insufficient use of documents and label clues. From the perspective of text polysemy, this paper proposes a clustering perceptual text multi-label classification model. Firstly, the deep learning model is used to obtain the original features of document. Then, multiple cluster center vectors combined with attention mechanism are used to extract text features in different contexts. Finally, these features are combined with the embedded representation of tags to do the dot product for classification. Experimental results in 4 data sets show that the performance of this method has been significantly improved in several evaluation indexes.