Multi-label Chinese question classification based on word2vec

In recent years, as one of the key technologies of question answering system (Q&A), question classification is more and more concerned by researchers. Question classification is to assign category label for a natural language question. Considering the linguistic features of Chinese and the complexity of the question itself, a question usually belongs to multiple categories. That is “multi-label Chinese question classification”. Through multi-label classification on questions, we can get better constraints and filter conditions in retrieval right answers. Therefore, the classification can improve the performance of the Q&A system in a large extent. We proposed a multi-label question classification with label-specific features (MQ-LIFT) model in this paper. A deep learning based tool called word2vec was used to train the Chinese word vector from large-scale corpus, then we obtain the semantic features vector of a Chinese question based on the trained word vector. After that the multi-label classification algorithm can be applied to do the classification job. The average classification accuracy up to 90.28%, which suggests that MQ-LIFT can effectively solved the problem of multi-label Chinese question classification in a Q&A system.