A Hybrid Bi-Attention Mechanisms and Long Short-Term Memory Model for Chinese Question Classification

The question classification assigns a classify label to each question sentence. These category labels represent the answer type of the question or the intention type of user. The classification of question is not only an important part of the question answer(QA) system, but also the result of the question classification can influences the quality of the QA system directly. At early stages, scholars studied rule-based problem classification methods. Due to the rules were not universal, they were gradually transformed into methods based on machine learning and deep learning. This paper proposes a question classification model (BAL) that combines with bi-attention mechanisms and long short-term memory (LSTM) network. The model proposed in this paper can acquire multi-level text features and recognize the classification of questions better by combining with the attention mechanism of the word vector and the attention mechanism of part-of-speech. In the Chinese question classification data set, compared with common convolutional neural network, common long-short-term memory network, and convolutional neural network based on attention mechanisms, the model proposed in this paper can achieve better classification results.