Research And Implementation Of Multi-labels Text Classification Via Deep Learning

Multi-label text classification is a core research field of natural language processing and a key technology for intelligent systems.These intelligent systems such as retrieval systems,dialogue systems and question answering systems,need to use multi-label text classification technology.In these complex applications,the content maybe long text,short text or carry contextual information.These problems exist in the application such as the number of labels is huge,the class sample is imbalanced and dependency between labels.For example,the intention classification in multi-turn dialogue systems is a typical multi-label text classification problem.The characteristics of the dialogue data include short text,contextual information,imbalance and dependency between labels.The traditional machine learning methods are hard to handle the short text and are more difficult to analyze the short text with contextual information.Deep learning plays an important role in automatic learning and knowledge representation.And it has good modeling ability.Aiming at solving the multi-label text classification problem in the intention classification of the multi-turn dialogue systems,we proposed a multi-label text classification model based on deep learning.The main research works are as follows:1)We proposed to use the parameter attention mechanism to extract context information.The parametric attention mechanism can not only extract similar information from context,but also extract correlation information.2)We proposed to add a forget gate which can control context information.The forget gate can improve the generality of the model by controlling the information and avoid the classification of noise.3)Using the open source dialogue dataset published by Microsoft,the performance evaluation and comparison of the original model and the improved model are carried out.And the results of these experiments are analyzed from various aspects.Experimental results show that the proposed model achieves better classification results on the dataset.