Context And Topic Aware Question Answering Study Based On Deep Learning

Thanks to the development of big data and deep learning technology,the interaction level of neural network machine question answering system has made great progress,and this field is attracting more and more attention.The great progress of deep learning neural network has prompted the landing progress of question answering machine.With the intelligentization of people's life,intelligent dialogue devices are constantly in the sight of people,providing people with information services,interactive control and chatting entertainment or other functions.This paper mainly research the neural network question answering model based on topic model.the traditional neural network applied to open question answering fields existing many problems,one of which is that it is easy to generate broad and nondistinctive simple reply,because the model can neither perceive context nor supplement external knowledge in the generated reply.In order to enable the question answering neural network to generate rich and diverse responses,this paper studies the question answering model of Seq2 Seq neural network and topic model:(1)LDA is an algorithm of extracting text topic,the LDA model trained by large-scale corpus can extract topic words from each group of dialogue as short text,then these words will be introduced into the question answering model of Seq2 Seq neural network as exogenous knowledge.Therefore in a question answering model,besides learning the question and answer mode of training corpus,it can also dynamically generate replies by using topic words,so as to enrich and diversify the content of the generated responses.(2)By integrating the extracted topic words into the question answering model of Seq2 Seq neural network,we get the topic aware question answering model and then trained it on the self-made Chinese corpus.The traditional question answering model of Seq2 Seq neural network is difficult to perceive the context,in this paper,by adding the joint attention mechanism to the model,which mainly include the bi-directional attention flow mechanism and topic attention mechanism,In this way,the new model can be improved so that the decoder can effectively utilize context semantics and topic semantics when generating answers,and promoting the generated responses to be more context-related and topic-related.(3)Inspired by the processing mechanism between questions and context in machine comprehension,we introduced the bi-directional attention flow network into the question answering model of Seq2 Seq neural network and get the Attention Augment Topic Aware Seq2 Seq model.The input of the bi-directional attention flow mechanism includes the context part and the question part encoded by the Message Encoder.Because different words in context play different roles in the generation of questions,the intermediate vector encoded by the Context-level Encoder pays attention to the importance of different words through the attention mechanism of interrogative sentences.Finally,it flows into the bi-directional attention flow network through the context encoder.Bi-directional attention flow mechanism can capture the relationship between context and question,which enables the decoder to generate more close responses to question.Finally,the model is trained by self-made Chinese corpus,and the effectiveness of the model is verified by comparing the experimental results with other mainstream queation answering models.