Answer Selection And Natural Language Inference Model For Intelligent Question Answering Systems

In the context of an explosive increase in the total amount of information,Internet users need a way to the target information,so that they can more effectively obtain a valuable part of the mass of information,and a question answering system can accomplish this task.At present,question answering systems has been applied in many scenarios.In the open field,it can be used as an enhanced version of the search engine to accurately answer the factual questions entered by users.At the same time,question answering systems are also widely used in customer service in various fields.Compared with the method of hiring a human customer service,the deployment of an online question answering system can effectively reduce labor costs and can maintain a long-term online status.At the same time,deep learning has made rapid progress in many research areas,but the existing question answering systems still has shortcomings in the application of deep learning related methods.Therefore,this article mainly studies how to use related technologies of deep learning to deal with intelligent question answering systems.The specific research work includes the following three parts:(1)Aiming at the task of ranking candidate answers in a question answering system,this paper proposes an answer selection model based on residual depthwise separable convolution and self-attention mechanism.In this model,depthwise separable convolution and multi-head self-attention mechanism are used to encode the question and answer sentences,extract the semantic features in the sentence,and obtain the vector representation of the question and answer.The score is calculated for each candidate answer,and all candidate answers are sorted according to the score.Experimental results show that the model can rank answers more accurately than other deep learning-based answer selection models.(2)Aiming at judging the inconsistency between the user's description in multiple rounds of interaction with a question answering system,this paper proposes a natural language inference model based on hybrid pooling.In this model,sentences are encoded using a bidirectional long-short term memory network,and then the result is pooled by max-pooling and multi-dimensional attention to obtain a vector representation of the sentence,and finally multi-layer forward neural network gets the final classification results.Experimental results show that the model can classify the relationship between two sentences more accurately based on the meaning of the sentence compared with other deep learning-based natural language inference models.(3)The Django framework is used as a server to build an intelligent customer service system for the insurance field.The system has three modules: intelligent question answering,question searching and question list.The intelligent question answering module uses RDSCS proposed in this article as an answer selection model,and provides users with a service that can directly obtain the answer to a question by entering a question.The question search module can search for questions based on keywords,while in the question list module questions are presented to users in the form of a list.This system can help users from multiple aspects to achieve efficient customer service.