Application Research On Question Answering Service Robot In Teaching Field Of Colleges

With the increasing expansion of college enrollment,college teaching consultation work is facing tremendous pressure,student counseling needs are not timely to meet,the college teaching management pressure surge,making the most of colleges set up special teaching consultation,and some secondary colleges also through the school office,BBS,president's mailbox etc,to carry out online and offline teaching consultation services for the majority of students.Generally speaking,there are the following problems in such practice: online and offline consultation all rely on manual recovery,resulting in heavy workload,information asynchrony,poor interaction and timeliness,and can not achieve 7 hours and 24 hours all-weather service.Therefore,service robots emerge as the times require.In order to provide students with adequate and timely teaching consultation services,Meanwhile to reduce the burden of teachers,the author choose the research of college teaching question answering service robot,automatic replies by robots instead of manual process.Assisting the university teaching management organization to complete daily teaching consultation work,reducing and solving the cumbersome consulting burden of its staff,improve the efficiency of teaching management.The aim of this paper is to build a question answering service robot for university teaching management,to implement a software system to assist teachers to complete various teaching consultations,a new and intelligent quiz platform is set up for college students.Therefore,based on natural language processing technology,focusing on the application of text classification and text similarity technology,and FAQ,a question analysis,question retrieval and answer feedback as the basic steps of system,the key steps including "Students propose a text form to robot,robot preprocessing and text representation,short text classification retrieval and return related questions-answer set,short text similarity calculation to select better answers,return the answer to the student".In this process,short text classification model and short text similarity computation are the key technical steps,and also the core work.Considering the above objectives,in order to make a clearer presentation of the results of the application,the author divides the thesis into 5 components:(1)Introduction.It mainly includes the research background and research status of the Q&A system.The research background generalize the significance and importance of the research,and the current research situation is introduced,theindustry research status of Q&A robot is introduced.the research status of short text classification and short text similarity computing technology is reviewed.(2)Text preprocessing and representation.It mainly includes the processing of natural questions to computable forms.This process involves the key technology contents of text preprocessing and text representation.The text preprocessing is for the original question of regularization denoising,word segmentation and POS tagging,stop words filtering;and text representation methods are based on preprocessing,using the vector space model,LDA topic model and word vector model for text representation,provide the basis for the classification and similarity comparison of the following questions.(3)Short text classification method.This paper mainly introduces the input matrix of traditional neural network,which only extracts the word vector of the word granularity level as the feature representation,ignoring the overall semantic features of the text granularity level,this problem will lead to the problem of insufficient representation of text features.Therefore,in this chapter,a text representation matrix combining word vector and topic vector is proposed,which generalizes the semantic and semantic features.Matrix input convolution neural network method to improve the accuracy of short text classification.(4)Short text similarity calculation method.Mainly describes traditional LDA topic similarity calculation method is lack of accuracy problems,put forward a short text similarity algorithm based on multi feature fusion based on LDA,and extract the topic similarity factor and word co-occurrence factor,established policy of expanding and union similarity model,to further improve the accuracy of short text.(5)Q&A service robot implementation.The construction of the FAQ knowledge base,the application of the algorithm and the realization of the prototype system are mainly completed.The construction of the FAQ is based on the 10 year historical data of the president's mailbox of our school computer science,to construct the domain keyword dictionary,two tag clustering algorithm and knowledge database;based on the FAQ knowledge base,the application of classification and similarity algorithm,realizes the algorithm of the system architecture,which Q&A service robot software must ask,realization of prototype system adopt Python WEB technology,the visual simulation system is more intuitionistic.