Design And Implementation Of Question Answering System Based On Search Engine

With the rapid development and popularization of Internet,it has become a tool to access information and communicate with others.In spite of enriching the Internet content and leading us towards the knowledge era,the Internet traps us into the dilemma of information.Although traditional search engines and existing question answering systems alleviate the contradiction between user needs and information supply,On one hand,based on query requests submitted by users,search engines return a collection of long related texts,then the user still needs to manually retrieve out the answers.On the other hand,existing question answering systems are always based on a knowledge base.An exceptionally large knowledge base needs to be constructed and maintained.To address this problem,we design and implement an open-oriented QA system based on search engine.This system is divided into six sub-modules:Web service,question analysis,information retrieval,answer mining,answer merging and answer ranking.By analyzing the user's intentions,it digs out all candidate answers from the vast amount of related texts,then after merging sematic similar elments,ranks the remaining ones to output the top-1 answer.Besides,we propose a Chinese question classification algorithm named MGE-CQC and a question answering matching algorithm named MQAMA.As for MGE-CQC,firstly,it carries out segmentation,part of speech tagging,syntactic structure analysis on raw question to mine purified features of each granularity;then passes the fine-grained vector representation of the question to the coarse-grained feature vector layer by layer,finally output question type through softmax function.Differ from MGE-CQC,MQAMA does not carry out syntactic structural analysis during feature engineering and directly learns the raw question features of each granularity.Then we employ an attention-based network to distinguish the importance of each question term and output the final ranking score through sigmoid function.According to the results of algorithm evaluation and system test,the two algorithms help to improve the accuracy of question answering,also the practicality of the system has also been verified.