| In recent years,with the rapid development of machine learning,especially research in deep learning,many AI products have entered into people's daily life.According to the current development trend,the future AI will continue to exert strength in many positions,at the same time it will produce many new positions.Among them,NLP is an important part of AI technologies,aiming to enhance the "understanding ability" of machines continuously,the question and answer robot is one of the typical applications.Nowadays,a variety of question and answer robots appear in scenes as online shopping,chaperone and intelligent audio.They can make clear the intention of customers and respond appropriately through language communication,which reduces the expenditure on human and material resources of enterprises greatly.In terms of function,the question and answer robots can be divided into two categories: open domain oriented and limited domain oriented.Online education platform is a "gold mine" which is worthy of mining and is oriented to limited fields.The existing online education platforms can meet the basic needs of knowledge dissemination,but there is a general lack of personalized interactive function in the process of teaching and answering question.Question and answer robots can make up for the shortcomings of the existing system to a certain extent.Up to now,there are two ways to realize question and answer robots:(1)knowledge base type of question and answer robots.This type of question and answer robot usually combine synonyms form synonymous phrases.The more similar the meaning,the higher the degree of association between groups.In addition to synonymous relations,other relations also include implications and antonyms.The question and answer robot determines the most appropriate response according to the relationship between words.In addition to relational human-machine question and answer,there are also human-machine question and answer based on grammar tree.This type of question-answering robot will construct a tree structure according to some defined grammar rules to find the most matching questions and answers;(2)deep learning type of question and answer robots,this type of question and answer robots can be roughly divided into two ways: end-to-end and sentence calculation similarity.The end-to-end method will automatically get answers according to the customer's questions,while the way to calculate sentence similarity needs to provide a core question-and-answer library for the robot to choose answers.In this paper,on the basis of the deep learning type of question and answer robot,a combined classifier based on CNN-GRU deep learning is introduced to form a classified question and answer robot system.Taking five key subjects of junior middle school as the object,the model of classified question and answer robot system is built.The whole process of the system is as follows: the input sentences are segmented and deactivated,the words are transformed into word vector form,and the problems are classified into corresponding subjects using CNN-GRU deep learning combined classifier.For each subject,the question database of the corresponding subject is used as the core question library,and the question and answer robot of the subject is constructed based on sentence calculation similarity.When the user interacts with the robot,the system locates a subject according to the input,and then calculates and finds the most similar question in the question-answer library of the subject,and returns the corresponding answer to the user.In this process,the question and answer robot system is equivalent to "laying emphasis first",and then search the corresponding answer.The accuracy of the answer is better than that of the question and answer robot system which checks all subjects' core question answer library.At the same time,the response time is obviously faster than that of the question and answer robot system which checks all subjects' core question answer library. |
PDF Full Text Request