Research And Realization Of Robot Conversation System Based On Convolutional Neural Network

With the deepening of algorithms such as deep learning,human beings have entered the era of artificial intelligence.In 2016,Google's ALphaGo robot defeated the world go champion Li Shishi,which caused a new understanding of artificial intelligence.In the future,artificial intelligence will play an important role in all walks of life,including transportation and medical fields,etc.More and more work will be changed accordingly.Nowadays,all kinds of robots emerge one after another.There are manufacturing robots in industrial production,and chat robots that can interact with people.They have appeared in airports,ports and other places to provide services for people.Through communication,we can not only answer questions for users,but also become friends of human beings in our daily life.The main idea of the dialogue with the robot is that the user raises the problem,the system calculates the problem and feedback the answer to the user.In a simple way,it is a chat system,in which the answers given by the robot are exactly matched with the user's problems.At present,there are two ways to implement a chat system.The first is the chat system based on the knowledge base,which refers to the combination of the synonyms,the formation of synonyms,the establishment of the relationship between the sets,the higher relevance of the words with similar meanings,the existence of antisense,and some other common relations,according to the words.To find the best answers.It also establishes the form of a grammatical tree,which represents the structure of a sentence with a graphic representation,representing the derivation of a sentence,the tree structure formed in a certain rule,and the grammatical structure of the sentence,so as to determine whether a sentence is similar.The other is a chat system based on sentence similarity,which first constructs a chat library,which contains the question answer pair,the user input problem,and the system accountingcalculates which problem is the highest in the chat library,and returns its corresponding answer to the user.This is actually the similarity degree of the two sentences,due to the number of sentences.It is based on the form of question answer pairs,so we only need to find out the most similar questions.On the basis of deep learning,this paper puts forward a chat and dialogue system based on the convolution neural network.Firstly,the words are divided into words,the words are removed and the words are converted into word vector form.The semantic matching vector is obtained by using the words in a sentence and the other sentence,and the semantic matching vector is obtained.Matching vectors are decomposed into similar and dissimilar matrices.Then the two are input to the convolution neural network,and the feature vectors of the sentences are obtained.Finally,the correlation degree of the sentences is calculated by cosine.The weights of the convolution neural network are adjusted continuously by training the similar values of the given sentences until the error is less than a certain range.When the user interacts with the robot,the system will find the highest similarity in the chat library according to the input,and return the answer to the user,which is more accurate than the previous one.