Research And Implementation Of Intelligent Chatbot Based On Deep Learning

With the introduction of artificial intelligence technology into our lives,chat robots with good human-computer interaction are also constantly being developed.Chat robots have been used in many human-computer interaction scenarios,such as question answering systems,personal assistants,e-commerce,teaching guidance,etc..Nowadays,deep learning technology is widely used in the field of natural language processing,and it is very valuable and promising to study the application of deep learning technology to intelligent chat robot systems.The intelligent chatbot is composed of two parts: a text preprocessing module and a natural language generation module.The text preprocessing module helps the model understand natural language,and the natural language generation module helps the model learn the way of human dialogue.Both modules are built using deep learning models.The text preprocessing module includes data acquisition,data cleaning,and text representation.The natural language generation module improves the shortcomings of the traditional Encoder-Decoder model.The main contents of this article are as follows:1.This article studies the text preprocessing work.The subtitle corpus is selected as the training data,and the garbage data is cleaned.The BERT model is selected to generate word vectors with rich semantics through the study of the principles of various word vector technologies,saving a lot of time and cost.2.This paper studies and improves the long-distance dependence problem of the traditional Encoder-Decoder model.The Attention mechanism is used to provide multiple semantic encodings for the Encoder-Decoder model,so that the model can selectively pay attention to the relevant information in the encoder when generating the sequence.Using the GRU network to build an Encoder-Decoder model allows the model to selectively forget unimportant information.The Bi-GRU network is used to improve the encoder part of the model,while considering the past and future information,so that the encoder can better understand and extract the information of the input sequence.3.This paper studies and improves the security response problem of the traditional Encoder-Decoder model.Using the cluster search algorithm to improve the selection of the decoder in the prediction phase can allow the chatbot to get better and more diverse output.At the end of the article,the construction process of the chatbot model developed based on the Tensor Flow framework is introduced.The model is trained through Chinese movie subtitles,and the generation effects of different structural models are compared,which proves the feasibility of the intelligent chat robot designed in this paper.