Research And Implementation Of Customer-based Chatbot Based On Specific Domain

With the exciting game of Alpha Go VS chess master Li Shishi,artificial intelligence is once again popular in the world.Coupled with the development of the Internet and the upgrade of speech recognition technology and natural language processing technology,intelligent chatbots are also developing rapidly.and chatbots are also seen as the key to the future of human-computer interaction and the entrance of Internet.However,the development of gossip chatbot has not reached the goal of free communication.Instead,a customer-based chatbots based on specific fields has become popular.This kind of domain-specific customer service chatbot has strong application scenarios and a broad market.Whether in the field of education,finance,or medical,there are a wide range of applications.By using customer service chatbot,efficiency can be improved and human resources can be reduced.And this domain-specific customer service chatbot combines the hottest natural language processing technology with deep neural network technology.This thesis proposes the design of a customer-based chatbot based on a specific domain and the implementation of related algorithms.This is a customer service chatbot that combines multiple rounds of dialogue with a single round of dialogue.The customer service chatbot of this thesis adopts a combination of a variety of natural language processing tasks and deep learning techniques.It supports both the single-round mode of the fixed knowledge base and the multi-round mode of obtaining user information through multiple rounds of dialogue.First,a single round mode subsystems is implemented.This subsystem mainly involves the implementation of key technologies such as Chinese word segmentation,word vector conversion,and similarity matching algorithm.The Chinese word segmentation introduces and implements the open source algorithm Han LP tool.The word vector conversion first crawls the People's Daily corpus resources through the crawler,and then builds and trains the Fast Text model.Finally,the similarity matching algorithm selects the cosine similarity between vectors.At the same time,the single round model service is deployed on multiple server nodes and load balancing is performed by the reverse proxy technology Nginx.Second,a multi-round dialogue subsystem is implemented.The subsystem mainly proposes a multi-round dialogue configuration model,designs and implements an algorithm model such as intent recognition algorithm and entity recognition algorithm.The intent recognition is realized by constructing and training the Text CNN model.The entity recognition is realized by the Bi LSTM-CRF model training.Finally,the system is published as a RESTful API through the Flask framework.This method can easily provide services to the PC or mobile.Finally,this thesis thoroughly tests and verifies the implementation of the algorithm and the construction of the model involved in the customer service robot system.Through the interface test method,a variety of different tests are performed for the system interface call,and each of them is analyzed.In the end,verifies that the system meets the requirements.