USING MULTI-CLASS SUPPORT VECTOR MACHINE MONITORING ALGORITHM IN BUILDING CHATBOT SYSTEM FOR VIETNAMESE

Abstract

Chatbot can be defined as AI based computer program that simulates human conversations. The Chatbot system is designed, implemented and tested in this thesis. The system responds to the user's question through two main steps: analyzing the user's question and looking for answer to respond. In particular, we use the technique of BoW (Bag of Words) in combination with the method TF-IDF (Term Frequency - Inverse Document Frequency) to build feature vector of Vietnamese text sentences, use Multi-Class SVM algorithm to train and perform classification. Finally we compare the accuracy with kNN algorithms. The bot understands user intent through language similarity the meaning between the input question and the answer-space set is used in the training step. The numberical test results on a real dataset with 1325 intent corresponding to 175 questions collected from 3 websites : “Phòng công tác sinh viên”, “Tuyển Sinh”, “Phòng tài chính” and “ Quy định công tác học vụ” show that our model achieves an accuracy 78.83%. The system has simple interface for convenient communication.