The Key Technology Research Of Intelligent Chat Robot

Chat robot is one kind of question answering systems, which is characterized by the imitation of language habit of people. A chat robot (Chatbot) can return the most appropriate answer according to the input question by pattern matching methods. For the question which is similar with that in a question-answer database, Chatbot can present a satisfied answer.For the question which is dissimilar with the case in the database, Chatbot can generate a candidate answer for people. This paper presents the emphasis on the answer extraction and generation. The main work in this paper contains the following aspects.Firstly, this paper presents an efficient similarity measure algorithm for Chinese sentences, in which word segmentation operation is avoided. This is because the word segmentation results largely affect the final system accuracy and real-time performance. The proposed method does not need to cut word, while according to the length of consecutively same string of sentences they are given corresponding weight.The longer and the more the consecutively same string, the more similar the two sentences. Experimental results show that our method can quickly return more accurate search results which are more suitable for real-time chat robot application.Secondly, a review on existing mainstream answer extraction methods is given in this paper. Their advantages and disadvantages are analyzed detailedly,then a novel extraction method of candidate answer information points is proposed. In the proposed method, the potential relations of a candidate question and corresponding answer are presented by a matrix model, as well as the potential relations of different candidate questions. Then the question-answer information point pairs can be obtained for sequent answer generation.Thirdly, some main evaluation functions of items are introduced in this article, together with some knowledge on the generation of an answer. Then a novel answer generation strategy is proposed which can be described as follows:(1) Analyzing the relationships between an input question and its corresponding retrieval questions;(2) Obtaining some information point pairs on the input question and the corresponding answer;(3) Applying a scoring mechanism for the answer sentence selection from all candidate ones. Finally, the experiment is performed based on the corpus provided by Intelligent Information Processing Laboratory of Kunming University of Science and Technology. After the analysis and comparation on the experimental results, we can know the proposed answer generation method is promising.