A Dialogue Assisting System Based on Team Learning and Hierarchical Reasoning

Abstract

The invention discloses a dialogue auxiliary system for team learning and hierarchical reasoning. This method first crawls and generates a pan-industry knowledge graph, which includes general entities and attribute entries outside the industry, and can conveniently meet the needs of users for general knowledge queries. It is necessary, and at the same time, it is also conducive to the rapid migration of knowledge bases between industries; secondly, the multi-level reasoning network proposed by the present invention can realize complex semantic reasoning capabilities, and can realize man-machine dialogue based on the reasoning process. business recommendation and product marketing; finally, the policy learning network based on reinforcement learning can use the accumulated historical interaction experience to learn sorting strategies and continuously improve user experience. By deploying this system, the function expansion and effect upgrade of the original dialogue system can be realized, and the user experience can be improved.

Description

technical field [0001] The invention relates to a dialogue assistance system based on team learning and hierarchical reasoning, which belongs to the field of human-computer interaction in artificial intelligence. Background technique [0002] The traditional dialogue system is mainly based on knowledge engineering to build a knowledge tree. During the dialogue process, the method of searching and matching the knowledge tree is used to obtain the answers to the edited questions. The multi-round interaction process is mainly based on the matching of some words and entity slot values. This kind of dialogue system can meet the needs of self-service voice interaction in vertical industries to a certain extent, but there are still several problems: First, the common-sense question-and-answer ability of this kind of dialogue system is relatively weak, because the knowledge tree is manually edited and established Yes, so the questions that the robot can answer are limited to the ran...

AI Technical Summary

Problems solved by technology

This kind of dialogue system can meet the needs of self-service voice interaction in vertical industries to a certain extent, but there are still several problems: First, the common-sense question-and-answer ability of this kind of dialogue system is relatively weak, because the knowledge tree is manually edited and established Yes, so the questions that the robot can answer are limited to the range of manual presets; second, in daily business handling, customers sometimes raise some dialogues that require reasoning, such as recommending financial products based on risk preferences, and choosing products based on users’ emotions. Correspondingly speaking, the traditional dialogue system cannot carry out this kind of inferential dialogue; the third point is that it cannot adaptively learn the sorting strategy and dialogue of multiple-choice answers based on parameters such as the user's emotional feedback, the number of interaction rounds, and service satisfaction. strategy; the fourth point, a single machine has fewer dialogue logs and cannot provide effective learning samples

Images