Method for optimizing question and answer models on basis of adversarial network reinforcement learning

Abstract

The invention discloses a method for optimizing question and answer models on the basis of adversarial network reinforcement learning. Principles for the method for optimizing the question and answer models include that adversarial mechanisms are introduced into multi-question and single-answer problems and single-question and multi-answer problems in knowledge bases, in other words, question and answer interaction can be implemented by alternate questions and answers of two intelligent question and answer systems, and the intelligent question and answer system models can be ultimately optimized on the basis of reinforcement learning mechanisms and are provided with reward system models. The method has the advantages that the method is reasonable in design, and optimization indexes for adversarial question and answer interaction and modes for computing the optimization indexes of easy responsiveness, content diversity, theme evolution properties, semantic continuity and the like are defined; reward functions for optimizing the adversarial question and answer models are further defined, learning can be reinforced, the question and answer models can be continuously optimized, the question and answer interaction quality can be improved, and the experience of users can be enhanced.

Description

technical field [0001] The invention belongs to the field of computer programs, and more specifically relates to a question answering model optimization method based on confrontational network reinforcement learning. Background technique [0002] In today's society, information technology is developing rapidly. As people's urgent demand for information retrieval increases in the information society, ordinary information retrieval systems can no longer meet the needs of users, and intelligent question answering systems developed on the basis of information retrieval technology can meet people's needs. The intelligent question answering system allows the user to input a question in the form of natural language, and finally returns to the user a short and accurate answer in the form of natural language. [0003] Automatic question answering is a research direction that has attracted much attention in the fields of natural language processing and information retrieval. Automat...

AI Technical Summary

Problems solved by technology

[0004] The recently appeared dialogue generation neural network makes the construction of the question answering model a step further, but the current neural network implementations have certain limitations, that is, they only consider how to generate the next sentence response at the current moment, ignoring its impact on the subsequent development of the entire question answering interaction. The impact of the question and answer, and the sustainable interactivity of question answering is very critical for an intelligent question answering system

Images