Question and answer corpus learning method with reinforcement learning function

Abstract

The invention discloses a question and answer corpus learning method with a reinforcement learning function. The question and answer corpus learning method comprises the following steps: receiving a plurality of questions input by a user through a model A, and respectively outputting corresponding answers for the plurality of questions; forming a question-answer pair by each question in a plurality of questions input by the user and the corresponding answer; scoring the plurality of question and answer pairs through the model B; and carrying out incremental learning on the model A through the plurality of scored question and answer pairs. According to the question and answer corpus learning method with the reinforcement learning function, the model A can be automatically trained, adjusted and updated through reinforcement learning, continuous self-fine adjustment, self-improvement and self-learning of the model A are achieved, in this way, answers output for questions can be optimized, a large amount of manual annotations can be saved, and labor cost is saved.

Description

technical field [0001] The invention relates to a question and answer corpus learning method with a reinforcement learning function. Background technique [0002] At this stage, in the field of intelligent question answering, the core method is to manually mark a small number of samples from large-scale chat corpus and build a model for learning. This method requires a lot of manual labor and has low generalization, which cannot be consistent with the hotspot aging. Contents of the invention [0003] The invention provides a question-and-answer corpus learning method with a reinforcement learning function, which adopts the following technical scheme: [0004] A question and answer corpus learning method with a reinforcement learning function, comprising the following steps: [0005] Receive several questions input by the user through model A and output corresponding answers to the several questions; [0006] Each of the several questions input by the user and its corres...

AI Technical Summary

Problems solved by technology

This method requires a lot of manual labor and has low generalization, which cannot be consistent with the hotspot aging

Images