Event corpus purification method based on multi-agent reinforcement learning

Abstract

The invention relates to an event corpus purification method based on multi-agent reinforcement learning. The method comprises the following steps: carrying out initialization reset on an environment and agents before model training, and setting corresponding training parameters; enabling the intelligent agent to perform corresponding purification and optimization actions in the environment to form a series of data required by training, sampling the data, and storing in a data cache region for subsequent training; when the data quantity in the data cache region reaches a preset value, starting to use the data to train and update the real networks of all agents; after the real network is updated, updating target networks of all agents through an irregular parameter copying method; and repeating the above steps until the training frequency reaches a preset training frequency. The labeled data is purified and optimized, so that the problem of data label noise encountered in the training process of the sequence labeling joint extraction model is solved, and the effect of the event entity relation joint extraction task is improved.

Description

technical field [0001] The invention relates to the field of multi-agent reinforcement learning methods, in particular to a method for purifying an event corpus based on multi-agent reinforcement learning. Background technique [0002] Reinforcement learning (MARL) is a method of machine learning. According to the number of agents, it can be divided into single-agent reinforcement learning and multi-agent reinforcement learning. Among them, multi-agent reinforcement learning has a wider range of application scenarios and is the solution A key tool for many real-world problems. In multi-agent reinforcement learning, according to the different task relations of agents, it can be divided into: fully cooperative tasks, fully competitive tasks and mixed tasks. Here we only consider fully cooperative tasks. [0003] In the multi-agent reinforcement learning training under fully cooperative tasks, the agent aims at maximizing the joint reward, selects an action according to its ow...

AI Technical Summary

Problems solved by technology

This leads to the problem of label noise in the large amount of labeled data sets generated by distant supervision methods, and these noises will adversely affect the joint extraction model.

Images