A Reinforcement Learning Knowledge Graph Reasoning Method Based on Path Quality Discrimination

Abstract

The present invention proposes a knowledge graph reasoning algorithm RLKGR‑PQD based on path quality evaluation. The algorithm includes: improving the benchmark algorithm and adding the path quality assessment module and giving the corresponding overall frame diagram, and then performing a test on the benchmark model and the improved RLKGR‑PQD model on two sets of public data sets (FB15K‑237 and NELL‑995). Experiment, and finally the experimental analysis verifies the effectiveness of the RLKGR‑PQD algorithm, and the experimental results show that the improved algorithm can effectively improve the MRR index in query question answering.

Description

technical field [0001] The invention belongs to the field of natural language processing. Background technique [0002] The mainstream method of knowledge graph reasoning is to infer new facts from the constructed knowledge graph. The reinforcement learning-based knowledge graph reasoning methods MINERVA, MultiHop-KG and RLKGR-CL do not measure the path quality. The reasoning methods for modeling knowledge graphs based on reinforcement learning all have the problem of false paths, that is, there is no practical and high-quality path for training, and the model may be misled by false paths. The Action Drop method proposed by Multihop-KG avoids the agent being misled by the initially found path by masking out a part of the outgoing edges when adopting the action set, forcing the agent to fully explore all possible paths. The Reinforcement Learning Knowledge Graph Reasoning Method Based on Curriculum Learning (RLKGR-CL), which integrates the curriculum learning method on the b...

AI Technical Summary

Problems solved by technology

However, both methods, Multihop-KG and RLKGR-CL, avoid evaluating the quality of the path, so it is impossible to know the quality of the samples used for training from beginning to end.

Images