A Systematic Literature Review of Reinforcement Learning-based Knowledge Graph Research
Zifang Tang,Tong Li,Di Wu,Junrui Liu,Zhen Yang
DOI: https://doi.org/10.1016/j.eswa.2023.121880
IF: 8.5
2023-10-12
Expert Systems with Applications
Abstract:Knowledge graphs (KGs) model entities or concepts and their relations in a structural manner. The incompleteness has turned out to be the main challenge that hinders the application of KGs. Recently, reinforcement learning (RL) has been recognized as an effective method to deal with such a challenge, which models research tasks into a sequence decision problem without labels. Although an increasing number of studies investigate and analyze KGs using RL, there lacks a systematic literature review that comprehensively and quantitatively analyzes the landscape of RL-based KG research (RL-KG for short). As a result, researchers may have encountered difficulties in appropriately adopting RL techniques in KG research, even reinventing the wheels. In this paper, we follow the Systematic Literature Review (SLR) methodology to survey, screen, and investigate papers of RL-KG. Specifically, we identify 109 highly related papers from 1542, and systematically investigate them with regard to the following five research questions: (1) to what extent RL-KG have been investigated; (2) what application domains have been covered; (3) what RL techniques have been mainly considered; (4) whether there is a connection between the influence and reproducibility of these papers; (5) what specialized datasets, evaluation metrics, and publication venues have been applied. Through an in-depth analysis of the review results, we systematically and comprehensively identify some significant phenomena and analyze the reasons and difficulties of these phenomena. Based on such analysis, we tentatively propose promising future research topics to promote the RL-KG.
computer science, artificial intelligence,engineering, electrical & electronic,operations research & management science