Reasoning for Local Graph Over Knowledge Graph With a Multi-Policy Agent

The reinforcement learning framework for multi-hop relational paths is one of the effective methods solving knowledge graph incompletion. However, these models are associated with limited performances attributed to delayed rewards and far-fetched search trajectories. To overcome challenges, we propose searching window multi-policy agent. provides a large action space, so that agent can backtrack based on newly obtained information establish local instead path chain. Based window, double long short-term memory (DBL-LSTM) policy network introduced encode relation sequence, after which encoding used by select correct entity grow graph. Furthermore, separately infers through three different networks, then, all graphs integrated into an information-rich Experiments using WN18RR dataset revealed reasoning had greater than reasoning, proposed DBL-LSTM improved HITS@N(N = 1,3,5,10) compared prior works, achieved higher hit rates single-policy