Fig. 1
From: Reliable knowledge graph fact prediction via reinforcement learning
Illustrations of RL-based KG fact prediction. (a) A training sample and its partial walking space for an RL-based agent. The black relation is a rule head for rules extracted. The blue path is an effective reasoning path that can extract a rule for fact prediction. The green path is an ineffective reasoning path. (b) A prediction triple and its subgraph containing multiple paths linking the head and tail entities of the triple. The blue path matches the rule Athlete_plays_for_team \(\rightarrow\) Team_plays_in_league. Therefore, the prediction relation indicated by the black dashed line is considered true. (c) Another prediction triple and its subgraph containing multiple paths. The orange path matches the rule Athlete_home_stadium \(\rightarrow\) League_stadiums_inverse. Therefore, the prediction relation indicated by the black dashed line is considered true. However, this black dashed line relationshipno with walking mechanisms is a false positive