TRFR: A ternary relation link prediction framework on Knowledge graphs

Artificial intelligence has been widely used in daily wireless networks for its flexibility and adaptability in solving extremely complex problems in real-time. These wireless network applications using artificial intelligence will generate a lot of valuable data. Knowledge graph is able to mine and store useful information from these large amounts of data and make data-driven wireless network applications more intelligent. Existing knowledge graph represents knowledge facts in triples whose examinations are restricted within two entities and binary relations. This nature induces the weakness of expanding to complex learning scenarios with multi-entity relations, which motivates the research insights into the n-ary relation. However, existing n-ary relation link prediction all embedding-based which are limited by their interpretability. We investigate the ternary relation link prediction task and propose a novel unified framework TRFR which is the first path-based model on n-ary relational data. TRFR incorporates the hierarchical attention mechanism and reinforcement learning technique. In addition, we release a newly constructed dataset NELL-995-3 to fill the shortage of learning resources for n-ary link prediction. Extensive experiments demonstrate the superiority of the proposed framework compared with a wide variety of state-of-the-art knowledge graph-based approaches in ternary relation link prediction task on one benchmark dataset and one newly constructed dataset.