CAFE: Knowledge graph completion using neighborhood-aware features

Knowledge Graphs (KGs) currently contain a vast amount of structured information in the form entities and relations. Because KGs are often constructed automatically by means extraction processes, they may miss that was either not present original source or successfully extracted. As result, might lack useful valuable information. Current approaches aim to complete missing have two main drawbacks. First, some dependence on embedded representations, which impose very expensive preprocessing step need be recomputed again as KG grows. Second, others based long random paths cover all relevant information, whereas exhaustively analyzing possible between is time-consuming. In this paper, we an approach evaluating candidate triples using set neighborhood-based features. Our exploits highly connected nature relations surrounding any given pair entities, while avoiding full recomputations new added. results indicate our proposal able identify correct with higher effectiveness than other state-of-the-art approaches, achieving average F1 scores tested datasets. Therefore, conclude vicinities within triple can leveraged determine whether from not.