Knowledge Graph Completion Using Embedding Models and Neo4j

A knowledge graph stores facts as (head, relation, tail) and is always incomplete. This project predicts missing facts on CoDEx-S, CoDEx-M and WN18RR and compares simple predictors, rules and embedding models under one filtered ranking protocol with average-rank ties. CoDEx-S: best filtered MRR on the test triples is ComplEx (0.436 (0.005), Hits@10 0.627); CoDEx-M: best filtered MRR on the test triples is AnyBURL (0.318 (0.000), Hits@10 0.452); WN18RR: best filtered MRR on the test triples is AnyBURL (0.484 (0.001), Hits@10 0.556). A leakage check with WN18 as positive control shows how much of each benchmark can be read off the training graph; hard negatives, per-relation tables, predicted new facts, Neo4j and a web application complete the study.

Category: Information Science and Technology

Information Science & Technology Project IST06 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.