Entity Profiling with Graph Rules

Zhenzhen Mai, Wenjun Wang, Xueli Liu, Xiaoyang Feng, Bowen Dong

Abstract: This paper proposes a class of Graph Entity Profiling Rules, denoted as EPRs, for profiling entities. An EPR  consists of a graph pattern and a dependency to capture topological and semantic constraints in graphs. As opposed to the previous rules, EPRs  specify rules for both entity resolution and association deduction to enrich the semantics for entities in graphs, and it unified machine learning (ML) and rule-based methods by embedding ML models for ER and link prediction as predicates in logic rules. To apply EPRs, one challenging problem is the interaction between association deductions and entity resolution. We formalize the interaction process of EPRs  with chase and prove its Church-Rosser property. Then we propose an efficient algorithm for entity profiling by chasing EPRs. Experiments on real-life datasets demonstrate the effectiveness and efficiency of our algorithm.

External IDs:dblp:conf/dasfaa/MaiWLFD24