| Title | : | Predicate Similarity and Mining Property Axioms in Linked data |
| Speaker | : | Rajeev Irny (IITM) |
| Details | : | Fri, 11 May, 2018 3:00 PM @ A M Turing Hall |
| Abstract: | : | Recent advances in Information Extraction techniques have led to the creation of large cross-domain, multipurpose linked datasets like DBPedia, YAGO, NELL etc. The ontologies associated with these datasets conceptualize a part of the the domain information in the form of Concept and Property Axioms. These axioms are useful in several downstream applications like Query Answering, Entity Summarization, Automated Reasoning etc. However, our study of popular linked datasets and their associated ontologies has shown that they contain very few inverse and symmetric property axioms. To this end, we propose PRAXIS; a data-driven, schema-agnostic, unsupervised method to discover predicates that are potentially inverses of each other as well potential symmetric predicates from linked datasets. We also introduce a semantic similarity-measure called Predicate Semantic Similarity (PSS) to aid in extracting useful property axioms. PSS leverages the semantic information present in linked data and helps to rank-order the candidate axioms. PSS computes similarity along two facets, namely context and relevance. Evaluation results show that PSS out-performs existing similarity measures. Qualitative and quantitative evaluations show that PRAXIS discovers latent property axioms at a greater accuracy compared to the state-of-the-art method. |
