Searching and ranking in entity-relationship graphs

The Web bears the potential to become the world’s most comprehensive knowledge base. Organizing information from the Web into entity-relationship graph structures could be a first step towards unleashing this potential. In a second step, the inherent semantics of such structures would have to be exploited by expressive search techniques that go beyond today’s keyword search paradigm. In this realm, as a first contribution of this thesis, we present NAGA (Not Another Google Answer), a new semantic search engine. NAGA provides an expressive, graph-based query language that enables queries with entities and relationships. The results are retrieved based on subgraph matching techniques and ranked by means of a statistical ranking model. As a second contribution, we present STAR (Steiner Tree Approximation in Relationship Graphs), an efficient technique for finding “close” relations (i.e., compact connections) between k(≥ 2) entities of interest in large entity-relationship graphs. Our third contribution is MING (Mining Informative Graphs). MING is an efficient method for retrieving “informative” subgraphs for k(≥ 2) entities of interest from an entity-relationship graph. Intuitively, these would be subgraphs that can explain the relations between the k entities of interest. The knowledge discovery tasks supported by MING have a stronger semantic flavor than the ones supported by STAR. STAR and MING are integrated into the query answering component of the NAGA engine. NAGA itself is a fully implemented prototype system and is part of the YAGONAGA project.