Frame-Based Ontology Population from Text: Models, Systems, and Applications

Date: April 26, 2016
Time: 9:00
Location: Meeting Room Ofek - Polo Scientifico e Tecnologico "Fabio Ferrari" (Building Povo 1, via Sommarive 5 – Povo, Trento)

Speaker 
Francesco Corcoglioniti, University of Trento

Abstract
We focus on Ontology Population from text where the goal is populating an RDF/OWL ABox with semantic frames, i.e., events, situations, and other structured entities reified as ontological instances (e.g., a sell event) connected to related instances via properties specifying their role (e.g., seller). This representation supports n-ary relations and allows leveraging complex NLP techniques such as Semantic Role Labeling (SRL), which annotates frame-like structures in text according to domain-general predicate models (e.g., FrameNet). We contribute to the considered problem from multiple directions. We start by developing an ontology (PreMOn) for predicate models and their mappings to FrameBase, a frame-based ontology. Based on this, we propose an Ontology Population approach (PIKES) for English texts using a SRL-based NLP pipeline, whose output is aligned to DBpedia, Yago, and FrameBase. We represent all the involved data according to an ontology (KEM) enabling navigating from any piece of extracted knowledge to its textual mentions, and propose a scalable storage system (KnowledgeStore) for this model. To support the necessary RDF processing tasks, we propose a non-distributed tool (RDFpro) leveraging streaming and sorting. We apply these solutions in the NewsReader Project and in an Information Retrieval approach (KE4IR) enriching document and query vectors with semantic terms extracted from text.

Contact: Francesco Corcoglioniti, f.corcoglioniti [at] unitn.it