PhD proposal public presentation

14 outubro 2014, 22:12 • Pável Pereira Calado

Tomorrow (15th Oct), instead of the usual lab class, all students are invited to attend the public PhD proposal presentation by the PhD candidate David Batista, occuring in room l, door 20, of Pav. de Informática II at 15:00.

The title is "Semantic Relationship Extraction for Information Discovery".

Abstract:

Relationship Extraction transforms text into structured data by extracting relational triples from textual documents. One way to achieve this is using semi-supervised bootstrapping systems. Such systems are initialized with seed instances of a relationship type, the occurrence of such instances in a document collection are used to generate extraction patterns. Using the extraction patterns the systems extracts new instances, which are used to feed the system again. This is done iteratively un- til certain stop criteria is reached. The main problem with bootstrapping systems is semantic drift, which occurs when the extraction of relationships instances or concepts start to drift away from the semantic meaning of the initial given seeds.

To address this problem I propose a new method to deal with semantic drift by relying on distributional semantics. Distributional semantics capture the relation- ships between words, based on co-occurrences. I proposed a method to explore the relationships among the words that mediate a relationship as way of discarding non-related sentences and decreasing the semantic drift.