Monday May 6, 2013 at 16h30 in Celestijnenlaan 200A (auditorium 00.225)
On the Importance of Relations for Natural Language Learning
by Mathias Verbeke (PhD student DTAI)
Memory-based learning (MBL) is a technique that is founded in the well-known k-nearest neighbor (kNN) algorithm. Since a lot of natural language learning tasks are characterized by only a few clear generalizations, with many conflicting sub-regularities and exceptions, this approach has proven to be successful for many natural language processing problems. In recent years, learning from structured data has attracted a lot of attention, due to which also the syntactic and semantic relational features have gained importance. In the first part of this seminar, we will therefore present a relational memory-based learning approach that integrates the powerful formalisms of statistical relational learning (SRL) in a memory-based setting.
In the second part of the seminar, we will illustrate the importance of context with an SRL approach to identifying evidence-based medicine categories, a problem in biomedical information retrieval. Evidence-based medicine is an approach to clinical practice whereby clinical decisions are supported by the best available findings gained from scientific research. This requires efficient access to such evidence. To this end, abstracts in evidence-based medicine can be labeled using a set of predefined medical categories, the so-called PICO criteria. Since both structural and sequential information are important for this classification task, we will illustrate the advantage of using an SRL approach for this multiclass multilabel classification problem.