What's on in Computing Science?

Date: Thursday, 30 April, 2009
Time: 11:00
Location: Sir Alwyn Williams Building, 422 Seminar Room
[ Inference Seminar ] Using Semantic Spaces to Improve Sentence Classification
Tamara Polajnar
In this talk I will present two semantic models: Hyperspace Analogue to Language (HAL) and Bound Encoding of the Aggregate Language Environment (BEAGLE). I will also show my recent work where I use these models to improve protein interaction sentence classification. I use the models to generate word-word co-occurrence matrices from a large unlabelled data set. The matrices are then integrated into the kernel created from a smaller labelled sentence corpus. The kernel is used to train a Gaussian process (GP) classifier. I found that this procedure significantly increased quality of sentence classification as measured by the F-score and the AUC. These experiments, along with some future directions I would like to discuss with you, will be a large part of my thesis.

Contact: Dr Rónán Daly (rdaly@dcs.gla.ac.uk)

Add to my calendar