Combining labelled and unlabelled data
Cyril Goutte, Hervé Déjean, Éric Gaussier, Nicola Cancedda, Jean-Michel Renders · 2002
We address the problem of using partially labelled data, eg large collections were only little data is annotated, for extracting biological entities. Our approach relies on a combination of probabilistic models, which we use to model the generation of entities and their context, and kernel machines, which implement powerful categorisers based on a similarity measure and some labelled data. This combination takes the form of the so-called Fisher kernels which implement a similarity based on an underlying probabilistic model. Such kernels are compared with transductive inference, an alternative approach to combining labelled and unlabelled data, again coupled with Support Vector Machines. Experiments are performed on a database of abstracts extracted from Medline.