A Support Vector Machine Classifier for Gene Name Recognition

This summary describes our solution for task 1A of the BioCreAtIvE Challenge Cup 2003. Essentially, we reduce the entity recognition problem to the problem of classifying single words using a Support Vector Machine followed by a term expansion. Our research question is therefore to find those types of features that eventually yield the highest precision and recall. We implemented and evaluated different features and combinations of features, such as n-grams, neighborhood defined by a sliding window, classification results of preceding words, appearance of special characters or digits, or appearance of the word in a dictionary. Multi-word entity names are gathered in a context-sensitive post-processing step. Our best set of features on the training set leads to a precision of 71.4% and a recall of 72.8%, corresponding to an F-measure of 72.1%, for the closed division.