SVM Approach to GeneRIF Annotation

In the biological domain, to extract the newly discovered functional features from massive literature is a major challenging issue. To automatically annotate GeneRIF in a new literature is the main goal in this paper. We try to find function words and introducers in the training corpus, and then apply such informative words to annotate the GeneRIF. The experiments showed that 48.15%, 49.78%, 32.31%, and 35.63% for the measure of Classic Dice, Modified unigram Dice, Modified bigram Dice, and Modified bigram Dice phrases. After applying SVM learning mechanism combing new weighting scheme and position information, we get much better performance.