Automatic Classification Of Semantic Relation Between Nominals

The task of classifying semantic relations between two nominals in one sentence is important and challenging. Usually, the supervised classification method based on perspective features and effective classification algorithm can lead to a comparable performance. However, due to the lack of enough labeled data for training, the classification performance cannot be improved significantly. To solve this problem, the semi-supervised learning methods are introduced in this work as it can exploit the large amount of unlabeled data to improve the learning quality, and further to improve the classification performance. Since the unlabeled data are easily available, this work can save us a lot time and effort, and thus make the work more effective.This paper consists of two works. In the first work, we employed a supervised classification framework by designing 6 types of features, and using the support vector machine (SVM) classifier. By doing so, we got the fourth best performance on SemEval-2010 task 8 among 10 international teams. In the second work, we then applied a linear classifier based structure learning method on SemEval-2007 task 4 and SemEval-2010 task 8. We designed a couple of patterns to create the auxiliary problems based on the statistic and observation of the prepositions and verbs between the two nominals. The comparable results showed that the semi-supervised method can improve the performance significantly. Two papers corresponding to the two work have been published on ACL 2010 SemEval workshop and IALP 2010 conference.