Social Semantics And Its Evaluation By Means of Closed Topic Models: An SVM-Classification Approach Using Semantic Feature Replacement By Topic Generalization
Ulli Waltinger, Alexander Mehler, Rüdiger Gleim · 2009
Abstract. Text categorization is a fundamental part in many NLP ap-plications. In general, the Vector Space Model, the Latent Semantic Analysis and Support Vector Machine implementation have been suc-cessfully applied within this area. However, feature extraction is the most challenging task when conducting categorization experiments. Moreover, sensitive feature reduction is needed in order to reduce time and space complexity especially when deal with singular value decomposition or larger sized text collections. In this paper we examine the task of feature reduction by means of closed topic models. We propose a feature re-placement technique conducting a topic generalization comprising user generated concepts of a social ontology. Derived feature concepts are then subsequently used to enhance and replace existing features gaining a minimum representation of twenty social concepts. We examine the effect of each step in the classification process using a large corpus of 29,086 texts comprising 30 different categories. In addition, we offer an easy-to-use web interface as part of the eHumanities Desktop in order to test the proposed classifiers. 1