School of Computer Science and Engineering, Seoul National University

---

Abstract—Although documents have hundreds of thousands of unique words, only a small number of words are significantly useful for text analysis. Thus, feature selection has become an important issue to be addressed in various text analysis studies. A number of techniques and algorithms for feature selection are available, but unfortunately, it is hard to say that a certain algorithm overcomes the others, because feature selection results mostly depend on the source documents. We should pick and choose the appropriate algorithm and the best subset of feature words whenever we need to analyze source documents. In this paper, we present a framework named ‘PicAChoo’, which stands for ‘Pick And Choose’ that enables customizable feature selection environments by composing several primitive feature selection methods without hard-coding. As indicated in the name, this framework provides many strategies for extracting appropriate features and allows dynamic compositions among several feature selection methods. In addition, it tries to give users an environment that utilizes linguistic characteristics of textual data, namely part-of-speech, sentence structures, and so on. Finally, we illustrate that selected feature words can be used for various intelligent services.

Index Terms—text analysis, feature selection, dynamic composition, feature storing model, complex feature.

[PDF]

Cite: Jaeseok Myung, Jung-Yeon Yang, Sang-goo Lee, "PicAChoo: A Text Analysis Tool for Customizable Feature Selection with Dynamic Composition of Primitive Methods," Journal of Software vol. 5, no. 2, pp. 179-186, 2010.