A Study on Optimal Parameter Tuning for Rocchio Text Classifier

Current trend in operational text categorization is the designing of fast classification tools. Several studies on improving accuracy of fast but less accurate classifiers have been recently carried out. In particular, enhanced versions of the Rocchio text classifier, characterized by high performance, have been proposed. However, even in these extended formulations the problem of tuning its parameters is still neglected. In this paper, a study on parameters of the Rocchio text classifier has been carried out to achieve its maximal accuracy. The result is a model for the automatic selection of parameters. Its main feature is to bind the searching space so that optimal parameters can be selected quickly. The space has been bound by giving a feature selection interpretation of the Rocchio parameters. The benefit of the approach has been assessed via extensive cross evaluation over three corpora in two languages. Comparative analysis shows that the performances achieved are relatively close to the best TC models (e.g. Support Vector Machines).