Using Soft Similarity in Multi-label Classification for Reuters-21578 Corpus

Abstract

In classification tasks one of the main problems is to choose which features provide best results, i.e., Construct a vector space model. In this paper, we show how to complement traditional vector space model with the concept of soft similarity. We use the combination of the traditional tf-idf model with latent Dirichlet allocation applied in multi-label classification. We considered multi-label files of the Reuters-21578 corpus as study case. The methodology is evaluated using the multi-label algorithm Rakell. We used the traditional tf-idf model as the baseline. We present the F1 measures for both models for various feature sets, preprocessing techniques and vector sizes. The new model obtains better results than the base line model.