Improving VG-RAM WNN Multi-label Text Categorization via Label Correlation

In multi-label text databases one or more labels, or categories, can be assigned to a single document. In many such databases there can be correlation on the assignment of subsets of the set of categories. This can be exploited to improve machine learning techniques devoted to multi-label text categorization. In this paper, we examine a virtual generalizing random access memory weightless neural network (VG-RAM WNN for short) architecture that takes advantage of the correlation between categories to improve text-categorization performance. We compare the performance of this architecture, that we named data correlated VG-RAM WNN (VG-RAM WNN-COR), with that of standard VG-RAM WNN using four multi-label categorization performance metrics: one-error, ranking loss, average precision and hamming loss. Our experimental results show that VG-RAM WNN-COR has an overall better performance than VG-RAM WNN for the set of metrics considered.