An Algorithm for Tolerance Value Generator in Tolerance Rough Sets Model

The Tolerance Rough Sets Model (TRSM) is a tool to model document in text mining. Generation of a document representation based on TRSM basically depends on tolerance classes of terms which are created by setting up a tolerance value. Despite the fact that tolerance value is critical in TRSM, the manual process of setting this value is an exhaustive task. We conducted a study on our own corpus and were supported by two human experts when constructing the training data. We came up with a novel algorithm to generate tolerance value automatically from a set of training data. The heart of our algorithm is measuring the distance between document representation calculated using TFIDF weighting scheme and document representation yielded by TRSM both in reduced dimensional space generated by Singular Value Decomposition (SVD). In spite of the result, we recognize that further study is significant, i.e. to set up some properties (the size of training data and rank of SVD) as well as evaluating the algorithm in real data and scenario.