A Term Weighting Approach with Subjective Logic Reasoning for Text Categorization

A new term weighting approach is used to construct the simplest linear weighting classifier (SL). By probability standard deviation of terms as a base line weighting regulated with terms distributed parameters based on subjective logic reasoning, the weighting is computed. In the assessment process of terms distributed parameters, the model of the term reputation in documents categories based on Beta probability density function has been discussed. Some problems, such as parameters overfitting, documents length, documents number of categories, term selection etc., also have been considered. Experiments with 20Newsgroup, Fudan-Chinese, Reuters-21578 reported in the paper shows that SL generally outperformed traditional classifiers.