Extraction of Strong Associations in Classes of Similarities

Several algorithms are proposed to support the process of automated classification of textual documents. Each of these algorithms has characteristics that influence the classification result. Depending on the amount and nature of the data submitted, the quality of results may vary considerably from one algorithm to another. The generated classes are often noisy. In addition, the number of classes created can be significant. These constraints can easily become barriers to data analysis. This paper presents a method that exploits the notion of association rules to extract regularities in the classes of similarities produced by classifiers.