Association Rule Generation and Evaluation of Interestingness Measures for Artwork Tags

Finding associations between groups of items within a data set is referred to as the problem of association rule mining. This can be decomposed into the two subproblems frequent itemset mining and association rule generation. The open source data mining framework ELKI used throughout this work for experiments so far only supports the first subproblem. Therefore the implementation of association rule generation and its integration in ELKI is part of this work. The ARTigo project is an online platform with the purpose of powering a semantic artwork search engine. By playing special games on the projects website users contribute descriptive tags to artworks. These collected keywords can be analyzed using various data mining techniques. This work concentrates on applying association rules mining to the ATRigo dataset, using various interestingness measures. The main part of this thesis is to evaluate appropriate interestingness measures for the given ARTigo data. Therefore, ten interstingness measures are first searched for similarities. Second, their ability of eliminating uninteresting and locating interesting association rules is compared. Last their one thousand highest ranked rules are analyzed by the means of five proposed properties.