Weakly Supervised Learning Algorithm to Eliminate Irrelevant Association Rules in Large Knowledge Bases

Abstract

The construction and population of large knowledge bases have been widely explored in the past few years. Many techniques were developed in order to accomplish this purpose. Association rule mining algorithms can also be used to help populate these knowledge bases. Nevertheless, analyzing the amount of association rules generated can be a challenge and time-consuming task. The technique described in this article aims to eliminate irrelevant association rules in order to facilitate the rules evaluation process. To achieve that, this article presents a weakly supervised learning technique to prune irrelevant association rules. The proposed method uses irrelevant rules already discovered in past iterations and prunes off those with the same pattern. Experiments showed that the new technique can reduce and eliminate the amount of rules by about 60%, decreasing the effort required to evaluate them.