Effectiveness Research of the Apriori Algorithm Implementations as Part of the Recommendation System

Abstract

The article presents a comparison of the author's implementation of the Apriori algorithm with implementation presented in the Accord.NET library. A comparison of various parallel implementations of this algorithm is also shown. The main objective of the article is evaluating the efficiency of the Apriori affinity analysis algorithm implementation by the speed of creating an association rules list. The results of testing on the data provided by the company “FB Consult” are given. The test results show an increase in performance of the optimized parallelized implementation by about 5.28 times compared to the implementation from Accord.NET. This article clearly shows that optimization of the algorithm in order to reduce the running time by using binary vectors and operations on them gives more significant speedup than parallelization. Whereas when parallelizing, attention should be paid to the computation of support for association rules, because based on the results of this function, most of the association rules will be eliminated.