Organizational Co-evolutionary Algorithms For Web Log Mining

Relevant technology about web log mining has been studied in this paper. Based on organizational evolutionary algorithm two kinds of algorithms are proposed in this paper.The main research work and results are as follows:1. The OCEC (Organizational CoEvolutionary algorithm for Classification) is different from the GA based classification methods available. The individual in OCEC does not need to be coded, and the individual is evolved in organizations. So three evolutionary operators, add and subtract operator, exchange operator and unite operator are constructed in OCEC.The evolutionary operations of OCEC do not act on rules, but on the given data directly. Based on the characteristics of the OCEC, organizational co-evolutioary algorithm for web log mining is proposed in this paper. The unite operator is improved, and the parameters of add and subtract operator, exchange operator are modified to percent in this algorithm. It is shown that this alogrithm is effective and available, and its convergent speed is fast with computer simulations.2. Based on OCEA, a Multiple-Level Organizational Evolutionary Algorithm for Association Rules Mining(MLOEA) is proposed in this paper. A new evolutionary operator named as gather operator is constructed, a same attribute set form and two populations are defined in this algorithm. The algorithm regards each of data as an organization at the beginning, then join all organizations in population A. Through the evolution of population A, the organizations with same attribute set can be found out, then these organizations are moved to population B. The different organization with the uniform same attribute sets in population B will merge into a new greater organization by using the new gather operator on the organizations of population B. After evolving, association rules can be extracted from the same attribute sets of organizations in population B. This algorithm does not need to calculate the support values of many invalid large itemsets, and the population A and population B are evolved together, so its runing speed is faster. It is shown that the proposed algorithm has faster convergent speed, and achieves higher distilled rate of rules with computer simulations.