Maintenance of Generalized Association Rules Under Transaction Update and Taxonomy Evolution

Swarm Intelligence Based Technique for Rule Mining in the Medical Domain

“…Most approaches to rule mining have been based on candidate generation using an Apriori [9] style algorithm or FP-tree [10] style approaches to mine rules without candidate generation. Efforts have been made to improve the performance of these techniques by either i) reducing the number of passes over the database [11] [12], or ii) sampling data [13] [14] [15], or iii) adding extra constraints on the structure of rules [16] [17] or iv) parallelization of operations [18] [19] [20] or v) a combination of these. But these different strategies still do not return accurate results in a reasonable time.…”

Recommender systems over distributed architectures

“…The majority of GARM algorithms, GP-Close (Jiang & Tan, 2006;Jiang et al, 2007) included, is not incremental, since they are unable to deal with the potential need to update the taxonomy information, as more transactions arrive (Kotsiantis & Kanellopoulos, 2006) 43 ; nevertheless, the potential insertion of new items in the taxonomy, the deletion of other ones, the possible renaming of certain elements, as well as the potential reclassification of certain taxonomy items may render infrequent elements in the database constructed so far frequent or vice versa (Tseng et al, 2005). To tackle these issues, the IDTE algorithm, which caters for the incremental update of the generalized association rules, as the taxonomy evolves, by either inserting new items, deleting other ones, renaming certain elements or reclassifying certain items, has been proposed in (Tseng et al, 2005). IDTE tackles the aforementioned four challenges by taking into account that the sets of elements can be divided into two categories, the ones, which are affected by the taxonomy modification and those which are not; the update procedure needs to be performed only for the former during the examination of the candidate frequent itemsets.…”

“…at fixed time periods or when a number of new entries has been accumulated)], the entire structures utilized have to be reconstructed from scratch. It would be therefore better to resort to an incremental approach (Ezeife & Su, 2002;Tseng et al, 2005;Li et al, 2009), eliminating the need to rescan the database, each time new information arrives, in a fashion that frequent patterns generated at time point t i can be processed together with web log entries recorded in the period [t i , ti +1 ] to mine frequent patterns for time point t i+1 , removing thus the need to store and process web log entries recorded in the period [0, t i ]. In this section, we present iFP-Growth (Giannikopoulos et al, 2012), a trie-based approach (Tryfonopoulos et al, 2004;Tryfonopoulos et al, 2009;Aoe et al, 1992;Baeza-Yates & Gonnet, 1996;Rivest, 1976;Nilsson & Karlsson, 1999) for storing the frequent-pattern tree.…”