Research Status of Association Rules

Since R. after Agrawal and others proposed the issue of mining association rules in 1993, many researchers have conducted a lot of research on this issue. So far, the main research directions include: multi-cycle mining algorithms (hierarchical mining algorithms) incremental update algorithm, distribution, parallel mining algorithm, multi-layer association rule mining algorithm, multi-value association rule mining algorithm, and concept grid-based association rule mining algorithm.

   Multi-cycle mining algorithmsThe core idea is "level-wise algorithms". As its name implies, an algorithm divides the entire mining process into several layers. After each layer is mined, the final result is merged. These algorithms include Apriori, AIS, AprioriTid and apriorihybrid proposed by Agrawal, DHP proposed by Park, partition proposed by savadere, sampling proposed by Toivonen, and FP-growth proposed by Toivonen; dic. Among them, the most effective and influential algorithms include the Apriori and FP-growth algorithms.

   Incremental update Mining AlgorithmThere are two situations: 1) update when the database records change (add or delete); D. w. cheng and others provide the update algorithm FUP corresponding to the hierarchical algorithm. Based on this, FUP2 is proposed, which can not only process the increase of transactions, but also delete or modify transactions. 2) updates when the metrics of association rules (such as support, confidence level, and interest level) change. Feng Yucai studied this situation and proposed the corresponding algorithms iua and piua. Feldman proposed an association rule update technology called the border algorithm. The algorithm only needs to check that all real subsets are frequent project sets when the minimum supported degree specified by the user is absolute and constant, but it is not a frequent project set (these project sets are called border ). However, this algorithm still needs to store related frequent project set results to reduce the update cost of association rules.

   Parallel/Distributed Association Rule Mining AlgorithmThe data to be processed in Data Mining is usually very huge, and the data is distributed across regions. Currently, the distributed algorithms for mining association rules proposed by most documents are based on distributed processors (each processor excludes its own memory and disk space, and the processors communicate with each other through internal connection mechanisms such as networks). Algorithms include CD, PDM, FPM, DD, IDD, and hPa. These algorithms can be seen as the parallel version of the Apriori algorithm.

   Multi-layer Association Rule Mining AlgorithmIt defines the minimum support threshold value for each abstract layer of the concept layer and uses multiple policies to mine multi-layer association rules, different from the previous method based on the support-credibility framework. At present, many algorithms have been proposed to mine multi-layer association rules. ml_t2l1 and its variants ml_t1la, ml_tml1, ml_t2la, and r proposed by Han and others. such as Cumulate, stratify and its variants etimate and estmerge proposed by srikant.

   Multi-value Association Rule Mining AlgorithmIs different from boolean association rules. Currently, most of the proposed multi-value attribute association rule mining algorithms convert the problem of multi-value attribute association rule mining into a boolean association rule mining problem, which divides the values of multi-value attributes into multiple intervals, each interval acts as an attribute, and each category of a category attribute is considered as an attribute. G. michael and others proposed a multi-value attribute association rule in the form of X = QX þ y = QY. Both the preceding and subsequent items correspond to a single value rather than a single interval; however, when you need to mine association rules between all attributes, you will be faced with the combination explosion of rules.

   Association Rule Mining Algorithm Based on Concept LatticeIt is the most widely used and fruitful field of concept lattice in Data Mining. scholars at home and abroad have conducted in-depth research on association rule mining based on concept lattice. Godin and others proposed a conceptual lattice model to extract the implication rules, but the implication rules are deterministic rules. This method does not have the ability to describe the approximation rules; R. missaoui and other algorithms are proposed to extract approximate rules from the concept lattice. Based on the godin incremental construction of the concept lattice algorithm, Hu keyun and others, a more effective association rule algorithm for basket analysis is proposed, visualization of association rule mining is achieved. Petko valtchev and other algorithms are proposed to mine frequently closed project sets using concept lattice; wang Dexing and others proposed an algorithm to quickly discover frequently closed project sets using the concept lattice of pruning.