Research On Multi-relational Association Rule Mining

Multi-relational data mining is developing rapidly in recent years, and it is one of the important areas of data mining. For multi-relational data mining, how mining more efficiently and how improving the scalability of the algorithm, has been the focus of our study. Compared to the traditional data mining algorithms, the complexity of specific performance of the algorithm in the multi-relational data mining put forward higher requirements. The search space of Multi-relational data mining algorithm becomes larger and more complex.At present, many relational data mining focused on two aspects of the research. One is finding the association rules among many tables based on the ILP theory of technology. It is using the logical expression of key sets of atoms, through Prolog queries to calculate item set's support, learn from a typical single-table association rules mining algorithm by sub-stack generation method. The problem of the approach towards multi-relational association rule mining will avoid statistical deviation. Another method is based on Un-ILP technology of multi-relational association rule mining algorithm. This approach is discovering and solving the performance issues in the algorithm.To solve the above problem, we do the following work:First of all, the theory of data mining and the association rules algorithm in data mining is summarized, and focus on the multi-relational association rules mining algorithm. It describes the popular technology of the multi-relational association rules mining, the classical algorithm of the multi-relational association rule mining based on the ILP techniques and the tuple ID propagation theory used to solve the problem of the joining among the relation tables based on the non-ILP technique.Second, this article introduces the FP-Growth into the multi-relational mining, proposes the multi-relational FP-Growth and gives the detail of how to build the multi-relational data mining FP tree and how to find frequent item sets in the tree, and gives a specific example to demonstrate the algorithm. Finally the algorithm based on the existence of non-ILP technical issues of statistical skew has been further optimization.Finally, given by PKDD conference data set, the algorithm is experimentally verified and compared with other algorithms. The experimental results are analyzed.