The Theory And Application Of Multi-Relational Data Mining

In traditional data mining algorithms, the knowledge discovery in database is mainly in the form of propositional logic: each sample is expressed in the form of tuples of attribute- value. It is difficult to target the expression of the complex relationship in the internal object and the pattern can only be found from a single model. However, most practical information in the relational database is stored in multiple relations, and it is difficult to express many complex patterns in propositional logical language. The learning of attribute - value in a single table can not directly exploit such links and the implication of the information they content. The algorithm is limited by a table or relationship constituted in database, and it can not simplely perform the relatively complex patterns in the form of attribute- value ,which can not find more complex hidden pattern.in the real world data. For many applications, when finding a multi-relational data pattern, the pattern will naturally involve a number of relations. When using traditional data mining algorithms, we should integrate the data from multiple relations into a single relationship before dataing. This requires not only a large number of pre-processing work and careful design, and may lead to information loss, semantic bias, reduce the efficiency and so on. Multi-relational data mining algorithm emerged a new research field in the context of this study.Based on inductive logic programming, MRDM focused on discovery the complex pattern in relational databases. MRDM can directly analyse the data in a number of relations without needing to convert to a single data table. Because of using first order predicate logical language, which is more expressive than propositional logical language, MRDM is able to express more complex patterns and facilitate the use of the background (domain) knowledge, the application of MRDM is more extensive and it has becomed a research hotspot.This paper relatively systematically and completely descriptes and analyses the general research and research methods of MRDM, and focuses on how to upgrade the traditional data mining algorithms to the field of MRDM.The main work is as follows:①Integrated the work of predecessors, according to the relevant literature,through the contrast with traditional data mining, focus on analysis the advantages of MRDM, and throw out a minnow to catch a whale, hoping to initiate more research attention. ②Research the inductive logic programming technology; highlight the strong ability of expression of first-order predicate,specifically to achieve in the international chess application.③Analyse multi-relational association rule, illustrate the algorithm WARMR, analyze its strengths and weaknesses.④Analyse how to extend traditional classification and clustering algorithm to the field of multi-relations.