Based On Frequent Itemsets Markov Network Building And Its System Design And Implementation

With the development of computing, communications and mass data, The data has been greatly enriched. However, the phenomenon of a paucity of knowledge spread. It is difficult to understand the information contained in the data and obtain valuable information. To solve the above problems, The knowledge discovery technology which mine the potential pattern in the data and find out valuable information and knowledge appeared. Knowledge discovery system is a specific application of knowledge discovery techniques in practice, It’s architecture can be divided into the data source layer, pending mining layer, mining layer, knowledge evaluation and presentation layer, user interface and control layer. Mining layer is the core of the system, it is the specific areas of applications of knowledge discovery methods, techniques, algorithms. The uncertainty of knowledge representation based on probability theory and association rule mining based on frequents are important contents in knowledge discovery research areas. Probabilistic graphical model, including the model of the directed graph and undirected graph model, is an effective tool for uncertainty knowledge inference, Bayesian network and Markov network are typical representatives of it. Apriori algorithm based on frequent itemsets can mining association rules between items efficiently. However, neither the Bayesian network nor Markov network took advantage of the results of association rule mining. Although association rule mining can mine the frequent itemsets fully, it can not reflect the association relation between different frequent intemsets.Based on above background, this paper first proposed a method of building a Markov network model based on association rule mining, namely building a directed graph model-Markov network with analysing and testing of conditional independence based on the frequent itemsets mined by the association rule. Then we designed and implemented the Knowledge Discovery system with this method as mining layer. The system includes data preprocessing, the methods of design mining layer, results evaluation and presentation, graphical user interface, and so on. The main contributions and novelties of this thesis can be summarized as follows:To find frequent itemsets according to the Apriori algorithm in association rule mining and then construct a Markov network model in line with coditional independence tests based on frequent itemsets. So as both to take full advantage of the results of association rule mining to construct a probabilistic graphical model, and to mine association between different frequent itemsets. The thesis also design and implement the knowledge discovery system of setting the method proposed in this paper as mining layer. In the system, Markov network constructed in this way will be compared and analysed with the Markov network built based on dependency analysis results to objectively assess and demonstrate the mining results of this method.