The Study Of Electronic Commerce Transaction Knowledge Acquisition Based On Rough Set

The rapid development of E-Commerce has brought the opportunity for the enterprise, but with the deepening of EC applications, large amounts of transactions data generate;how to extract the potential knowledge to support business decision making has become the first task for the operator and the manager currently. The emergence of data mining technology provides strong technical support to solve this problem. Compared with other DM technologies, Rough Sets has the unique advantage taking the data provided as the only without any prior knowledge, so the research of RS has become the academic upsurge in the domain of Knowledge Acquisition in recent years.This dissertation intensively studies the application of RS in the Knowledge Acquisition from EC transactions data based on the predecessors'achievements and experience.Firstly, the data preparation is discussed, including EC transactions data induction and categorization, the general process of data precondition and the Web data precondition, especially the way to supplement and disperse the decision table based on RS.Then, this dissertation makes researches on the classification and association knowledge acquisition model based on RS deeply. In the process of building the classifying model, the selection before method of the attribute reduction, starting with the core set is explored. To the disadvantage that the current algorithms can not compute the core from some inconsistent information systems, an improved algorithm based on discernibility matrix is introduced, using the concept of partial condition entropy. To the problem that the computation load of the general heuristic algorithms of attributes reduction on the whole decision table is large, an improved algorithm based on information entropy is presented. When evaluating the rules, this thesis proposes a both-sides comprehensive measurement, combining the objectively deterministic factor (rule confidence) and the introduced subjective relative weight concept. In the process of constructing the association knowledge acquisition model, the equivalence class of RS is applied to classify the affairs database according to whether the transaction item appears or not, so that it just needs to do some set operations to these transaction items' equivalence classes instead of scanning the affairs database, which can reduce the time cost. Mean while, in regarding of the commercial application, the minimum support is redefined with the commodity profit restraint, which can avoid the commodity with low appearance frequency but high profit from being filtered, so that the final rules have greater commercial value.Finally, the EC transaction data of Gazelle.com provided by the KDD Cup 2000 is used to validate and analyze the feasibility of the two knowledge acquiring models.