The Research Of Classification Based On Multi-Relation Domain Knowledge

With the explosive growth of the size of databases, knowledge discovery in databases (KDD) are facing new problems such as focusing search to relevant portion of data, making the discovered patterns more meaningful and so on. Additional knowledge, called domain knowledge (DK), is used to help the steps of the discovery process such as data pre-processing, finding the strong relevant attributes, generalizing the concepts to more interesting levels, guiding the discovery process to extract more useful rules, interpreting the discovered results and making the results be more understandable to the end user and so on, so as to make the discovery process more efficiently and effectively. However, the research and application referring to this research field are at a primary stage, thus deserve our exploring and further research. KDD based on DK is a newly promising research field.Concept lattice, also called GCL (Galois Concept Lattice) is a complete form of knowledge representation. In the dissertation, incorporating domain knowledge based on concept lattice is proposed, which is intend to find the inner relationship among attribute values. The content of the dissertation is as follows:1. The role of domain knowledge in the process of KDD is discussed in detail. Based on different representation of domain knowledge, the effect and the corresponding mechanism of using domain knowledge are different.2. The essential purpose of KDD is to find the potential relationship among data that described in multi concept levels. However, the data stored in the real world databases correspond to a certain specialized level, which can't reveal the common information among these data. This dissertation incorporates multi-relation domain knowledge that can be used to find the inner relationship of attributes into knowledge discovery and utilizes concept lattice for describing the closest relationship among attribute values.3. A classification algorithm is also presented, which demonstrates the superiority of using multi-relation domain knowledge represented by concept lattice.4. Based on the work stated above, a prototype system that can utilize multi-relation domain knowledge in knowledge discovery in database is implemented.