Research Of The Case Retrieval Technique Based On Multi-Dimension Optimization

The main function of the intelligent data mining system is automatically doing Data Mining and finding the corresponding algorithm in accordance with the mandate of the users. Our group has been done research on Intelligent Data Mining System since 2006.The previous student designed of intelligent data mining system based on the case-based reasoning. By the application in the banking system proves the feasibility of the system, but also exposes the system's many shortcomings. There are three main aspects. First, the case expression is too simple to express the integrity features of the assignment. Second, there is no advisement in the technology-case retrieval algorithm optimization, resulting in poor outcomes in cases matching. Third, they didn't consider self-learning and adaptation of cases. The case elected is not the most suitable for the current situation, and we need adapt it to the current situation.The thesis does study on case-based reasoning technology and focus on the case retrieval technology optimization. This thesis does research on feature selection, feature weighting, case retrieval algorithm and the current optimization technology of case retrieval algorithm. It establishes a multi-dimensional optimization model. This model combines the genetic algorithms and analytic hierarchy process. Second, this paper expands the description of cases .There are five aspects including the description of the problem, data sources, the target type, use's opinion and system's requirement. Because in a certain period users often do the same missions. This thesis establishes user database, making use of the user's information as feedback information for the case retrieval. Thereby it reduces the time-consuming brought by the case reasoning.Finally, this paper increases a database based on revising case in the case database. It includes experience of adapting cases. Make use of these experiences to adapt case. Different algorithms produce different reactions for different data types, so adaptation is mainly directed against the parameter of algorithm. Revising cases also adopts the technique combining adaptation experience and modify function of algorithms. To retaining case, it takes the user's evaluation and the average utilization rate as a basis for excluding and update case.In experiments we have done two major tests. One is the multi-dimensional optimization algorithm. Select those representative models based on genetic algorithms in the single-objective and two-dimensional optimization of case retrieval technology contrasting to the model designed by this paper. From the search space and case-based reasoning results, prove the superiority of the multi-dimensional optimization. The other is on an evaluation of the holistic performance of the system. Mainly compare with the current good machine learning tool-Weka. In terms of the running time and veracity we are relatively satisfied with the results.