Granular Ranking Algorithm Based On Rough Sets

The increasingly data resources have led to the emergence of Data Mining technique. It's a technique that aims at large-scale source data analysis and revealing the intrinsic knowledge of the data. At present time, it is one of the hottest research topics, in the field of artificial intelligence and database, and has been widely applied in Targeted Marketing. For meeting various business requirements in the targeted marketing, Response Models can be built by different data mining techniques, such as regression, decision tree, neural network, association rule, etc. The potential values (i.e. Response Values) of customers or products, calculated by the response models, are instructive information for decision making in the targeted marketing. Consequently, the efficiency of marketing activities can be improved, while the marketing cost can be cut down.The idea of Granular Computing was emerged in 1970's, soon put into development and wide application. Its main idea is to imitate the thinking model of human beings. That means people can observe and analyze a same problem on completely different granularities, and can easily transfer from one granularity world to another.In recent years, granular computing has been applied in data mining field. With initial-stage accomplishment have been achieved, it has become a new and prospective research topic in the academic study of data mining. In this dissertation, granular computing is introduced into response modeling. And related research is done including:1) Designed a general algorithm framework for Market Value Function (MVF), which is a linear response model proposed by Prof. Y.Y. Yao. Thus, different kinds of utility functions and weight functions can be combined discretionarily, calculating the market values. Then, proposed the Market Value Function 1-1 Algorithm (MVF11A) byusing the framework and selecting u_a~1 and w_a~1 as utility function and weight function respectively.2) Introduced the idea of Rough Sets in granular computing into response modeling, and put forth the Granular Ranking Algorithm (GRA) with the time complexity of O(nm). The core of the algorithm is the construction of Granular Ranking Function (GRF), a response model based on rough sets, which guides testing sample sets finish ranking. The new algorithm is validated by a contrastive test with MVF on a standard dataset provided by KDD Cup. Compared with traditional data mining techniques, the new algorithm preserves many advantages of response modeling, and the ranked result has a good readability. Compared with the linear market value function model, the computation result of new algorithm approaches to that of MVF but with less time consumption. Thus, the new algorithm can be applied for the investigation of targeted marketing, identifying potential values of customers or products.3) Put forth the Granularity Set Combination Algorithm (GSCA) with the time complexity of O(nm) by extending the granular ranking algorithm, obtaining the Incremental Granular Ranking Algorithm (IGRA). The algorithm is validated by an incremental test. In the production environment, the incremental algorithm can update the response model partially to solve the problem caused by business data increase. The incremental algorithm is also a good solution to the problem of one-time granular ranking in large-scale databases.