Knowledge Discovery Based On Rough Sets And Applications To Chinese Processing

Rough Sets theory is a new mathematical approach to imprecise, vague and uncertain information analysis. It has been applied in many fields such as machine learning, knowledge acquisition, decision analysis, expert system and pattern recognition. Knowledge discovery based on Rough Sets theory is the process of finding new, applicable and no-trivial patterns by using Rough Sets theory, which has been applied to medicine, finance, engineering, language processing and so on.Rough Sets theory emphasize the research on the knowledge reduction for information system, especially for decision table, to obtain concise representations of information, or useful rules for predicting decision values of unknown objects. In this thesis we research on knowledge reduction and its algorithms as well as the application to natural language process, and achieve the following important results.A new conditional entropy and knowledge reduction algorithms are proposed. The knowledge reduction algorithms based on existing conditional entropy result in high time complexity and core and reducts obtained may not be equal to the core and reduct in the algebra view. In order to compute the core and the reduct in the algebra view by means of the information view, the defect of the existing conditional entropy is analyzed and the new conditional entropy is proposed. This new conditional entropy can represent the core and the reduct in the algebra view equally. By using the new conditional entropy, the algorithm for the core and the reduct is proposed. Because the new conditional entropy can represent the core and the reduct in the algebra view equally, the results of the algorithm proposed in this paper is equal to those in the algebra view. Theoretical analysis and experimental results show that this algorithm is efficient and superior to algorithms based on the positive region and the existing conditional entropy in finding optimal or sub-optimal reducts.The conversion algorithm for decision tables is proposed. Some efficient algorithms for core and reducts can only be applied to consistent decision tables, and is not suitable for inconsistent decision tables. To overcome this shortcoming, an algorithm is proposed to convert the computation for the original inconsistent decision...