【摘要】 在分析了 COP- COBWEB算法之后 ,给出了一种结合背景知识计算分类效用的方法 .它考察数据对象之间的已知关联 ,通过比较预期关联数与实际关联数 ,来获得限制系数 .限制系数将直接参与分类效用的计算 ,得到基于限制的分类效用 CCU .有关实验证明 :利用基于限制的分类效用 ,COBWEB算法将更为有效 ;在有噪声的情况下 ,基于CCU的 COBWEB算法明显优于 COP- COBWEB算法更多还原

【Abstract】 After analyzing COP-COBWEB algorithm, an approach for combining background knowledge into category utility was proposed. It takes into consideration of the instance level constraints: Must-links and Cannot-links. For a given class, its Constraints Index (CI) will be computed by comparing the number of constraints in it and the number of constraints expected. Then CI will be combined with Category Utility (CU), and gets Constraints Based Category Utility (CCU). Using CCU to guide the construction of the classification tree, COBWEB is improved in accuracy. The experiments also suggest that CCU directed COBWEB is superior to COP-COBWEB when clustering noisy data.更多还原