中文摘要：

并不是所有的先验知识都能提高聚类质量，因此，评估先验知识的质量对半监督聚类极其重要。在现有的可有效评估成对约束形式的实例层知识的指标——信息量和一致性基础上，提出了可有效评估属性排序形式的属性层知识的指标，即信息量和有效性。并证实了提出指标的有效性和潜力。

英文摘要：

Not all the prior knowledge can improve clustering quality, so evaluating the merit of prior knowledge is extremely important for semi-supervised clustering. In recent years, some researchers haw~ proposed an effective measure to evaluate instance level knowledge in the form of pair-wise constraints, in formativeness and coherence. In this paper, we have further extended the work into the evaluation of altribute-level knowledge in the form of attribute order preferences. The proposed measure includes two as pects, informativeness and effectiveness, provided that instance-level knowledge in the form of pair wise constraints are also available. Experiments prove the effectiveness of the proposed method.