中文摘要：

多值属性的约简大多采用将多值属性单值化的方法,从而增加了概念格节点数量,带来数据处理的复杂性。针对此问题提出了基于粒计算的多值属性概念格约简方法。应用相容函数和信息粒定义了概念粒和相容概念粒集,根据概念粒的偏序关系给出了概念粒集的分层算法,同时给出了概念粒的合成和分解算法;通过概念粒的分量IG得到概念粒的属性域,依据核心属性得到各个属性的概念粒分辨率,为属性约简提供排序依据;应用概念格的分层建格算法构建概念粒集的格结构,通过格同构性得到形式背景的核心属性集合。此方法避免了复杂的模式匹配和多次遍历数据库的操作,为概念格在信息检索、知识发现和数据挖掘等领域的应用奠定了技术基础。

英文摘要：

Multi-valued attribute uniformization method is frequently employed in the reduction of multi-value attributes,for which the number of concept lattice nodes increases and the data processing becomes complex. To solve this problem,the paper put forward a multi-valued attribute reduction method,basing on granular computing. After presenting definitions of the concept granule and the compatible concept granular sets by applying compatible function and information granule,the paper provided hierarchical algorithm grounding on the partial-order relation of concept granule,and the merging and decomposing algorithm of concept granule. Ultimately the attribute domain of concept granule was obtained through the component IG. On the basis of the core attributes,the paper obtained concept granule resolution for each attribute,which provided a basis for ordering in attribute reduction. At last,lattice structure of concept granular set was built by the application of hierarchical lattice-building algorithm. In this way, the core attribute set in the formal context could be obtained through lattice isomorphism. The method avoids complex pattern matching and traversing the database operation for many times,and lays the technical foundation for the application of concept lattice in the fields of information retrieval,knowledge discovery,data mining etc.