Edit Operations on Lattices for MDL-based Pattern Summarization
Edit Operations on Lattices for MDL-based Pattern Summarization
复制标题
DOI：
--
复制
发表时间：
2015
期刊：
影响因子：
--
通讯作者：
Keisuke Otaki;Akihiro Yamamoto
中科院分区：
文献类型：
--
作者：
Keisuke Otaki;Akihiro Yamamoto
封闭性是FCA中的一个基本概念，对于解决模式挖掘中的模式冗余问题也很重要，其中一些枚举模式从属于其他模式，并且它们是冗余的。虽然已经提出了许多有效的挖掘算法，但发现整个枚举模式的特征和易于解释的子集，称为模式摘要问题，仍然是具有挑战性的。一个常用的基于信息论的标准;最小描述长度（MDL）原则有助于我们解决这个问题，但它要求我们根据模式的类型从头开始设计MDL评估。在本文中，我们提出了一个新的框架，适用于各种模式使用格，这是有益的制定MDL评估。一个关键的想法是修改现有的模型，通过使用编辑操作定义的概念之间的格子，这使我们能够考虑额外的信息，如背景知识，并帮助我们设计的MDL评估。我们实验我们的方法，看看我们的建议，帮助我们从整个集合中获得信息的结果，并确认我们的模型是适用于各种模式。
The closedness, a fundamental conception in FCA, is also important to address the pattern redundancy problem in pattern mining, where some enumerated patterns subsume others and they are redundant. Although many efficient mining algorithms have been proposed, finding characteristic and easy to interpret subsets of the whole enumerated patterns, called the pattern summarization problem, is still challenging. A well-used Information Theory-based criterion; the Minimum Description Length (MDL) principle helps us to tackle the problem, but it requires us to design the MDL evaluation from scratch according to types of patterns. In this paper we propose a new framework applicable to various patterns using lattices, which are beneficial to formulate the MDL evaluation. A key idea is revising an existing model by using edit operations defined on lattices among concepts on them, which enables us to consider additional information such as background knowledge and helps us to design the MDL evaluation. We experiment our method to see that our proposal helps us to obtain informative results from the whole sets, and confirm that our model is applicable to various patterns.