Niche Finding - A New Task of Knowledge Discovery

利基发现——知识发现的新任务

• 批准号: 9988084
• 负责人: Raul Valdes-Perez
• 金额: $ 29.64万
• 依托单位: Carnegie-Mellon University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2000
• 资助国家: 美国
• 起止时间: 2000-04-15 至 2004-03-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9988084&HistoricalAwards=false
• 关键词: Niche; Finding; New; Task; Knowledge

The starting point for many tasks of data-driven knowledge discovery is a data matrix whose rows are individuals, columns are numeric and/or symbolic features, and entries are feature values. The usual goals of discovery or data analysis include finding classifiers, clusters, association rules, and Bayesian networks. The new question posed by the PI in this work about a data matrix is: Given one individual (or row) that is of special interest, what is unique about that individual? That is, concisely articulate that individual's niche within the overall context. One challenge the PI intends to address here is to better understand what statements of individual uniqueness people tend to find understandable and interesting. Preliminary results indicate that the criteria for these two properties are mostly general, and only partly domain-dependent. A prototype implementation of niche finding, using a set covering approach, has been applied to a newly assembled dataset on 236 major U.S. universities described by 123 features. The PI will extend the methods by adding new feature types, scaling up to larger datasets, comparing against alternative approaches and different approximation algorithms, and by applying niche finding to significant problems in science and other professional fields. The overall goal is to make niche finding as usual a discovery task as the others cited. The PI expects to show that there is a new, simple, and interesting question to ask about datasets that has broad applicability in science and other professional fields. He also intends to enhance public appreciation for knowledge discovery and information processing by developing special versions for baseball, with plans for use on public media and as a museum exhibit

数据驱动的知识发现的许多任务的起点是数据矩阵，其行是个体，列是数字和/或符号特征，并且条目是特征值。 发现或数据分析的通常目标包括发现分类器、聚类、关联规则和贝叶斯网络。 PI在这项关于数据矩阵的工作中提出的新问题是： 给定一个特别感兴趣的个体（或行），该个体的独特之处是什么？ 也就是说，简明扼要地阐明个人在整体背景下的利基。 PI打算在这里解决的一个挑战是更好地理解人们倾向于认为可以理解和有趣的个人独特性陈述。 初步结果表明，这两个属性的标准大多是一般的，只有部分域相关。 一个原型实现的利基发现，使用一套覆盖的方法，已被应用到一个新组装的数据集上的236所美国主要大学描述的123个功能。 PI将通过添加新的特征类型，扩展到更大的数据集，与替代方法和不同的近似算法进行比较，以及将利基发现应用于科学和其他专业领域的重大问题来扩展这些方法。 总体目标是使利基发现像往常一样，发现任务，因为其他人引用。 PI希望表明，有一个新的，简单的，有趣的问题要问的数据集，在科学和其他专业领域具有广泛的适用性。 他还打算通过开发棒球的特殊版本来提高公众对知识发现和信息处理的赞赏，并计划在公共媒体上使用和作为博物馆展览