|
无主观相似信息的聚类
Clustering without any subjective similarity information |
|
| |
| 课程网址: |
http://videolectures.net/icml2010_ben_david_cwsi/
|
| 主讲教师: |
Shai Ben-David
|
| 开课单位: |
滑铁卢大学
|
| 开课时间: |
2010-07-20
|
| 课程语种: |
英语
|
| 中文简介: |
考虑基于这些页面之间的链接图来聚类大学网页的任务。是否可以从此链接结构中检测到“功能相似”页面的集群？注意，这是一个聚类任务，其中一个人在没有任何关于域元素之间的任何相似性或距离度量的先验知识的情况下开始。输入中的所有信息都是对象之间的客观，观察，二元关系。这些关系不是相似性链接。例如，教授页面群集具有非常内部链接，而服务页面群集具有许多内部链接。我们正在寻找的是集群，其成员与其他集群共享类似的链接模式。我们为这种聚类任务提出了一个正式的模型。我们的模型基于一个目标函数，它测量集群链接之间的同质性。我将讨论以最小的客观成本找到聚类的计算复杂性，并描述一些硬度结果以及有效的近似算法。谈话（部分）基于与Sharon Wulff的合作。
|
| 课程简介: |
Consider the task of clustering university web pages based on the graph of links between these pages. Can clusters of "functionally similar" pages be detected from just this link structure? Note that this is a clustering task in which one starts without any prior knowledge of any similarity or distance measure between the domain elements. All the information in the input comes as objective, observed, binary relations among the objects. These relations are not similarity links. For example, the cluster of professors pages have very internal links, whereas the cluster of service pages have lots of internal links. What we are looking for are clusters whose members share similar link patterns with respect to the other clusters. We propose a formal model for such clustering tasks. Our model is based on an objective function that measures the homogeneity of between-clusters links. I shall discuss the computational complexity of finding a clustering with minimal objective cost and describe some hardness results as well as efficient approximation algorithms. The talk is (partly) based on work with Sharon Wulff.
|
| 关 键 词: |
链接图; 大学网页; 功能相似
|
| 课程来源: |
视频讲座网
|
| 最后编审: |
2019-04-25：cwx
|
| 阅读次数: |
21
|
|
图 书 馆