Publication Type

Conference Proceeding Article

Version

publishedVersion

Publication Date

7-2008

Abstract

With an explosive growth of blogs, information seeking in blogosphere becomes more and more challenging. One example task is to find the most relevant topical blogs against a given query or an existing blog. Such a task requires concise representation of blogs for effective and efficient searching and matching. In this paper, we investigate a new problem of profiling a blog by choosing a set of m most representative entries from the blog, where m is a predefined number that is application-dependent. With the set of selected representative entries, applications on blogs avoid handling hundreds or even thousands of entries (or posts) associated with each blog, which are updated frequently and often noisy in nature. To guide the process of selecting the most representative entries, we propose three principles, i.e., anomaly, representativeness, and diversity. Based on these principles, a greedy yet very efficient entry selection algorithm is proposed. To evaluate the entry selection algorithms, an extrinsic evaluation methodology from document summarization research is adapted. Specifically, we evaluate the proposed entry selection algorithms by examining their blog classification accuracies. By evaluating on a number of different classification methods, our empirical results showed that comparable classification accuracy could be achieved by using fewer than 20 representative entries for each blog compared to that of engaging all entries.

Keywords

Blog profiling, Entry selection, Blog classification

Discipline

Computer Sciences | Social Media

Publication

AND '08: Proceedings of the Second Workshop on Analytics for Noisy Unstructured Text Data: July 2008, Singapore

First Page

55

Last Page

62

ISBN

9781605581965

Identifier

10.1145/1390749.1390759

Publisher

ACM

City or Country

New York

Additional URL

http://dx.doi.org/10.1145/1390749.1390759

Download

Share

COinS