New Event Detection in the Context of Web Forums
Date Issued
2007
Date
2007
Author(s)
Chuang, Ting-Wei
DOI
zh-TW
Abstract
As the emergence of Web forums, tremendous archived messages regarding diverse subjects become valuable knowledge repositories. Unfortunately, the continuously increasing and heterogeneous messages compound the navigation load. The existing human-generated category hierarchies provided by most Web forums are not self-revealing. Besides, the thread structures created according to the subject headers of messages only give limited information about actual contents of messages. Thus, the forum messages are actually lacked for a better organization to guide browsers in the information space.
In this work, we employ the conventional incremental clustering algorithm which is well-studied in TDT as the baseline to group together topically related messages in streams of text messages. In order to detect the topic from one single short message, we propose six different models to exploit temporal information and context information obtained from parent messages. We utilize some extern features extracted from Web as well. Our preliminary experiments confirmed our assumption that incorporate features relevant to the Web forum environment can enhance the performance the basic clustering algorithms.
Subjects
網路論壇
事件偵測與追蹤
新事件偵測
分群技術
網路社群
Web Forum
Topic Detection and Tracking
New Event Detection
Clustering
Web Community
Type
other
File(s)![Thumbnail Image]()
Loading...
Name
ntu-96-R94725023-1.pdf
Size
23.31 KB
Format
Adobe PDF
Checksum
(MD5):6475f43630408c6fee0bcd3e056f99d5