Overview - Web 2000

Proceedings | Data | Results | Runs | Participants

TREC-9 marked a broadening of the range of of search task types represented in the Web track and a serious attempt to determine whether hyperlinks could be used to improve retrieval effectiveness on a topic-relevance ad hoc retrieval task. The Large Web Task compared the ability of systems to locate online service pages within the 18.5 million page VLC2 collection. In this case the question is not whether the page is relevant to the topic, but whether it provides direct access to the desired service. In contrast, the Main Web Task compared link-based and non-link methods on a task involving topic relevance queries and a 1.69 million page corpus (WT10g) which was carefully engineered to ensure a high density of inter-server links and (relative) ease of processing. The Main Web task topics were in TREC Ad Hoc form but were reverse engineered from query logs. Ternary relevance judgments were obtained and, in addition, assessors were asked to identify 'best' documents for each topic. As in TREC-8, no significant benefit associated with the use of link information in a topic-relevance retrieval task was demonstrated by any of the participating groups, whether or not additional weight was given to highly relevant documents.

Track coordinator(s):

• D. Hawking, CSIRO Mathematical and Information Sciences

Tasks:

• main: Main Task
• large: Large Web Task

Track Web Page: https://trec.nist.gov/data/t9.web.html

---