Pushing task relevant web links down to the desktop

Searching the web has become a task in many people's work, without which subsequent tasks would be hard to carry out or even impossible. But as people tend to have less time for querying the web or even for searching their personal computer for information they need, it becomes common to skip information gathering activities like trying to find useful resources on the web because of the "effort" it takes to query a web search engine. In this paper we propose to use software agents that collect useful web specific related information which would otherwise not be viewed at all. More specifically, we present two new algorithms to automatically search the web and recommend URLs relevant to user's current work, defined through his or her active personal desktop documents. Our experiments show our proposed algorithms, Sentence Selection and Lexical Compounds, to yield significant improvement over simple Term Frequency based web query generation, which we used as a baseline.