The physical context for the information filtering process is the existing networked computing infrastructure. The relevant portion of the physical context may consist of, for example, isolated workstations monitoring a common newsfeed, a workgroup computing environment supported by an intranet, or the entire Internet. With a few notable exceptions (SIFT and Tapestry), in our descriptions we have placed more emphasis on effectiveness than efficiency when describing design features and performance evaluations. This is not surprising, since most experimental work on text filtering has sought to demonstrate effectiveness and a small user population suffices for that purpose. Even the TREC evaluation, which requires filtering hundreds of thousands of pages of text, specifies only 50 topics each year.
Once adequate effectiveness has been demonstrated for small user populations, the task of engineering efficient implementations for widespread use of such systems remains. One alternative is to simply replicate the filtering system and then provide all of the content to each filtering system. Tapestry implemented a more sophisticated approach, demonstrating that an appropriate division of effort between server-side and client-side computing can improve overall efficiency.
In general, he goal of distributed computation is to optimize the tradeoff between distributing the workload and minimizing communication requirements. Yan studied this issue rigorously in conjunction with his work on SIFT, developing optimal assignments of computational tasks among a group of cooperating servers . The GroupLens project has chosen an alternative approach that exploits an existing infrastructure for document distribution. By augmenting this infrastructure with distributed annotation servers, GroupLens expects to achieve acceptable efficiency in a manner compatible with the existing physical and social structure for Internet News. Thus, one of the key issues to be addressed as the number of users scales up is which constraints to accept and which to attempt to change.