What roles can software agents play?
The simplest way to implement web “thinking” is to create a specialized software agent. This is a program that works as a “delegate” of its user, autonomously collecting information that is likely to be interesting to its user. The agent can learn the user’s preferences simply by observing which pages the user actively uses, or it can receive specific instructions (e.g. keywords) from the user. Given that preference profile, the agent can locate pages that satisfy the profile, and then use “spreading activation” to find further, related documents. It does this by “activating” pages in proportion to their degree of interestingness, and then propagating that activation according to the hyperlinks and their weights as learned from other users. Thus, it can discover new documents, that may not contain any of the initially given keywords, but that are still highly relevant to the query. This is especially useful when the user cannot clearly formulate the query, but only has an intuitive fee
