Agents for the Web ??

Yezdi Lashkari (yezdi@media.mit.edu)
Sun, 6 Mar 1994 17:20:31 --100


Hello everyone. Our group at the Media Laboratory builds
autonomous interface agents for various classes of
applications that use machine learning and AI techniques
to learn their users patterns and take actions autonomously
on their behalf after a point. We've successfully built agents
for a variety of applications such as netnews, email etc.

We are seriously exploring the possibility of an agent for
WWW/Mosaic. The problem with the WWW is that new exciting
information comes into existence almost every day all over
the world. Some of it may be potentially interesting to
one. What we'd like is to embed an agent in one's favourite
WWW browser that learns what types of documents its user finds
interesting, contacts other users' agents over the net and
asks them to recommend new stuff they've discovered. Over time
agents learn which agents to trust and which not to, for different
classes of documents. Agents may also recommend peers or forward
URLs proactively to other agents that they feel may be interested.

Note that mapping the web is not the solution as

- the map is probably out of date pretty soon
- resource limitations probably prevent it being comphrehensive
- browsing through a massive database of www stats is really not
a solution to our problem

What we want is a distributed form of collaborative filtering for
all the dynamic information out there on the Web. In addition we
plan to combine this with a form of social filtering for groups
of peers and URLs which is very effective for non text documents.

Does anyone know of any work in this area ? I'd really like pointers
to any work that people know about vis-a-vis the Web/Mosaic and for
this sort of task. I also welcome comments/critiques on the proposed
approach's feasibility/performance/scalability etc.

thanks in advance

Yezdi