This page is maintained by Ian Soboroff [email], and was last updated on 17 March 1998. This page is ever changing; here is an older version, which was less focused on agents for IR. Please contact me with ideas, comments, corrections, or additions!
Oren Etzioni and Mike Perkowitz have a paper in IJCAI '97 entitled Adaptive Web Sites: an AI Challenge. In it, they propose that web servers will analyze user request patterns, and use this data to dynamically restructure their pages.
Sue Bogar and Ian Soboroff worked on using agents to understand specific types of web pages (i.e., university course descriptions), mark them up with semantic information using SHOE (Simple HTML Ontology Extensions), and draw inferences over the collection for interesting search problems, such as student academic planning.
Daniela Rus has several papers on using agents for document structuring and information access.
The Natural Language Processing lab at the University of Massachusettes (part of the Center for Intelligent Information Retrieval) conducts research into extracting information from unrestricted text documents.
The Aristotle project at Iowa State is the source of several projects such as Phoaks. The interest of the Aristotle project is developing systems to automatically categorize Web resources.
The Intelligent Software Agents group at Carnegie Mellon have two projects regarding information retrieval. DVINA is an agent for monitoring the web, USENET, or email using both statistical and knowledge-based techniques; as of this update, few details are available. WebMate is a personal Web agent integrating parallel search techniques, relevance feedback, similarity-fetching, offline browsing, etc.
The Softbots group at the University of Washington has been the source of several projects on mediating Internet searches, such as Ahoy and Metacrawler.
The Nobots group at the Stanford Robotics Laboratory as subsumed the Nobotics group. They are interested in adaptive IR agents, among other things, and have a collection of publications.
The local bibliography contains many references on collaborative tools for searching.
The KQML agent communication language provides semantics for mediated queries between agents.
Statistical approaches, such as n-grams and latent semantic indexing are particularly interesting for analyzing text objects, because they are independent of the language of the text, are resistant to noise (i.e., mispellings), and allow the application of many known mathematical techniques to natural language analysis.
The SHOE project (Simple HTML Ontology Extensions) at UMCP defines a set of HTML tags which can be used to embed semantic markup in web pages. Using SHOE, a personal web page could define the author as a graduate student in Computer Science at Wossamatta University, working with Prof. J. Q. Hacker, all in a software-readable fashion. Agents can then read these tags as well as an ontology definition, and compile a knowledge base on web pages.
AgentSoft has an XML demo to show how one could use XML in developing semantic markup for querying web resources such as push channels.
The Software Agents group of the MIT Media Laboratory has conducted several projects in constructing agents and collaborative tools for information filtering and discovery.
Marko Balabanovic at Stanford has written the FAB system, which presents a user with possibly interesting web pages, which the user evaluates. The evaluations are used to try to perform better searches.
Alexa is a web-page recommender service, where users rate pages they see. Alexa provides a compact interface which shows recommended "next stops" on the web based on your current page.
Project centers at Stanford University and The University of Michigan are using agents to construct intelligent interfaces, facilitate high-level searching, organize collections, and sophisticated searching over specific collection classes.
Site listings from UMich: