ADAM

ADAM - Adaptive Information Memories

The objective of the ADAM project is to find new approaches to enhance existing Information Retrieval (IR) systems. We currently deal with two main ideas, both concerning Web IR systems.

SEREBIF

One approach is to include implicitly collected information about user preferences and behaviour into a typical web search engine and evaluate how to use this information for enhancing the search engine. More precisely, we observe what the users are doing in interaction with a search engine (i.e., which queries are being entered, which results are chosen, and how long these results are approx. viewed). This information can be of great value because it gives hints as to which of the results seem to be indeed relevant for the users. To store and process the gathered information, we use the ANIMA system that has previously been developed in the MINE group.

We want to enhance the results of the search engine by this way, e.g. by re-ranking them according to the information derived from the captured user interaction. Additionally, the collected information about entered queries can be used to suggest new search terms to users, allowing them to refine a search into the desired direction and possibly eliminating ambiguities that might exist for a query.

The prototype we implement to realize this is called SEREBIF. For more detailed information about this approach, check out the detailed page about SEREBIF.

FAQ Generation

Another idea we're working on is to analyse web content for the purpose of generating FAQ (Frequently Asked Questions) documents. Specifically, we focus on a certain, limited document collection (the contents of an internet forum). Using the data contained in there, we try to discover patterns out of the threads and posts in the forum as well as getting information about the structure. In particular, our goal is to find out what common questions are being asked in a forum and what possible answers people are getting for these questions.

An automatically created FAQ repository like this could e.g. be helpful to find out about the most relevant topics in the forum, or for an administrator to manually create a refined version of an FAQ on the basis of the results given by the automatic analysis.

More detailed information about this project will be available soon.

Project Members

* Christoph Schommer
* Jayanta Poray
* Ralph Weires

Keywords:

Information retrieval; implicit feedback; search engines; re-ranking; FAQ generation

"ADAM" is mentioned on: Jayanta Poray