FAQ Finder Information Maintenance System

The AI Group of the ICS Department at UC Irvine

Customer Contact: Robin Burke

Summary

Build a information maintenance system to keep the FAQ Finder question answering system synchronized with an on-line FAQ repository.

Background

FAQ Finder is a natural language question answering system that serves as a testbed for research in natural language processing, intelligent information retrieval, and machine learning. The system answers questions by matching them against a corpus of questions and answers drawn from regularly-posted USENET FAQ files.

FAQ Finder was running for approximately 15 months at the University of Chicago as a public web utility, and is about to be started up at UC Irvine.

Problem

FAQ Finder currently contains approximately 2000 files mirrored from the RTFM FAQ archive in 1996. Many of these files are now out-of-date or obsolete, and there are new files in the archive that have never been integrated into the system. The original design of the system called for continuous updating of the FAQ corpus, but this part of the work was never completed.

There are a number of steps that must take place to keep an up-to-date FAQ archive:

1. Mirroring: The RTFM archive must be regularly mirrored, downloading new and changed files.
2. Flattening and collating: The files in the archive are organized in a directory hierarchy that is incompatible with FAQ Finder's information retrieval system, so the hierarchy must be collapsed into a single large directory. Also, long USENET FAQs are typically stored in several chunks, which must be combined together in the correct order to create a single file.
3. Database update: FAQ Finder works from a flat file database of information about its FAQ files. As new files come in and old files change, this database will need to be updated.
4. File preparation: There are a number of existing tools (mostly written in perl) that are used to prepare files before they can be handled by FAQ Finder. The tasks handled by these tools include the identification of file type, the extraction of question/answer pairs, the identification of important phrases, and other steps. The maintainer system will need to organize the application of these tools to new and updated FAQs, and provide a web interface to those tools that are only semi-automated.
5. Versioning: For research purposes, it will be important to maintain snapshots of the state of the FAQ archive at different points in time. For example, we may need to run a series of experiments against a fixed FAQ corpus to test new algorithms, meanwhile keeping an updated corpus in the public version of the system.

Project

Define, specify and build a mechanism that will keep FAQ Finder up to date and manage the steps required to integrate new and changed FAQ files into the system. The interface to the system should be web-based, and allow for complete management of the updating process.