|ASIS&T 2005||START ConferenceManager|
In June of 2003, a Raytheon intranet search and browse survey was administered to a random set of Raytheon employees and had 199 respondents. Only 26% of respondents indicated the then-current capabilities were “Good” or “Excellent.” In November of 2004, 3 months after a new deployment of enterprise search capabilities to 60,000 users, 69% of respondents to the same survey indicated the new system was “Good” or “Excellent.” This turnaround was part of a $2.3M effort that involved 3 internal company groups and 3 external vendor companies. Internal knowledge representation, content management, and intelligent search teams participated closely with external vendors across the project life cycle.
The Information and Knowledge Management program was funded in 2004 for $2.3M to begin to develop a multi-pronged system-based approach to this problem. A “knowledge representation mirror” concept was developed using a metadata registry to keep a consistent, manageable representation of taxonomies and metadata, and to evolve into an ontological registry for the company. The knowledge representation team developed and deployed over 20 taxonomies and metadata schemas. The content management team employed the metadata schemas for content publishers. The intelligent search team used metadata schemas and taxonomies for information retrieval and navigation in enterprise search.
Search, classification, and recommendation engines -- combined with a content management system and a metadata registry -- form the core systems supporting publishing, classifying, storing, and retrieval of document and web site information for the enterprise search application. Many lessons have been learned in the deployment of taxonomies for enterprise search. We believe the most significant problems were uncovered and addressed before deployment because of the user-centered design approach we followed.
While the project has initially been successful, many challenges remain. The recommendation engine currently captures objects transactions in the search engine, consisting almost entirely of unstructured information. Thousands of structured repositories behind hundreds of applications store related valuable information that could be leveraged in intelligent systems integration scenarios. A metadata registry approach is being taken to attempt a semantic integration of these systems.
We are extracting the models for a range of system types, and using the models to integrate the systems into our existing classification and navigation schemes. The set of recommended object types within enterprise search can then be expanded. Most importantly, we can continue our user-centered design on specific high-value opportunities in an “extensible search” environment we are deploying in 2005.
Proposed audience discussion items include:
How far can recommender systems take us in solving knowledge transfer problems? Is enterprise search the right place to start? If not, what is the best opportunity? What about personalization? Especially in portal environments? Has this failed? What do we now know that can take us to a next-generation digital workspace? Can we expect semantic integration technology to keep apace of our problem?
Discuss this on the ASIS&T 2005 Annual Meeting wiki!
|START Conference Manager (V2.49.6)|