 |
 |
 |
 |
| |
|||||
|
|
|
"Who knows what in your company?" That is the question that the Expert Finder project wants to answer. We built Expert Finder to help users quickly find people who know about a particular subject area. In creating this system, we wanted to use only existing search engines. At MITRE we use the MITRE Information Infrastructure (MII), an automated information center, which allows employees access to technical documents, resumes, and basic employee information. By using just the MII, we avoid creating a separate, single-purpose "skills database." Experience has shown that a skills database is difficult to maintain and quickly becomes obsolete. By mining documents already on the MII, the results from the Expert Finder system are always current. As information about new technologies is added to the MII, the Expert Finder system can locate the people associated with those technologies with no additional effort. Expert Finder uses the native MII search engine to examine the many sources of information available on the MII for documents containing a keyword phrase such as "human computer interaction." The list of documents returned is categorized into one of two types: publishing documents, (documents published by employees) and mentioning documents (documents such as in-house newsletters that mention employees together with projects and technologies). Once the documents are found, the search for relevant experts begins. Since papers written by employees are indexed by employee number, it is simple to find the employees associated with each document. The employees found are then ranked according to the number of publications. Locating the MITRE employees in the mentioning documents is more complicated. Expert Finder uses a COTS (Commercial Off-the-Shelf) name extraction tool to locate the proper names within each document. It then performs a statistical analysis of names and query term, examining factors such as the proximity of name to the relevant terms and the number of times name and query term appear together. The names retrieved from the mentioning documents are combined with the names gathered from the published documents, and each name is matched with names in the MII phone book. The final ranking of each name is determined by factors such as number of documents, distance between name and query term, and types of HTML tags that appear between name and term. The names are then presented in order, as in the picture at the top of this article. The initial goal for the Expert Finder prototype was to create a system that could find people who were a phone call away from an expert on a subject. That way even if the person called didn't know a lot about the subject, he or she would know who did. The prototype that we developed has met this goal in most test cases. This system was created by Barbara Gates, David House, Inderjeet Mani, and David Mattox. For more information, please contact David Mattox using the employee directory. |
Solutions That Make a Difference.® |
|
|
