Expert Recommender Systems in Practice: Evaluating

Summary:
This article had to deal with the concept of knowledge management (KM) which has had several leaps and bounds with how it is being used. A second wave of KM applications that would share knowledge among social networks and human actors (those seeking the knowledge) are now being postulated. This paper also has to deal with expert recommender systems (ERS ) which allow the finding of appropriate knowledge carriers based on an expertise profile. The difficult thing about expertise profiles for the actors is the quick, effective and useful design of the profile to be used in the ERS . This paper discusses an ERS which combines self reported information with keyword mining from a user's files. The basis of this ERS system as well as any other ERS is to take an input description of a needed piece of knowledge and output a list of sources (users with profiles) who have the highest outcome of possessing that knowledge. This ERS system was studied on a European industrial association called the NIA which offers services such as networking among member companies, legal regulations, standardizations and many other things to its users. The NIA is a highly decentralized organizational structure with many gaps in the transfer of knowledge between sections. This study was aimed at the increased sharing of knowledge between departments and members of the association. The ERS tested on the NIA was called the ExpertFinding which was created by doing the following: Studying the organizational needs, design a prototype to meet those needs, Roll out and evaluate the designed system. ExpertFinding's main purpose is to help redirect question requests to the person in the association with the highest likelihood of being able to competently answer the question. Profiles for ExpertFinding were created based off of two things. First, there was a large scale keyword list based off arbitrary text documents which would help establish a user's competencies. The second thing it was based off of was a listing of contact information, and other facts about the user that would help personalize their profile (education, job description etc.). The patterns of usage among the testers was that they need to feel adequately represented by the system to want to use it. Some of the testers wished there was a filter system to take out irrelevant terms in the searches. Some of the problems encountered where: the LSI matching algorithm was not the most efficient and the selection mechanisms were not best for the file system that emerged. One of the proposed solutions was the creation of a NIA specific thesaurus which would help filter out irrelevant words and phrases. The findings showed that this ERS could generate accurate profiles for the job, but not always complete.

Discussion:
I thought this was a rather interesting article to read because it seemed like a good idea to find out and use in the working world. I know at my work there are times when I find out someone has already solved a problem I was working on after I solved it. It would be a great help to be able to look up who has worked on something similar before and go talk to them to see their findings. I think that a company specific thesaurus would be a good idea for ways to improve it.

Comment:
http://computerhumaninteractionblog.blogspot.com/2010/02/
social-computing-privacy-concerns.html#comment-form

http://shauntgo.blogspot.com/2010/02/uist-predicting-tie-strength-with.html
#comment-form