Towards a Domain Analysis Methodology for Collaborative Filtering

Collaborative filtering has the ability to make personalised information recommendations in the absence of rich content
meta-data, relying instead on collations between the preferences of similar users. However, it depends largely on
there being sufficient overlap between the profiles of similar users, and its accuracy is compromised in sparse domains
with little profile overlap. We describe an extensive analysis that investigates key domain characteristics that are vital
to collaborative filtering. We then explain how knowledge of these characteristics has helped to drive the design of a
collaborative recommender system for the JobFinder online recruitment service.