Personalized information retrieval and search promises to improve the Internet experience. An important requirement for building personalized web applications is to build user profiles that represent the users' interests. There are two representations commonly used for user profiles. One is using frequently occurring words in user documents. This creates large profiles where profile terms have low precision and have insufficient context to determine the user interests. The other is using a pre-existing ontology such as DMOZ. While this approach alleviates the ontology creation and maintenance problem, it requires constructing classifiers for each DMOZ node. Besides, of all the topics in the DMOZ ontology, most people will have only a small fraction of the topics as their interests and hence most of the ontology is redundant for capturing the interests of a specific user. This paper presents an alternative method to construct a hierarchical user profile using Wikipedia as the vocabulary for describing the user interests. The profiles created in this manner are more compact and have high precision compared to profiles that use words. We also discuss a method to tag concepts in these profiles as being of recreational or transactional interest.
PDF
download
878987797
028-85220240
