Use of Granularity and Coverage in a User Profile Model to Personalise Visual Content Retrieval

The enormous volume of visual content available from cameras and from recordings stored in data centres and the phenomenal number of users presents a major challenge to the research community. This major challenge is the use of adaptive techniques for personalised retrieval and filtering mechanisms in order to find relevant visual content appropriate to individual needs but without overloading users by retrieving uninteresting content. This paper presents a user modelling framework, which integrates a statistic model based upon latent semantic indexing (LSI) and a first-order logic (FOL) type knowledge-based technique, in order to acquire static and dynamic user preferences. Consequently, the framework is able to detect shifts in user interests. The framework is also able to construct a user profile at appropriate levels of granularity and coverage by taking advantage of concept properties, relations and the distance between nodes in users' model of a domain with respect to a common or global domain model. In addition, terminological problems such as ambiguities are solved by exploiting an external lexical reference system, WordNet.