Deliverable D4.3 Content and Concept Filter V1
LinkedTV is dedicated to the widespread and rich domains occurring in multimedia content on the Web. In such rich domains it is essential for the users to get support in finding the kind of content they are interested in and to make use of the rich relations between multimedia items on the Web. User models are used to represent the different kinds of interests people may have in multimedia content. In this document we describe how a user model (introduced in deliverable D4.2) can be used to filter multimedia content in various ways and to support the user in this way to manage the large amount of multimedia information available on the Web. A user model contains two main aspects: a description of the user himself (age, profession, social status, etc.), and a representation of those things in the world he is interested in. Whereas his personal description results in a set of data the representation of his interests needs a more complex form. User interests typically cover a broad spectrum of topics represented in a user model ontology (LUMO). It represents the mental model of the user, i.e., the main concepts, topics, concrete entities, and semantic relationships between them he maintains about the world. The entities in this user model ontology are related to items in various LOD ontologies like DBPedia, schema.org, the music ontology, etc. This enables us to use the LOD universe as semantic background for user modelling. The different degrees of interest a user has in various topics are represented as weights for each element in the user model ontology. The semantic annotation process in LinkedTV enables fine grained annotations of media fragments (see the LinkedTV deliverables D2.2 and D2.3). A video as a whole as well as scenes in it or even single shots can be annotated. The multimedia fragments are annotated with elements from LOD ontologies (URI) like DBPedia, music ontology, etc. They are interlinked to other entities on the Web. Our content filtering is based on weighted semantic matching. It can be used in different ways: enriching information about an object shown in a video scene or frame with linked information from the Web; ranking annotation elements occurring in a frame according to the user’s special interest; or determining semantic similarity between media fragments and providing user recommendations. Six concrete user models are described in this document in order to illustrate our approach showing how different user interests can be and what it meaans for their media consumption. A first version of the LinkedTV semantic filter LSF has been implemented. It takes semantic user models and semantically enriched media fragment annotations to compute rankings of media content w.r.t. specific user interests. Additionally, we show how a logical reasoner (f-PocketKRHyper developed by our Partner CERTH) can be used with its logic based user model components to post-process the filtering results by using fuzzy logic reasoning.