Adaptive User Experiences in the Cultural Heritage Information Space

摘要

Given the thematic diversity, richness and variance in exposition of published cultural heritage information and artefacts, accessing pertinent information can be a cumbersome task. The TAGCLOUD project aims to create an adaptive cultural heritage experience for individuals based on their personal preferences, allowing users to navigate with ease around both cultural artefacts and the related information space. Users will establish a narrative between themselves and their cultural heritage experience. We propose metrics and methods for making the transition from a pull-based dynamic to a successful push-based methodology. Users are inevitably overwhelmed by the volume and specificity of cultural data, so traditional query-based interaction (e.g. filtering and sorting) is insufficient to guarantee a relevancy to the user of the retrieved information. Further, the small form factor of mobile devices poses strict limitations on the complexity of the interface and interaction methods available. The TAGCLOUD system applies content personalisation and context aware techniques from web search and marketing, to the realm of cultural heritage. We incorporate the geographical, chronological, historical and narrative relationships between cultural items, and span levels ranging from entire cities to individual artefacts. For each of these levels it is important to broadly define the possible ways the experience can be tailored. Information may be presented via different modalities, including audio, text, and augmented reality; and can vary according to an individuals interests and level of understanding. The context of the user can affect how and what is delivered, and may depend on their location, familiarity with their surroundings, or who they are with. Information and media should be presented so as to complement the experience and not detract from it. We investigate how we can retrieve information about the user both passively and actively. Information from the users device allows us to investigate their interaction with artefacts, and enables the system to form assumptions of their respective interest levels. Additional information is procured from social networking information, such as local graph traversal, and interactions related to the cultural heritage experience. We investigate how preference is extracted from the user model, how the system mitigates against destructive feedback that would show inappropriate suggestions. We propose the use of non-normative expressions of preference, to circumvent the tendency towards the populist mean, a generic weakness of ratings-based recommender systems.