The Predict project – Novelty, diversity, context and time: new dimensions in next-generation information retrieval and recommender systems
Funded from January 2012 to December 2013 by the Spanish Ministry of Science and Education (TIN2011-28538-C02), the Predict project addresses the extension of information access technologies towards the incorporation of new dimensions, namely novelty and diversity, subjectivity, context, and time, as essential aspects of information access effectiveness and tasks that are not captured in traditional Information Retrieval (IR) paradigms and models, in order to enhance the quality, effectiveness, and user satisfaction with retrieval systems, in further ways than state of the art technologies and theories in the field currently procure. The research focuses on these problems within the specific scope of personalized Information Retrieval and Recommender Systems.
Acknowledging the importance of novelty, diversity and discovery as fundamental qualities of the effective information value for users, the project seeks the definition of new methods and techniques to enhance novelty and diversity in personalized search and recommender systems, as well as methodologies and formally grounded metrics to assess novelty and diversity. Making progress in these directions involves overcoming classic IR paradigms, and revising traditional assumptions on relevance such as being topical, independent, consistent, and stable. Context modeling, and the fundamental relations of novelty and diversity to context, personalization, and time, to a large extent unexplored in the literature, will be another key focus of the project.
New methods and paradigms will be researched that properly represent and leverage contextual information to enhance the effectiveness of retrieval and recommendation, focusing on the social user context and the temporal dimension involved in several elements of information access tasks and IR personalization. The use of recommender systems for marketing purposes considering the business context, recommender system hybridization, and the integration of heterogeneous sources of evidence, will be researched as further diversity-enhancement approaches.
The proposed research also refines the notions of personalization in information access towards their application in structured retrieval tasks, revising the definition of appropriate personalization strategies in the context of a flexible information and document granularity. Related with personalization, since the set of retrieved/recommended items depends on the user, we will explore techniques able to explain which are causes leading the system to match these particular items with a particular user.
Finally, the project addresses fundamental methodological aspects in experimental research related to the evaluation of recommender systems, addressing the current gaps in this particular area. The sound adaptation of IR evaluation methodologies to recommendation tasks, and the unification of IR and Recommender Systems models and paradigms, will be undertaken as background objectives in the project.