Smart Data is the new trend following the Big Data hype. The focus shifts from pure mass towards quality of data and the added value that data analysis can provide. To this end, Location-Based Social Networks (LBSN) are emerging as new area of application due to a vast amount of available personal information and the predominant demand for personalized user recommendations. In this social environment, personalization is extended from individual users to groups of users. Application scenarios include the recommendation of items to a group of users, recommending one user to another, or forming groups of users via recommendation of users to groups.
As a work of applied computer science, the goal of this thesis is to systematically establish solutions for group preference problems of mentioned application scenarios. Technical challenges are met by following a database approach which is scalable to LBSN datasets of Big Data magnitude and allows for a fast computation of recommendation results. Semantic aspects are met by application of Preference SQL as database-driven preference framework which facilitates data-adaptive recommendations. By this design, user models can be formulated and evaluated on social network data and recommendations can be easily integrated into existing system architectures.
The development of individual solutions requires extensions of Preference SQL towards geo-social domains and group preferences. For item-to-group recommendations, novel means for the individual statement of preferences as well as the formation of group preferences become of importance. Applications in user-to-user recommendation require data aggregation from different user profiles and subsequent preference analytics for the generation of extended implicit preferences. In addition, the definition of data-adaptive similarity measures based on social and psychological aspects of real-life interactions is a major factor. These concepts are leveraged for user-to-group scenarios to implement effective strategies for problem instances of large user sets.
In the course of solving these practical application problems, superior theoretical questions emerged, such as heuristic algorithms for hyper-exponential solution spaces, that are also addressed by this work. The validity of proposed solutions is verified by demo applications and the implementation of semantic benchmarks.
