Abstract
Online Social Networks (OSN) are frequently used to find people with common interests, though such functionality is often based on mechanisms such as friends-of-friends that do not perform well for real life interactions. We demonstrate an integrated database-driven recommendation approach that determines reciprocal user matches, which is an important feature to reduce the risk of rejection. Similarity is computed in a data-adaptive way based on dimensions such as homophily, propinquity, and recommendation context. By representation of dimensions as unique preference database queries, user models can be created in an intuitive way and can be directly evaluated on datasets. Query results serve as input for a reciprocal recommendation process that handles various similarity measures. Performance benchmarks conducted with data of a commercial outdoor platform prove the applicability to real-life tasks.
Authors
Florian Wenzel, Werner Kießling
Publication
LNCS 9828, Springer, pp. 3-10, 2016