SessionPrint: Accelerating kNN via Locality-Sensitive Hashing for Session-Based News Recommendation

Traditional kNN methods, while proven to be accurate in session-based scenarios such as news recommendation, suffer from computational inefficiencies, especially when dealing with large datasets typical of real-world applications. This can lead to high costs for computing infrastructure as well as slow response times, during online recommendation generation. We propose an approach, called SessionPrint, that employs locality-sensitive hashing to reduce the time it takes to find neighboring sessions. Furthermore, we devise a multi-stage variant of our approach as well as a version that utilizes a final precision pass so as to drill down to the most fitting set of neighboring sessions in the most efficient way possible. We evaluate the performance of our approach in terms of both accuracy and efficiency on four real-world news datasets of varying sizes. The results confirm that SessionPrint not only reduces the time to generate recommendations but also maintains high accuracy compared to a traditional session-based kNN implementation, providing a scalable solution for real-world applications where rapid response times are crucial.