TOWARDS EFFICIENT PROCESSING OF NEIGHBOURHOOD ANALYTICS FOR ADVANCED APPLICATIONS

Abstract

With the increasing variety and volume of the data produced by today's applications, the adoption of effective analytics becomes remarkably demanding. Window functions, being an important part of SQL family, have proven numerous successes in relational analytics. A window function assigns each tuple a set of related tuples, on which analytics can be applied. However, the window function de nes the related tuples based on sorting which limits its usage in the domains where sorting may not be meaningful. In this thesis, we generalize the concept the window function to neighborhood analytics which eliminates the stringent sorting requirement. We propose three domain-specifi c queries tailored for advanced applications on the basis of two simple neighborhood functions. Then, we study how to process these queries efficiently given today's data scale.