Browsing by Author "Derakhshan, Behrouz"

Maximizing influence in graphs, typically applied to Social Networks, is the problem of finding a set of nodes with the highest overall influence on the entire graph. In marketing domain for example, it is used to find the set of people who have the highest influence on their local communities. As a result, instead of blindly marketing a product to a large group of people, the product is marketed to this group of selected users, and they will in turn help spreading the word. The problem has been studied extensively, and several state of the art methods have been proposed. But all of these methods have one common flaw, none of them are scalable. Even on small graphs, current methods take extremely long amount of time and introduction of bigger data sets have rendered some of these methods completely useless. Over the past two decades, collection of data has become easier and a very common practice. This is mostly credited to the advancements in hardware and software technologies as well as the introduction of World Wide Web. To overcome issues related to big data sets, large scale data processing platforms have been developed to tackle scalability issues of problems similar to the influence maximization. Most notably are the two frameworks called Hadoop and Spark that contain many features for simple data processing, machine learning and graph processing. In this thesis work, some of the current influence maximization algorithms are implemented in these two frameworks, some new methods are proposed, experiments on graphs of different sizes are performed and the results are reported.