Next Generation Applications of Social Systems

点击翻译按钮获取中文摘要

Technology has heralded the advent of Online Social Systems such as Social Networks (SN) and Recommender Systems (RS). Data pertaining to these systems--both those that describe them and those that are generated by their users, are essential for research aimed at advancing their state of the art. Recently, increasing availability of real data sets on social systems, and at an unprecedented scale, has facilitated cutting edge research into the design and use of these systems. An important application domain that is common to both SN and RS is marketing.**In case of SN research, a major application domain is Viral Marketing (VM). The vision behind VM is to give free samples of a product to a small number of ``seed'' users (consumers) in a SN in the hope that adoption of the product will propagate virally through the network from the seeds to their followers, and their followers, and so on. The seed users chosen must be influential in effecting large cascades of adoption by others. The underlying thesis here is that users are susceptible to the influence of their neighbors: when they observe their neighbors perform an action like buying a product or adopting an innovation, then with some probability, they will be tempted to do the same. Numerous models have been proposed for capturing propagation phenomena which cover many applications including viral marketing, spread of infections, spread of innovation/rumor, etc. In spite of a decade of research with many advances, its penetration into real-world marketing has been modest. RS are the backbone behind the success of companies like Amazon and Netflix: indeed, recommending products to users by leveraging their past profiles has been found to boost product adoption and ultimately sales. RS build profiles on users/items using past user feedback and use the profiles to make future recommendations. The majority of research to date has focused on improving the accuracy of prediction models. While this is important, an equally important business perspective has been largely ignored.**In the proposed research, we will focus on this marketing domain and study several fundamental data mining and computational questions that will enable the development of novel next generation applications. Most prior work on VM ignores competition, is based on unrealistic models, ignores the fundamental role of the network owner, or proposes algorithms which do not scale to real SN with hundreds of millions of users. A specific goal of this program is to develop novel models for VM that close the gap between research and ground reality by lifting the above limitations, and design scalable algorithms. We will study novel data mining questions on recommendations motivated by a business perspective: how to make strategic recommendations that optimize expected revenue, while accounting for pricing and boredom effects? how to make efficient and flexible recommendations of not only items but of packages such as travel itineraries, shopping lists, and playlists, with minimal user input? how to recommend events which are evolving or being composed from scratch, to a big collection of users, such as attendees of a large convention? We also intend to study questions at the interplay of SN and RS, such as how to exploit the hidden influence channels in a RS to launch an effective targeted marketing campaign, taking into account complex recommendation models, competition between rival campaigners, and non-monotonic behavior of users who may switch loyalties.**Cutting edge research fueled by these questions will drive the technology forward and just as importantly, take both viral or network-driven marketing and recommendation-driven marketing much closer to reality, leading to industry-strength systems and applications.