PhD Dissertation Defense by XIE Wei | Real-time Bursty Topic Detection and Virality Forecasting in Microblogs

Microblogs such as Twitter have become the largest social platforms for users around the world to share anything happening around them with friends and beyond. A bursty topic in microblogs is one that triggers a surge of relevant tweets within a short period of time, which often reflects important events of mass interest. How to leverage microblogs for early detection and further impact analysis of bursty topics has, therefore, become an important research problem with immense practical value.

To address the above problem, we propose in this thesis a framework which contains the following three parts. The first part is to select a budgeted set of users as information sources for early detection of bursty topics. An efficient algorithm is proposed to find such set of users from a large social network. The second part is that of detecting bursty topics from a tweet stream. We propose a two-stage integrated solution TopicSketch. In the first stage, we design a small data sketch which efficiently maintains at a low computational cost the acceleration of two quantities: the occurrence of each word pair and the occurrence of each word triple. In the second stage, we propose a sketch-based topic model to infer both the bursty topics and their acceleration based on the statistics maintained in the data sketch. A dimension reduction technique based on hashing is also proposed to achieve scalability. Once a bursty topic is detected, the third part will predict the further cascade size of this bursty topic. Based on a general time-aware cascade model, in which an appropriate hazard function is designed specifically for microblog networks, a simulation-based approach is proposed to forecast the virality of the cascade. Although each part can work separately, when putting them together, we get a comprehensive solution which is able to benefit many real world applications such as social network monitoring and viral marketing.