Events Calendar

PhD Defense - Yi Chang

Title: Time-Sensitive Social Media Mining and its Applications

PhD Candidate: Yi Chang

Date / Time: Jan. 11th (Monday), 1:00~3:00 PM

Place: SAL 213

Committee:
Prof. Yan Liu (Chair)
Prof. Cyrus Shahabi
Prof. Lian Jian (external)

Abstract

Social media is growing at an explosive rate and it becomes increasingly difficult for users to consume and digest useful information from massive and high-velocity data. To overcome this information overload problem, in this thesis, we have studied several key challenges, which could effectively re-structure and re-organize massive information on social media sites.

First, it is critical to effectively detect and model a burst of topics on social media, which is reflected by extremely frequent mentions of certain keywords in a short time interval. We propose a novel time-series modeling approach which captures the rise and fade temporal patterns via life cycle model, then invent a probabilistic graphical model to automatically discover inherent temporal patterns within a collection of buzz time-series. Second, as each individual tweet is short and lacks sufficient context information, users cannot effectively understand or consume information on Twitter, which can either make users less engaged or even detached from using Twitter. In order to provide informative context to a Twitter user, we initiate the task of Twitter cascade summarization, and propose a supervised learning framework with a set of novel features to generates a succinct summary from a large but noisy Twitter context cascade. Third, we address the challenge of timeline detection from social media, which is to detect a chain of spiking events in chronological order, and it can help social media users not only rediscover the most important historical events about entities but also understand the order and trends of those events. In order to capture the life circle patterns of events in timelines and combine temporal shapes with content information, we propose a novel probabilistic framework to effectively detect timelines of entities in social media. Finally, we address the challenge of timeline abstracting from social media, which is to detect a list of timeline events, and abstract each events with its representative social media post. In order to automatically identify the number of timeline events, we propose a non-parametric framework to effectively and efficiently detect and abstract timeline events. Gibbs sampling is employed to infer the model parameters, and a fast burn-in strategy based on temporal bursts is further introduced to speed up the model inference.

Location: Henry Salvatori Computer Science Center (SAL) - 213

Audiences: Everyone Is Invited

Contact: Lizsl De Leon