Quantified All: My primary project focuses on mining information about people’s actions and their outcomes from social media.  The goal is to build a large-scale knowledge base of actions that can be used to help people with decision-making, planning, reviewing, and goal achievement.

Social Media Bias: I’m interested in experiments and studies of self-reporting, population and other biases.

Heterogeneous-Intent Networks: Recognizing that people’s motivation for participating in social networks varies greatly (fame, friends, commercial, …), we study the implications of heterogeneous intent for modeling information diffusion in social networks.


Past Projects


U Rank, a search engine front end to support light-weight search editing and sharing. U Rank was prototyped externally in October 2008. [readwriteweb|pcworld|seattlepi]

The Social Web Experience browser toolbar analyzes Web pages and finds related content from your social networks. Your friends’ recent status updates and messages, favorite movies, interests, and other profile information are shown when its related to what you’re seeing on the Web.  This project motivated our work on entity recognition in social media, now in use in several products at Microsoft.

AjaxScope, aka AjaxView, is a tool that allows web app developers to efficiently monitor the real-world, in-browser behavior of their JavaScript code without special browser plugins or extensions. We have an SOSP paper about the project and released our research prototype. AjaxScope also became Visual Studio 2008 AJAX Profiling Extensions Power Tool (say that 3 times fast!).

Doloto is a system that analyzes application workloads and automatically performs code splitting of existing large Web 2.0 applications. Doloto improves page-load times for large complex web applications by 20-40%. See our FSE paper and download the prototype at DevLabs.

Internet service architectures and operations: Internally, I’ve also worked on surveying and collecting best-practices and lessons from the architectures, operations and reliability of Microsoft’s large-scale Internet services.

KISS: The Keeping Information Safe from Social networking apps” project protects the privacy of user data from untrusted 3rd-parties.  A combination of information flow control, sandboxing and differential privacy, implemented across a cloud hosting infrastructure and web app framework enforces strong limits on how user data can be used by application providers.

Application-level fault detection and diagnosis in large-scale internet services:  My PhD was on using machine learning to detect and diagnose  high-level application faults in Internet services. I continued this work at MSR, with more of an end-to-end perspective, looking at wide-area network failures as well as web application challenges.