Link Prediction

Apply Link Prediction algorithms on Massive Dynamic Social Network.

Perform in-depth comparison of existing metric-based and classification-based link prediction algorithms.

Take advantage of temporal information to improve Link Prediction performance.

Paper published in proceedings of IMC 2016.

Technical Analysis on U.S. and Chinese Stock Market

Apply technical analysis methods on stock historical price data of the largest stock exchanges in U.S. and China.

Quantitatively measure the predictive power of technical indicators in both markets.

Provide strong evidence to herd behavior on inefficient stock market.

Paper in submission.

User Behavior Clustering on Clickstream Data

Cluster user clickstream data of Whisper and Renren social network for user behavior modeling. The model is hierarchical and provides different levels of details of user behavior.

Develop a visualization tool for behavioral cluster display.

A user study shows that the model is easily understandable by experienced human annotators.

Paper published in proceedings of CHI 2016.

Investigate Online Stock Discussion Forum

Analysis sentiment of posts and user behavior of online stock discussion forums.

Identify key user groups and information flow between users, which suggests that users on online forums are isolated.

Find that user sentiment is not correlated to stock price, so that no sentiment-based trading strtegy is profitable.

The above 2 points provide evidence for echo chamber effect, which contributes to noise trading on stock market.

Paper published in proceedings of ICWSM 2017.

Large Scale Analysis of Venmo Transaction Behavior

Crawls full history of anonymized user transactions and social connections of Venmo.

Venmo's social and transaction network is densely connected and forms tight communities.

Venmo's network structure formation is driven by transactions in two different purposes: user-user transaction and user-vendor transaction.

Paper published in proceedings of ICWSM 2017.

Gender Bias in the Job Market

Develop large scale algorithm to detect gendered wording in job advertisement text.

Apply our algorithm on 10 years job posts on LinkedIn.

LinkedIn job market shows decreasing masculine bias over the 10 years. This can be account to different reasons including shift in job market, and changes in language use.

A user study finds that gendered wording only has limited effect on job application decision. Inherent bias (stereotypes) dominates more.

Paper to appear in proceedings of CSCW 2018.

User behavior on digital wallet sytems

Quantitatively measure network structure of Venmo transaction network and Venmo social network.

Identify different usage patterns from Venmo network analysis.

User study on social factors that drive adoption and affect transaction experiences on Venmo and WeChat Pay

Paper published in proceedings of ICWSM 2017 and to appear in proceedings of HICSS 2019.