Multimedia Search

LinkedIn Salary - Proposed a two-step framework that utilizes a novel, semantic representation of companies (Company2vec) and a Bayesian statistical model to computing robust, reliable insights at the company level when there is limited or no data at all. - Improved salary insights by 35 times, and active user coverage by 600%+. - Related publication: Chen, X., Liu, Y., Zhang, L. and Kenthapadi, K., 2018. How LinkedIn Economic Graph Bonds Information and Product: Applications in LinkedIn Salary. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (pp. 120-129). ACM.LinkedIn Jobs Marketplace- Application Forecasting: Proposed two-stage Bayesian Poisson-Gamma state space model with variational Bayes and Newton-Raphson method to deal with data sparsity and improve prediction accuracy.- Jobs Pacing: Proposed and implemented a new pacing algorithm that improves matching between demand-side and supply-side of job, and achieved a 1.5% lift in daily revenue.

RA/TA in the Department of Statistical Science. Publications see Google scholar homepage.Research Experiences Bayesian analysis, prediction & monitoring of dynamic networks for adaptive ad placement - Bayesian dynamic flow model for non-Gaussian streaming data, in contexts of web browser and user preference studies, focused on effective and scalable methods for comparisons and decisions - Bayesian methods for short-term forecasting, monitoring and structural change detection in dynamic networks, in contexts of adaptive optimization in e-advertising - Related publications: - Chen, X., Irie, K., Banks, D., Haslinger, R., Thomas, J. and West, M. (2018). Scalable Bayesian Modeling, Monitoring, and Analysis of Dynamic Network Flow Data. Journal of the American Statistical Association, pp. 1-15. - Chen, X., Banks, D. and West, M. (2018). Bayesian Dynamic Modeling and Monitoring of Network Flows. arXiv preprint arXiv:1805.04667. - Work presented at 2016 Joint Statistical Meeting, 2016 NBER-NSF Time Series Conference, etc. Dynamic relational data analysis with application in co-movement of international stock indexes - A multivariate Bayesian dynamic linear model with thresholding for streaming binary relational data - Modeling contemporaneous dependence structure through simultaneous parental specificationRiemann manifold Metropolis adjusted Langevin algorithm with quasi-Newton method - Markov chain Monte Carlo acceleration using Langevin diffusion as proposal with Metropolis adjustments - Fisher-Rao metric tensor as local configuration approximated with quasi-Newton methodTA Experiences: STA941 (2017 Spring): Bayesian Nonparametrics STA961 (2016 Spring): Statistical Stochastic Processes

Proposed a sequential Bayesian optimization algorithm with explore-and-exploit strategy for multiple-objective optimization with applications in job search re-ranking; implemented both offline large-scale data processing & online experiment sampling algorithm.

Developed online anomaly detection & offline hot-spotting algorithms for metrics monitoring at LinkedIn with applications in site speed monitoring for server status detection; proposed & implemented pattern recognition algorithms based on scan statistics & Bayesian dynamic modeling.