Rutgers University · Ph.D. in Computer Science
BiLink: Bidirectional Meta-paths for Link Discovery in Billion-Scale Heterogeneous Graphs
I am a Ph.D. candidate in Computer Science at Rutgers University, advised by Prof. Desheng Zhang. I work on distributed processing and learning over tables represented as graphs, especially on long-range structures under neighborhood explosion. An online demonstration of these methods predicts an author's potential new coauthors and references. I expect to graduate in Spring 2027 and am looking for research positions in this area.
Research
Enterprise data is typically stored across multiple tables with different schemas, so each prediction task requires its own manual feature engineering. If we choose certain columns as entity identifiers, each entity corresponds to one node, and each row connects the entities it involves, allowing a model to automatically learn from the tables as one unified graph. Because this data is usually too large for a single machine, my research focuses on distributed processing and learning over tables represented as graphs, especially on the effective and efficient use of long-range structures under neighborhood explosion.
Selected Publications
Complex-Path: Effective and Efficient Node Ranking with Paths in Billion-Scale Heterogeneous Graphs
Paths2Pair: Meta-path Based Link Prediction in Billion-Scale Commercial Heterogeneous Graphs
Outside In: Market-aware Heterogeneous Graph Neural Network for Employee Turnover Prediction
Applications
New Coauthor and Reference Discovery
Built a link discovery framework and used it to predict an author's new coauthors and references, on a billion-scale academic graph built from OpenAlex. It selected more than 1.4 billion candidate pairs for each task and ranked them by predicted probability. Using only the top 1% of the ranked pairs, it recalled over 31% of the actual new coauthors and references. A demonstration lets KDD 2026 attendees explore their own predictions.
Key Personnel Discovery
Applied the same framework to find the unknown key personnel of companies, on a billion-scale graph built from JD Logistics' internal data. Contacting the people it discovered raised the contact success rate by 86% over the previously deployed expert-designed rules. It was then scaled nationwide, selecting 1.8 billion candidate company–person pairs across 200 million companies.
High-Value Customer Ranking
Built a node ranking framework and used it to rank customers nationwide by their potential to become high-value, on a billion-scale graph built from JD Logistics' internal data. Calling the customers it ranked highest raised the success rate by 252% over the previously deployed expert-designed rules. Since December 2022, it has continuously ranked customers for sales outreach.
Education
Peking University · M.S. in Electronics and Communication Engineering
Northeastern University · B.S. in Electronic Information Engineering