Zikun’s research
1 Synthetic Data Augmentation for Database Entity Recognition
We study the Database Entity Recognition (DB-ER): identifying database tables, columns, and values in natural-language queries (NLQs), a core but under-isolated subtask in text-to-SQL systems. We reformulate DB-ER as a closed-domain NER problem grounded in database schemas and SQL structure. We construct a benchmark derived from Spider and BIRD, introduce an automatic SQL-guided data augmentation method to scale training data, and propose a T5-based two-stage tagging model. Experiments show that synthetic supervision and encoder fine-tuning substantially improve precision and recall, and that the proposed approach outperforms strong NER baselines such as LUKE and Flair on DB-specific entity types.
Key Contributions:
- DB-ER Benchmark from Text-to-SQL Data
- SQL-Guided Data Augmentation via Synthetic Annotation
- Specialized T5-Based DB-ER Model
- Evaluation and Ablation Analysis Against Strong Baselines
This work was under the supervision of Professor Ken Pu and Professor Kourosh Davoudi.
2 GREx An Educational Survey Dashboard (https://grex.eilab.ca/)
I was responsible for the development of the GREx An Educational Survey Dashboard, which is a web application that allows users to create and take surveys. I was responsible for the development of the front-end and back-end of the application. I was also responsible for the development of the database schema and the deployment of the application.
This work was under the supervision of Professor Roland van Oostveen, (https://eilab.ca/)