GradeSQL is a framework to generate training datasets and train Outcome Reward Models (ORMs) that score and rank LLM-generated SQL, offering an alternative to execution-based BoN and Majority Voting, with pipelines and benchmarks (Spider, BIRD).

Source code for the paper "Type-Less yet Type-Aware Inductive Link Prediction with Pretrained Language Models" accepted to EMNLP 2025

Comprehensive and Rigorous Framework for Reproducible Recommender Systems Evaluation

Accepted as full paper at CIKM '25

Universal Behavioral Modeling Data Challenge

Evaluates ChatGPT's recommendation capabilities beyond accuracy, focusing on fairness, diversity, novelty, and personalization. Includes code, datasets, and evaluation scripts for analyzing LLM-based recommender systems.

Accepted at ACM RecSys 2025