USTCAGI RAG Research Projects

Competition

🥈 Solution of Meta KDD Cup '24 CRAG

Our second-place solution for the Meta KDD Cup 2024 CRAG challenge.
Competition Research
Derived from KDD Cup '24

PruningRAG: Multi-Source Knowledge Pruning for RAG

A plug-and-play RAG framework that employs multi-granularity pruning strategies for optimizing the integration of relevant information. This work is derived from our experience in the KDD Cup 2024 competition.
Dataset

HoH: A Dynamic Benchmark for RAG

The first benchmark to evaluate the impact of outdated information on Retrieval-Augmented Generation (RAG), revealing that outdated information significantly degrades RAG performance by reducing accuracy and potentially leading to harmful outputs.
Survey

A Survey on Knowledge-Oriented Retrieval-Augmented Generation

Focusing on a knowledge-oriented perspective, this survey offers a comprehensive overview of RAG, examining its mechanisms, categorizing its approaches, discussing evaluation, highlighting applications, and outlining future directions within natural language processing.
Framework

Agent-R1: Training Powerful LLM Agents with End-to-End Reinforcement Learning

Agent-R1 is an open-source framework designed to accelerate research and development at the critical intersection of RL and Agent.