A Simple yet Effective Retrieval-Augmented Generation Framework for the Meta KDD Cup 2024

Liyang He, Rui Li, Shuanghong Shen, Junyu Lu, Linbo Zhu, Yu Su, Zhenya Huang

Keywords: retrieval-augmented generation, retrieval, ranking, large language model

TL;DR: We propose a RAG pipeline with data processing, retrieval, and chain-of-thought-based generation for Meta KDD CUP 2024 Task 1

Abstract: This paper describes our team's solution for the Meta KDD CUP 2024: CRAG Comprehensive RAG Benchmark Challenge Task 1 (Retrieval Summarization). The task involved building a retrieval-augmented generation framework and testing it on the CRAG benchmark. Our solution is a pipeline encompassing data processing, retrieval, and chain-of-thought-based generation. In this process, we also experimented with popular existing RAG techniques. Our framework ultimately won the Simple_w_condition, Set, and Aggregation questions in Task 1.

Submission Number: 2