The First Place Solution of WSDM Cup 2024: Leveraging Large Language Models for Conversational Multi-Doc QA

TL;DR

This document serves as a practical guide to using the ACM LaTeX acmart template for preparing publications across journals and conferences. It details the available template styles and parameter options, including mappings like acmsmall, acmlarge, acmtog, acmconf, and SIG variants, to meet different publication formats. It emphasizes metadata and formatting considerations such as anonymous submissions, author versions, screen options, and the incorporation of ACM references and CCS concepts. Overall, it provides implementation-focused guidance to produce consistent, publication-ready manuscripts with proper typography and metadata handling.

Abstract

Conversational multi-doc question answering aims to answer specific questions based on the retrieved documents as well as the contextual conversations. In this paper, we introduce our winning approach for the "Conversational Multi-Doc QA" challenge in WSDM Cup 2024, which exploits the superior natural language understanding and generation capability of Large Language Models (LLMs). We first adapt LLMs to the task, then devise a hybrid training strategy to make the most of in-domain unlabeled data. Moreover, an advanced text embedding model is adopted to filter out potentially irrelevant documents and several approaches are designed and compared for the model ensemble. Equipped with all these techniques, our solution finally ranked 1st place in WSDM Cup 2024, surpassing its rivals to a large extent. The source codes have been released at https://github.com/zhangzhao219/WSDM-Cup-2024.