AlexUNLP-FMT at ClimateCheck Shared Task: Hybrid Retrieval with Adaptive Similarity Graph-based Reranking for Climate-related Social Media Claims Fact Checking

Abstract

In this paper, we describe our work for the Cli-mateCheck shared task at the Scholarly Document Processing (SDP) workshop, ACL 2025.We focus on Subtask 1: Abstracts retrieval.The task involves retrieving relevant abstracts from a large corpus to verify claims made on social media about climate change.We explore various retrieval and reranking techniques, including fine-tuning transformer-based dense retrievers, sparse retrieval methods, and reranking using cross-encoder models.Our final and bestperforming system utilizes a hybrid retrieval approach combining BM25 sparse retrieval with a fine-tuned Stella model for dense retrieval, followed by an MSMARCO-trained MiniLM cross-encoder model for reranking.We adapt an iterative graph-based reranking approach that leverages a document similarity graph built over the document corpus to update the candidate pool for reranking dynamically.Our system achieved a score of 0.415 on the final test set for Subtask 1, securing third place on the final leaderboard.Our code is available on GitHub 1 .