MetaAdapt: Domain Adaptive Few-Shot Misinformation Detection via Meta Learning

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 5223–5239 July 9-14, 2023 ©2023 Association for Computational Linguistics MetaAdapt: Domain Adaptive Few-Shot Misinformation Detection via Meta Learning Zhenrui Yue Huimin Zeng Yang Zhang Lanyu Shang Dong Wang University of Illinois Urbana-Champaign {zhenrui3, huiminz3, yzhangnd, lshang3, dwang24}@illinois.edu Abstract With emerging topics (e.g., COVID-19) on social media as a source for the spreading misinformation, overcoming the distributional shifts between the original training domain (i.e., source domain) and such target domains re- mains a non-trivial task for misinformation de- tection. This presents an elusive challenge for early-stage misinformation detection, where a good amount of data and annotations from the target domain is not available for training. To address the data scarcity issue, we propose MetaAdapt, a meta learning based approach for domain adaptive few-shot misinformation detection. MetaAdapt leverages limited target examples to provide feedback and guide the knowledge transfer from the source to the tar- get domain (i.e., learn to adapt). In particular, we train the initial model with multiple source tasks and compute their similarity scores to the meta task. Based on the similarity scores, we rescale the meta gradients to adaptively learn from the source tasks. As such, MetaAdapt can learn how to adapt the misinformation detection model and exploit the source data for improved performance in the target domain. To demon- strate the efficiency and effectiveness of our method, we perform extensive experiments to compare MetaAdapt with state-of-the-art base- lines and large language models (LLMs) such as LLaMA, where MetaAdapt achieves bet- ter performance in domain adaptive few-shot misinformation detection with substantially re- duced parameters on real-world datasets. 1 Introduction Recently, significant progress has been made in misinformation detection due to the improve- ments in developing machine learning-based meth- ods (Wu et al., 2019; Shu et al., 2020c; Wu et al., 2022b). Such methods include large language mod- els (LLMs), which can be fine-tuned for detecting and responding to rumors on social media plat- forms (Jiang et al., 2022; He et al., 2023; Touvron Weights Source Text (e.g. Bleach, disinfectant, or similar products can prevent the spread of bacteria.) Source Prediction → True (p=0.92) Source-Trained Detection Model Deployed Detection Model Target Text (e.g. Coronavirus can be treated by injecting disinfectant into the body.) Target Prediction → True (p=0.73) Figure 1: Existing models (from source domain) fail to detect rumors on emerging topics (target domain). et al., 2023). However, misinformation on emerg- ing topics remains an elusive challenge for existing approaches, as there exists a large domain gap be- tween the training (i.e., source domain) and the target distribution (i.e., target domain) (Yue et al., 2022). For instance, existing models often fail to detect early-stage misinformation due to the lack of domain knowledge (see Figure 1). With the increase of emerging topics (e.g., COVID-19) on social media as a source of mis- information, the failure to distinguish such early- stage misinformation can result in potential threats to public interest (Roozenbeek et al., 2020; …

Content truncated to 500 words