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 remains a non-trivial task for misinformation detection. 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 target 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 demonstrate the efficiency and effectiveness of our method, we perform extensive experiments to compare MetaAdapt with state-of-the-art baselines and large language models (LLMs) such as LLaMA, where MetaAdapt achieves better performance in domain adaptive few-shot misinformation detection with substantially reduced parameters on real-world datasets.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that in social media on emerging topics (such as COVID - 19), due to the distribution differences between training data and target - domain data, misinformation detection in the early stage is challenging. Specifically, existing methods have difficulty effectively detecting misinformation in newly emerging topics in the absence of target - domain data and annotations. This is not only because of the data scarcity problem, but also because these models are often not optimized for cross - domain scenarios and cannot fully utilize limited target - domain examples to guide the effective transfer of source - domain knowledge to the target - domain. To solve the above problems, the paper proposes a meta - learning - based method - MetaAdapt for cross - domain few - shot misinformation detection. MetaAdapt achieves this goal in the following ways: 1. **Utilizing Limited Target - Domain Examples**: MetaAdapt can use a small number of target - domain examples to provide feedback and guide the knowledge transfer process from the source - domain to the target - domain. 2. **Adaptive Learning**: By calculating the similarity score between the source task and the meta - task, MetaAdapt can rescale the meta - gradient, thereby adaptively learning from the source task and improving the model's performance in the target - domain even with only a small amount of labeled data. 3. **Optimizing Model Performance**: MetaAdapt aims to maximize the model's performance in the target - domain while reducing the number of parameters and improving the model's efficiency and effectiveness. The paper verifies the effectiveness of MetaAdapt through extensive experiments. The results show that, compared with existing state - of - the - art methods and large - language models, MetaAdapt performs excellently in cross - domain few - shot misinformation detection tasks, especially on real - world datasets.

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

Meta-Learning Adversarial Domain Adaptation Network for Few-Shot Text Classification.

Early Detection of Misinformation for Infodemic Management: A Domain Adaptation Approach

Unsupervised Domain Adaptation Via Contrastive Adversarial Domain Mixup: A Case Study on COVID-19

Learning to Adapt to Evolving Domains.

MetaNER: Named Entity Recognition with Meta-Learning.

Few-Shot Text Ranking with Meta Adapted Synthetic Weak Supervision

Learning to Detect Few-Shot-Few-Clue Misinformation

When MAML Can Adapt Fast and How to Assist When It Cannot

Improve Meta-learning for Few-Shot Text Classification with All You Can Acquire from the Tasks

Meta-Curriculum Learning for Domain Adaptation in Neural Machine Translation

Adaptive Mutual Learning for Unsupervised Domain Adaptation

Learning to Diagnose: Meta-Learning for Efficient Adaptation in Few-Shot AIOps Scenarios

Meta-learning adaptation network for few-shot link prediction in heterogeneous social networks

Meta Navigator: Search for a Good Adaptation Policy for Few-shot Learning

MetaGAD: Meta Representation Adaptation for Few-Shot Graph Anomaly Detection

Meta-Learning Without Data Via Unconditional Diffusion Models

Cross-Domain Few-Shot Classification Via Adversarial Task Augmentation

Domain Adaptive Meta-Learning for Dialogue State Tracking

Sequence Labeling with Meta-Learning

Towards Well-Generalizing Meta-Learning Via Adversarial Task Augmentation