Rumor Detection with Adaptive Data Augmentation and Adversarial Training
Authors: Ying Wang, Fuyuan Ma, Zhaoqi Yang, Yaodi Zhu, Bo Yang, Pengfei Shen, Lei Yun
JAIR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct experiments on three real-world social media rumor detection datasets. Experiments demonstrate that our model has superior rumor detection capabilities compared to baseline methods and shows good robustness against perturbations. |
| Researcher Affiliation | Collaboration | Ying Wang EMAIL Key Laboratory of Symbol Computation and Knowledge Engineering College of Computer Science and Technology, Jilin University Changchun, Jilin, China Pengfei Shen shen EMAIL (Corresponding author) Lei Yun EMAIL (Corresponding author) Key Laboratory Ministry of Industry and Information Technology China Electronic Product Reliability and Environment Testing Research Institute Guangzhou, Guangdong, China |
| Pseudocode | Yes | Algorithm 1 The ADAAT training algorithm |
| Open Source Code | No | The paper does not provide concrete access to source code. It does not contain a specific repository link, an explicit code release statement, or mention code in supplementary materials. |
| Open Datasets | Yes | We select three public datasets to verify the effectiveness of ADAAT. Table 2 shows the statistics of these datasets, which are detailed as follows: Twitter15: The Twitter15 dataset is constructed by Ma et al. (Ma et al., 2017). Twitter16: The Twitter16 dataset is constructed by Ma et al. (Ma et al., 2017). PHEME: The PHEME dataset is constructed by Zubiaga et al. (Zubiaga et al., 2018). |
| Dataset Splits | Yes | In this experiment, the datasets are randomly divided into five parts for cross-validation. The training set comprises 80% of the data, while the test set contains the remaining 20%. |
| Hardware Specification | Yes | To maintain consistency, all experiments were conducted on a single hardware platform equipped with 1TB of RAM and an NVIDIA Ge Force GTX 1080 Ti GPU, which has 11264Mi B of storage. |
| Software Dependencies | No | The paper mentions using a "pre-trained BERT model" but does not specify any software libraries or frameworks with their version numbers. |
| Experiment Setup | Yes | The learning rate is set to 0.001 for all datasets. For the hyperparameters involved in the model, based on practical experience, we set gamma to 0.3 in FGM and beta to 0.01 in final loss. We employ an early stopping mechanism to select the converged model, with the convergence metric set to no better metric results for eight consecutive epochs. |