Text-Guided Fine-grained Counterfactual Inference for Short Video Fake News Detection
Authors: Linlin Zong, Wenmin Lin, Jiahui Zhou, Xinyue Liu, Xianchao Zhang, Bo Xu, Shimin Wu
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on publicly available fake short video news datasets demonstrate that our method outperforms state-of-the-art techniques. We experiment on the Fake SV dataset (Qi et al. 2023a), the only benchmark for short video fake news detection. Table 1: Comparative experiments on datasets partitioned by time and events. We conducted five ablation experiments to assess the impact of each component on the model’s performance, as shown in Table 2. Counterfactual inference strategy study. Loss constraint strategies study. Hyperparameter Study |
| Researcher Affiliation | Academia | 1Key Laboratory for Ubiquitous Network and Service Software of Liaoning Province, School of Software, Dalian University of Technology, Dalian, China 2School of Computer Science and Technology, Dalian University of Technology, Dalian, China EMAIL, EMAIL |
| Pseudocode | No | The paper describes methods using mathematical equations and structured steps in paragraph form, but there are no explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any explicit statements about releasing source code, nor does it provide a link to a code repository. |
| Open Datasets | Yes | We experiment on the Fake SV dataset (Qi et al. 2023a), the only benchmark for short video fake news detection. Fake SV includes rich content like videos, audio, comments, titles, and media information. |
| Dataset Splits | Yes | Following the original study, the dataset is segmented in two ways: by event and by time. Table 1: Comparative experiments on datasets partitioned by time and events. |
| Hardware Specification | Yes | The experiments were conducted on an RTX 3090 Ti GPU using the Py Torch framework, with Python version 3.9.18. |
| Software Dependencies | Yes | The experiments were conducted on an RTX 3090 Ti GPU using the Py Torch framework, with Python version 3.9.18. |
| Experiment Setup | Yes | The initial learning rate (lr) was set to 0.0004 for the teacher model, 0.0005 for the video model, and 0.01 for the audio model, with a batch size of 64. The model parameters were optimized using the Adam optimizer (Kingma and Ba 2014). |