Unsupervised Cross-Domain Rumor Detection with Contrastive Learning and Cross-Attention

Authors: Hongyan Ran, Caiyan Jia

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct experiments on four groups of cross-domain datasets and show that our proposed model achieves state-of-the-art performance.
Researcher Affiliation Academia Hongyan Ran, Caiyan Jia* School of Computer and Information Technology & Beijing Key Lab of Traffic Data Analysis and Mining Beijing Jiaotong University, Beijing 100044, China {hongyran,cyjia}@bjtu.edu.cn
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper links to 'Supplementary Material' (https://github.com/rhy1111/Supplementary Material) but does not explicitly state that the source code for the methodology is provided at this link or elsewhere.
Open Datasets Yes We evaluate the UCD-RD model on four groups of real-world cross-domain rumor datasets. The first group of data comes from PHEME (Zubiaga et al. 2015) dataset... The second group of cross-domain data is Twitter dataset (Ma, Gao, and Wong 2017) and Twitter-Covid19 dataset (Lin et al. 2022). The third group of datasets includes the Twitter15 dataset and the Twitter16 dataset (Ma, Gao, and Wong 2018). The fourth group of cross-domain data is the Chinese Weibo dataset (Ma et al. 2016) and the Weibo-Covid19 dataset (Lin et al. 2022).
Dataset Splits No The paper mentions training and testing but does not provide specific details on a validation dataset split (percentages, counts, or explicit use of a 'validation set'). The target data is used for pseudo-labeling, not explicit validation with ground truth.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory, or specific cloud instances) used for running the experiments, only general software frameworks are mentioned.
Software Dependencies No The paper mentions 'Keras3' and 'Pytorch4' which refer to frameworks but do not provide specific version numbers required for reproducible software dependencies.
Experiment Setup Yes The training process is iterated upon 300 epochs. The temperature τ is 0.1. For instance, for the Terrorist Gossip data, when these hyper parameters α1 = 0.9, α2 = 0.1, β1 = 0.7, β2 = 0.3, and γ1 = 0.8, γ2 = 0.1, γ3 = 0.1, UCD-RD achieves the best performance.