Debiasing Multimodal Sarcasm Detection with Contrastive Learning

Authors: Mengzhao Jia, Can Xie, Liqiang Jing

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments show the superiority of the proposed framework. We conducted experiments on a publicly available multimodal sarcasm detection benchmark dataset (Cai, Cai, and Wan 2019).
Researcher Affiliation Academia Mengzhao Jia1, Can Xie1, Liqiang Jing2* 1Shandong University, 2University of Texas at Dallas
Pseudocode No The paper does not contain a pseudocode block or a clearly labeled algorithm block.
Open Source Code Yes As a byproduct, we have released the source code and the constructed dataset4. 4https://sharecode.wixsite.com/dmsd-cl.
Open Datasets Yes We conducted experiments on a publicly available multimodal sarcasm detection benchmark dataset (Cai, Cai, and Wan 2019). The dataset comprises 24,635 samples, each comprising textual content, an associated image, and a corresponding label. As a byproduct, we have released the source code and the constructed dataset4. 4https://sharecode.wixsite.com/dmsd-cl.
Dataset Splits Yes Following the original setting, the sizes of the training set, development set, and testing set are 19,815, 2,410, and 2,409, respectively.
Hardware Specification No The paper does not explicitly describe the specific hardware (e.g., GPU models, CPU types) used for running the experiments, only mentioning pre-trained models and APIs like RoBERTa, ViT, and gpt3.5-turbo.
Software Dependencies No The paper mentions using 'RoBERTa-base', 'ViT', and 'gpt3.5-turbo' for various tasks, but does not provide specific version numbers for these software dependencies or other libraries.
Experiment Setup Yes Meanwhile, the hyperparameters K, τ, and λ are configured to 4, 0.07, and 0.9, respectively. We used the Adam optimizer to optimize our model with a learning rate of 1e 5. The mini-batch size is set to 16 and the maximum number of epochs for training is set to 20.