A Novel Energy Based Model Mechanism for Multi-Modal Aspect-Based Sentiment Analysis

Authors: Tianshuo Peng, Zuchao Li, Ping Wang, Lefei Zhang, Hai Zhao

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on three widely used benchmarks demonstrate that DQPSA outperforms previous approaches and achieves a new state-of-the-art performance.
Researcher Affiliation Academia 1School of Computer Science, Wuhan University, Wuhan, 430072, China, 2Hubei Luojia Laboratory, Wuhan 430072, P. R. China, 3Center for the Studies of Information Resources, Wuhan University, Wuhan 430072, China 4School of Information Management, Wuhan University, Wuhan 430072, China 5Department of Computer Science and Engineering, Shanghai Jiao Tong University
Pseudocode No The paper describes the architecture and components of the proposed model, including mathematical formulations for attention and loss functions, but does not include any explicit pseudocode blocks or algorithms.
Open Source Code Yes The code will be released at https://github.com/pengts/DQPSA.
Open Datasets Yes Following previous works, we adopt two widely used benchmarks: Twitter2015 and Twitter2017 (Yu and Jiang 2019) to evaluate our proposed DQPSA. Besides, we employ another Political Twitter dataset1 from (Yang et al. 2021) for JMASA task. In pre-training stage, we use COCO2017 dataset and Image Net dataset.
Dataset Splits No The paper mentions using a 'development set' for model selection but does not provide specific train/validation/test split percentages or sample counts for any of the datasets used (Twitter2015, Twitter2017, Political Twitter, COCO2017, ImageNet).
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments.
Software Dependencies No The paper mentions using specific pre-trained models and architectures like 'CLIP-Vi T-big G-14-laion2B-39B-b160k', 'FSUIE-base', and 'BERT-base architecture', but it does not provide specific version numbers for software dependencies or libraries (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes Hyper-parameters used during training are shown in table 9. We trained model for 50 epochs with an Adam W optimizer on the datasets of each task, and selected the final model based on the performance on the development set.