Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Beyond Single Emotion: Multi-label Approach to Conversational Emotion Recognition
Authors: Yujin Kang, Yoon-Sik Cho
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The empirical results on existing task with single label support the efficacy of our approach, which is more effective in the most challenging settings: emotion shift or confusing labels. We also evaluate ML-ERC with the multi-labels we produced to support our contrastive learning scheme. We conduct extensive experiments to verify the effectiveness of our proposed model. We integrate our multi-label scheme into existing single label ERC models, and show how our objective improves all of the existing baseline models. |
| Researcher Affiliation | Academia | Yujin Kang, Yoon-Sik Cho Department of Artificial Intelligence, Chung-Ang University, Republic of Korea EMAIL |
| Pseudocode | Yes | Algorithm 1: Learning procedure of ML-ERC for each batch B at each epoch. Once the model runs several iterations, we conduct soft-labeling. |
| Open Source Code | No | The paper does not provide an explicit statement or link to the source code for the methodology described. It only mentions that baseline results were reproduced using 'original code' or 'respective official codebase', referring to other works. |
| Open Datasets | Yes | We conduct experiments on three benchmark ERC datasets annotated with single labels. Emory NLP (Zahiri and Choi 2018) is labeled with joyful, mad, neutral, peaceful, powerful, scared, and sad from the Feeling wheel (Willcox 1982). MELD (Poria et al. 2019) is a multi-modal dataset with a label set that includes anger, disgust, fear, joy, neutral, surprise, and sadness. IEMOCAP (Busso et al. 2008) is a dyadic multimodal dataset with labels including excited, neutral, frustrated, sadness, happiness, and anger. |
| Dataset Splits | No | The paper mentions that 'The statistics for each dataset are provided in Table S4 in Appendix F.' which is not in the main text. It does not explicitly state the training, validation, or test splits (e.g., percentages or counts) for the datasets used in the main body of the paper. |
| Hardware Specification | Yes | All experiments are performed on an Nvidia RTX A6000 GPU. |
| Software Dependencies | No | The paper mentions using 'Ro BERTa-Large' as an embedding module but does not provide specific version numbers for any software libraries, frameworks, or programming languages used in the implementation. |
| Experiment Setup | Yes | To train our ML-ERC method, we set the learning rate, the number of batch sizes and epochs are 1e6, 16 and 30, respectively. We fix τ in Eq 8, 9 to 0.05. For α in Eq 12, we search the parameter using the validation set. We set α to 0.7 for Emorynlp, 0.1 for MELD, and 0.4 for IEMOCAP. The hyperparameter β, which is set to 0.5, controls the integration of LML-ERC with the original loss from the ERC model (LERC). |