SKIER: A Symbolic Knowledge Integrated Model for Conversational Emotion Recognition
Authors: Wei Li, Luyao Zhu, Rui Mao, Erik Cambria
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on three benchmarks demonstrate that both dialogue structure graphs and symbolic knowledge are beneficial to the model performance on the task. Additionally, experimental results indicate that the proposed model surpasses baseline models on several indices. |
| Researcher Affiliation | Academia | Wei Li, Luyao Zhu, Rui Mao, Erik Cambria School of Computer Science and Engineering, Nanyang Technological University, Singapore. {wei008, luyao001}@e.ntu.edu.sg, {rui.mao, cambria}@ntu.edu.sg |
| Pseudocode | No | The paper describes methods using mathematical equations and textual descriptions but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | 1https://github.com/senticnet/SKIER |
| Open Datasets | Yes | We test our model on Daily Dialog (Li et al. 2017), Emory (Zahiri and Choi 2018), and MELD (Poria et al. 2019). |
| Dataset Splits | Yes | Dataset Train Dev Test Label Metrics MELD u 9989 1109 2610 7/3 Weighted Avg F1 d 1038 114 280 Emory NLP u 9934 1344 1328 7/3 Weighted Avg F1 d 713 99 85 Daily Dialog u 87170 8069 7740 7(6) Macro & Micro F1 d 11118 1000 1000 |
| Hardware Specification | Yes | All experiments were conducted on a V100 GPU with 16 GB memory. |
| Software Dependencies | No | The paper mentions 'Ro BERTa-Large from Hugging Face' and 'Adam W' but does not provide specific version numbers for software libraries or programming languages used. |
| Experiment Setup | Yes | The optimizer was Adam W (Loshchilov and Hutter 2018) with an initial learning rate of 1e-5. We used a linear scheduler during training. The maximum value of 5 was used for the gradient clipping. The batch size was 1. The dropout rate was 0.2. λk in Eq. 6 was 0.5. The number of destination nodes was 3. |