Real-Time Emotion Recognition via Attention Gated Hierarchical Memory Network
Authors: Wenxiang Jiao, Michael Lyu, Irwin King8002-8009
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct experiments on two emotion conversation datasets with extensive analysis, demonstrating the efficacy of our AGHMN models. In this section, we will present the details of our experimental setup, including datasets, compared methods, implementation, and training. |
| Researcher Affiliation | Academia | 1Department of Computer Science and Engineering, The Chinese University of Hong Kong, Shatin, N.T., Hong Kong, China {wxjiao, lyu, king}@cse.cuhk.edu.hk |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | We implement sc LSTM and our proposed AGHMN models from scratch on the Pytorch5 framework. ... 4https://github.com/wxjiao/AGHMN |
| Open Datasets | Yes | Datasets. We train and test our model on two conversation emotion datasets, namely, IEMOCAP (Busso et al. 2008), and MELD (Poria et al. 2019a). IEMOCAP2: The IEMOCAP dataset contains the acts of 10 speakers in a dyadic conversation fashion, providing text, audio, and video features. We follow the previous work (Hazarika et al. 2018a) to use the first four sessions of transcripts as the training set, and the last one as the testing set. ... 2https://sail.usc.edu/iemocap/ MELD3: The MELD dataset (Poria et al. 2019a) is an extended version of the Emotion Lines dataset (Hsu et al. 2018). ... 3https://github.com/Sentic Net/MELD |
| Dataset Splits | Yes | IEMOCAP: ...The validation set is extracted from the randomly-shuffled training set with the ratio of 80:20. MELD: The data comes from the Friends TV series with multiple speakers involved in the conversations. It is split into training, validation, and testing sets with 1039, 114, and 280 conversations, respectively. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | Yes | We implement sc LSTM and our proposed AGHMN models from scratch on the Pytorch5 framework. ... We choose Adam (Kingma and Ba 2015) optimizer... |
| Experiment Setup | Yes | As for our AGHMN models, the hidden sizes of GRUs and AGRUs are also 100. By default, the context-window size K for the memory bank is 40 for IEMOCAP and 10 for MELD... We choose Adam (Kingma and Ba 2015) optimizer with an initial learning rate lr = 5 × 10−4. To regulate the models, we clip the gradients of model parameters with a max norm of 5 and apply dropout with a drop rate of 0.3. ...decay the learning rate by 0.95 once the m F1 stops increasing. The training process is terminated by early stopping with a patience of 10. |