Seed Emotion Recognition Eval

seed-emotion-recognition-eval

Deep Recurrent Semi-Supervised EEG Representation Learning for Emotion Recognition — Zhang et al. (2021) (arXiv:2107.13505, 2021)

What this evaluates

Evaluates a model's ability to classify EEG signals into three affective states (negative, neutral, positive) using a semi-supervised learning framework. It tests representation learning and classification performance on high-dimensional, noisy time-series data with limited labeled sessions.

Datasets

• SEED — total ?; splits: train (9), test (6)

Metrics

• accuracy (primary) — range: [0, 1]
  • Standard classification accuracy: the proportion of correctly predicted emotion labels (negative, neutral, positive) out of the total number of test instances.

Input / output format

Input: EEG time-series recordings from 62 scalp electrodes sampled at 1 KHz, structured into 15 sessions per experiment (5s pre-stimuli notice + visual stimuli).

Output: Predicted emotion label from the set {negative, neutral, positive}.

Scoring recipe

def compute_accuracy(predictions, gold_labels):
    correct = sum(1 for p, g in zip(predictions, gold_labels) if p == g)
    return correct / len(gold_labels)

Common pitfalls

• The dataset is split per subject/experiment (9 sessions train, 6 test), not randomly shuffled across all sessions.
• The evaluation is semi-supervised, meaning the model leverages unlabeled sessions during training alongside a small fraction of labeled data.

Evidence (verbatim from paper)

We use the SEED dataset to conduct emotion recognition experiments with three affective labels namely negative, neutral, and positive. ... In each of the 30 experiments, we use the first 9 sessions for training and the remaining 6 sessions for testing, as pre-defined in [[10]].

Citation

@misc{zhang2021deeprecurrent,
  title={Deep Recurrent Semi-Supervised EEG Representation Learning for Emotion Recognition},
  author={Zhang et al. (2021)},
  year={2021},
  note={arXiv:2107.13505}
}

• arXiv: 2107.13505