A Multimodal Myanmar Emotion Dataset for Emotion Recognition.

Area of Science:

• Cognitive Science
• Affective Computing
• Neuroscience

Background:

• Effective emotion recognition is crucial for human interaction and impacts fields like psychology, social sciences, and artificial intelligence.
• Existing emotion datasets often lack cultural diversity, limiting cross-cultural emotion recognition applications.
• Developing culturally specific datasets is essential for advancing inclusive AI and human-computer interaction.

Purpose of the Study:

• To introduce a novel, culturally validated Myanmar emotion dataset for enhancing emotion recognition.
• To evaluate the dataset's effectiveness using unimodal and multimodal classification approaches.
• To demonstrate the dataset's utility in diverse cultural contexts for emotion recognition technology.

Main Methods:

• Collected electroencephalogram (EEG) signals and eye-tracking data from 20 subjects.
• Utilized a novel emotion elicitation paradigm with 15 video clips (5 positive, 5 neutral, 5 negative) per session.
• Validated the dataset using unimodal, traditional multimodal, and deep multimodal (DCCA-AM) classification methods.

Main Results:

• Unimodal classification accuracies ranged from 62.57% to 77.05%.
• Multimodal fusion techniques achieved higher accuracies, ranging from 75.43% to 87.91%.
• The proposed dataset demonstrated significant value for cross-cultural emotion recognition applications.

Conclusions:

• The novel Myanmar emotion dataset effectively supports cross-cultural emotion recognition.
• Multimodal approaches significantly outperform unimodal methods when using this dataset.
• This research highlights the importance of culturally diverse datasets for advancing affective computing and AI.