Can we say a cat is a cat? Understanding the challenges in annotating physiological signal-based emotion data

Pragya Singh,Mohan Kumar,Pushpendra Singh

2024-06-21

Abstract:Artificial Intelligence (AI) algorithms, trained on emotion data extracted from physiological signals, provide a promising approach to monitoring emotions, affect, and mental well-being. However, the field encounters challenges because there is a lack of effective methods for collecting high-quality data in everyday settings that genuinely reflect changes in emotion or affect. This paper presents a position discussion on the current technique of annotating physiological signal-based emotion data. Our discourse underscores the importance of adopting a nuanced understanding of annotation processes, paving the way for a more insightful exploration of the intricate relationship between physiological signals and human emotions.

Human-Computer Interaction

What problem does this paper attempt to address?

The paper primarily explores the challenges in annotating emotion data based on physiological signals and highlights some limitations of current methods. Specifically, the paper attempts to address the following issues:

1. **The challenge of high-quality data collection**: Collecting high-quality physiological signal data that truly reflects emotional changes in everyday environments is very difficult. Although the development of wearable device technology has made continuous collection of physiological signals possible, these devices often do not provide raw data, which increases the difficulty of data collection.

2. **The complexity and bias of the annotation process**: Unlike fields such as computer vision, emotion annotation of physiological signals relies on psychological models, emotion theories, and human annotators' judgments, making the annotation process complex and prone to bias. For example, in computer vision, a photo of a cat can be clearly labeled as "cat," but in the field of emotion recognition, such straightforward labeling methods are not applicable.

3. **The inadequacies of existing annotation methods**: Currently used annotation methods (such as self-report, expert annotation, and physiological correlation annotation) all have certain limitations. For instance, self-reports may be influenced by the individual's mood, while expert annotation is difficult to implement in real-life or mobile environments. Additionally, annotation methods based on external stimuli cannot fully capture the subjective perception differences of individuals in different contexts.

In summary, this paper aims to reveal the shortcomings of current physiological signal emotion data annotation methods through discussion and calls for researchers to develop more refined and comprehensive annotation strategies to improve the reliability and accuracy of emotion recognition algorithms.