Recovering Individual Emotional States from Sparse Ratings Using Collaborative Filtering

Area of Science:

• Affective science
• Computational psychology
• Psychometric methods

Background:

• Measuring emotional states precisely is crucial but challenging.
• Existing methods can disrupt the natural emotion generation process.
• High-granularity, temporally precise emotion measurement is needed.

Purpose of the Study:

• Introduce and validate a novel approach for measuring emotional states.
• Utilize collaborative filtering (CF) to recover sparsely sampled data.
• Provide an open-source Python toolbox (Neighbors) for this method.

Main Methods:

• Developed a computational technique using collaborative filtering (CF).
• Sparselly sampled response data and recovered missing values.
• Validated the approach across three diverse experimental datasets.

Main Results:

• Collaborative filtering accurately recovered dense individual ratings from sparse data.
• CF outperformed traditional imputation methods (mean, multivariate).
• Effectiveness was particularly notable in datasets with high individual variability.

Conclusions:

• This CF-based approach enables high-dimensional emotion data acquisition with minimal disruption.
• The Neighbors toolbox offers a practical tool for affective science research.
• The method facilitates new research avenues in understanding emotional experiences.