[Springer] Deciphering the "Silent" Personality: Predicting Big-5 Traits from Daily Household Activities

TL;DR

Can the way you search for your keys or make a cup of tea reveal your personality? This study demonstrates that it can. By analyzing 3D skeletal motion and sensor data in a non-social "smart home" setting, researchers developed a framework that predicts Big-5 personality traits (Extraversion, Agreeableness, etc.) with up to 80% accuracy, proving that our personality is "written" into our most mundane movements.

The "Unconstrained" Challenge: Beyond Social Cues

Most personality computing research relies on social interaction—how you talk to a recruiter or post on social media. But what happens when you are alone?

The authors argue that social settings introduce expressivity biases; people act the way they think they should act. By shifting the focus to unconstrained, non-social scenarios (like a smart home), this research uncovers the "substrate" of personality. The core challenge is that daily activities (ADLs) are noisy and varied, requiring a bridge between Computer Vision (motion tracking) and Psychology (behavioral markers).

Methodology: Engineering the Digital Personality

The study utilizes a dataset of participants performing six specific tasks, ranging from the mundane (making tea) to the stressful (searching for an object that doesn't exist).

The Feature Engine

The framework extracts four primary types of behavioral features:

1. Posture: Euclidean distances between 11 upper-body joints while "standing still."
2. Movement Amplitude: The "stretch" of movements relative to a stable spine joint—predictive of Extraversion.
3. Pattern Regularity: Measured via magnetic sensors on drawers to track temporal consistency.
4. Movement Speed: Calculated via head displacement in 3D space, a known marker for Neuroticism.

Figure 1: The smart environment where subjects performed tasks like searching for keys or preparing tea.

Results: Which Tasks Reveal the Most?

The study used Support Vector Machines (SVM), Random Forests (RF), and Gaussian Process Classifiers (GPC). The GPC model was the clear winner, likely due to its ability to handle the non-linearities and the probabilistic nature of the small sample size.

Key Findings:

• Extraversion: Best predicted (80% accuracy) during "Task 3," where subjects were frustrated by searching for a non-existent item.
• Neuroticism: Strongly correlated with movement speed (Pearson’s R = 0.45-0.51) during stress-inducing or effort-requiring tasks.
• Conscientiousness: Most visible in Task 1 (finding keys), characterized by calm, clear posture patterns.

Figure 2: Performance comparison of SVM, GPC, and RF across different tasks.

Critical Insight: Task-Dependent Personality Display

The most profound takeaway is that personality is not displayed uniformly. Different traits "leak" out during different activities. High-pressure tasks (problem-solving) are excellent for identifying Extraversion and Neuroticism, while relaxed, low-stakes tasks (making tea) are better for observing Agreeableness and Openness.

Limitations and Future Outlook

While the results are promising, the study is limited by its sample size (46 individuals) and the reliance on upper-body joints due to camera occlusions. Future research could integrate Dempster-Shafer fusion to combine results from multiple tasks, potentially creating a "cumulative personality profile" that becomes more accurate as the smart home observes the user over weeks rather than hours.

Conclusion

This work bridges the gap between raw skeletal data and high-level psychological constructs. In the future, your smart home might not just adjust the lights—it might recognize when your movement patterns suggest rising Neuroticism or stress and suggest a break, all without you saying a word.