[Paper Review] Aligning Daily Activities with Personality: Towards A Recommender System for Improving Wellbeing

Mohammed Khwaja, Miquel Ferrer+3 more|arXiv (Cornell University)|Sep 9, 2019

Mental Health Research Topics4 citations

TL;DR

This paper proposes a personalized recommender system that aligns daily activities with users' Big-Five personality traits to improve subjective wellbeing (SWB). Using a machine learning model trained on smartphone-collected activity, personality, and SWB data, it predicts optimal activity distributions for high or low SWB, achieving up to 92% accuracy in identifying 'bad' activity patterns for individuals.

ABSTRACT

Recommender Systems have not been explored to a great extent for improving health and subjective wellbeing. Recent advances in mobile technologies and user modelling present the opportunity for delivering such systems, however the key issue is understanding the drivers of subjective wellbeing at an individual level. In this paper we propose a novel approach for deriving personalized activity recommendations to improve subjective wellbeing by maximizing the congruence between activities and personality traits. To evaluate the model, we leveraged a rich dataset collected in a smartphone study, which contains three weeks of daily activity probes, the Big-Five personality questionnaire and subjective wellbeing surveys. We show that the model correctly infers a range of activities that are 'good' or 'bad' (i.e. that are positively or negatively related to subjective wellbeing) for a given user and that the derived recommendations greatly match outcomes in the real-world.

Motivation & Objective

• To address the cold-start problem in wellbeing-focused recommender systems by leveraging personality traits as user profiles.
• To model the congruence between personality and daily activity distributions as a driver of subjective wellbeing (SWB).
• To develop a personalized activity recommendation system that predicts optimal activity patterns to enhance SWB.
• To validate the model’s ability to infer accurate 'white- and black-listed' activity distributions for high and low SWB outcomes.
• To evaluate the system’s performance under both ideal and realistic user behavior scenarios.

Proposed method

• Collected a longitudinal dataset via a smartphone app over 2–3 weeks, capturing five daily Ecological Momentary Assessment (EMA) probes for activity reporting.
• Integrated the Big-Five personality questionnaire and a life satisfaction survey as ground-truth SWB measures.
• Used a machine learning classifier to predict SWB based on the alignment between personality traits and activity distribution patterns.
• Defined optimal activity ranges for high and low SWB by simulating distributions that match predicted SWB outcomes.
• Evaluated model performance by comparing predicted optimal ranges against actual user activity distributions in a second dataset.
• Applied a binary classification model with a regularization parameter (λ=0.1) and identified 8 key activity clusters with significant variance.

Experimental results

Research questions

• RQ1 Can personality traits be used as a proxy to infer personalized activity patterns that enhance subjective wellbeing?
• RQ2 To what extent can a model predict the range of activity distributions that lead to high or low SWB for individual users?
• RQ3 How accurate is the model in identifying 'good' and 'bad' activity patterns when evaluated against real-world user behavior?
• RQ4 Does the model perform better under ideal (full compliance) or realistic (majority compliance) user behavior assumptions?
• RQ5 Can the model outperform baseline classifiers that use only personality and activity data without considering congruence?

Key findings

• The proposed model outperformed three benchmark classifiers by 9–18% in predicting SWB based on personality-activity congruence.
• When considering all eight key activities, 51% of users in the high SWB class and 74% in the low SWB class had activity distributions within the predicted optimal ranges.
• When evaluating the majority of activities (at least five), prediction accuracy improved to 71% for high SWB users and 92% for low SWB users.
• The model demonstrated stronger performance in identifying 'bad' activity patterns (low SWB) than 'good' ones, suggesting higher sensitivity to negative behavioral impacts.
• The results indicate that activity distribution congruence with personality is a strong predictor of SWB, supporting the feasibility of personalized wellbeing recommender systems.
• The system’s high accuracy under realistic compliance conditions suggests practical viability for real-world wellbeing interventions.