Inferring Mood-While-Eating with Smartphone Sensing and Community-Based Model Personalization

TL;DR

This study tackles the problem of inferring mood during eating episodes from smartphone sensing, addressing poor generalization of generic mood models to eating contexts and data-scarcity in personalization. It analyzes two datasets (MEX and MUL) and demonstrates that population-level and partially personalized models underperform for mood-while-eating; a novel community-based model personalization (CBM) is proposed to leverage similarities among users. CBM yields higher accuracies (63.8% for MEX and 88.3% for MUL) by forming target-user communities and training on both the target and similar users’ data, effectively mitigating data scarcity and cold-start challenges. The results highlight the importance of context-specific mood inference and offer practical implications for mobile food diaries and real-time interventions, while acknowledging limitations and outlining directions for broader, domain-adaptive personalization in multimodal mobile sensing.

Abstract

The interplay between mood and eating episodes has been extensively researched, revealing a connection between the two. Previous studies have relied on questionnaires and mobile phone self-reports to investigate the relationship between mood and eating. However, current literature exhibits several limitations: a lack of investigation into the generalization of mood inference models trained with data from various everyday life situations to specific contexts like eating; an absence of studies using sensor data to explore the intersection of mood and eating; and inadequate examination of model personalization techniques within limited label settings, a common challenge in mood inference (i.e., far fewer negative mood reports compared to positive or neutral reports). In this study, we sought to examine everyday eating behavior and mood using two datasets of college students in Mexico (N_mex = 84, 1843 mood-while-eating reports) and eight countries (N_mul = 678, 24K mood-while-eating reports), which contain both passive smartphone sensing and self-report data. Our results indicate that generic mood inference models experience a decline in performance in specific contexts, such as during eating, highlighting the issue of sub-context shifts in mobile sensing. Moreover, we discovered that population-level (non-personalized) and hybrid (partially personalized) modeling techniques fall short in the commonly used three-class mood inference task (positive, neutral, negative). To overcome these limitations, we implemented a novel community-based personalization approach. Our findings demonstrate that mood-while-eating can be inferred with accuracies 63.8% (with F1-score of 62.5) for the MEX dataset and 88.3% (with F1-score of 85.7) with the MUL dataset using community-based models, surpassing those achieved with traditional methods.

Figures (12)

• Figure 1: Original and Three-Class Mood Distributions of datasets
• Figure 2: (a) This shows the context-specific accuracies of the generic mood inference model trained for MUL for approaches PLM and HM. (b) This shows the overall mood inference accuracy (All) and accuracy for eating episodes (mood-while-eating) when the number of mood-while-eating reports in the training set changes, with HMs for MUL.
• Figure 3: MEX dataset: Distribution change in no. of users with the increase of no. of data points in the negative class.
• Figure 4: High-Level Architectural View of the Community-based Personalization Approach
• Figure 5: Mean accuracy values (column values in the graphs) [with standard deviations] calculated using the random forest for CBM for multiple threshold values [No. of Users] and averaged community sizes (line values in the graphs)