TL;DR
This paper introduces novel computational measures to quantify habitual food consumption from digital food diaries. By analyzing recurrence patterns in food choices, the authors reveal nuanced insights into individual eating habits, offering a new lens for personalized dietary interventions.
What Problem Does This Paper Address?
Understanding habitual food consumption is crucial for effective dietary interventions, yet it's historically been difficult to quantify. Traditional methods like food frequency questionnaires suffer from recall bias, and detailed dietary logs are burdensome. While smartphone apps have democratized food logging, extracting meaningful insights into habitual behavior from this data remains a challenge. This paper aims to bridge this gap by developing computational methods to analyze longitudinal food diary data and characterize habitual eating patterns.
Methods
The core of this work lies in developing "computational recurrence measures" to quantify habitual food consumption. The authors define two key characteristics:
- Relative Frequency: This measures how often a specific food item or meal is consumed within a given observation period. It's essentially a normalized count of occurrences.
- Span: This quantifies the temporal spread of a food item or meal's occurrences, measuring the time between its first and last appearance in the log. A larger span suggests a more consistent, long-term habit.
Using these, they define "recurrent choices" as food items or meals that meet specific thresholds for both relative frequency (at least once per week) and span (occurring over at least half the observation period). They then introduce three "recurrence strength" metrics:
- Food-item recurrence strength ($\alpha_{u,o}$): The proportion of days an individual consumes at least one recurrent food item during a specific meal occasion.
- Food-item-per-meal recurrence strength ($\beta_{u,o}$): The average proportion of recurrent food items within a meal.
- Meal recurrence strength ($\gamma_{u,o}$): The proportion of days an individual consumes a recurrent meal.
These three metrics form the "recurrence strength tuple," providing a multi-faceted view of habitual eating. The authors also establish a minimum data duration (28 days) for reliable estimation of these measures.
Datasets
The primary dataset used is the MyFitnessPal Food Diary Dataset. This public dataset contains 587,187 days of food diary entries from 9896 individuals over six months. Users logged their food intake, including meal occasion and food item names. The authors preprocessed this data to remove irrelevant information and focus on the food item names for their analysis.
Key Insights and Ideas
- Food-item habits are stronger than meal habits: Individuals tend to repeat specific food items more consistently than entire meals. This suggests that targeting individual food items might be a more effective strategy for dietary change than focusing on complete meals.
- Breakfast is the most habitual meal: Recurrence strength is highest for breakfast across all metrics, indicating that people have the most established routines for their morning meals. Dinner, conversely, shows the least habitual behavior.
- Weekdays are more habitual than weekends: Food consumption is generally more routine during weekdays, with a noticeable drop in recurrence on weekends, suggesting a shift towards more exploratory eating patterns.
- Habitual weekday eaters are also habitual weekend eaters: Despite the overall decrease in weekend recurrence, individuals who are highly habitual during the week tend to remain so on weekends, indicating a strong underlying habit-forming tendency.
- More items per meal, fewer habitual meals: A higher number of food items in a meal correlates with stronger food-item recurrence but weaker meal recurrence. This implies that while individuals might repeat certain items, they combine them in varied ways, leading to less consistent meal patterns.
- Food logging duration doesn't significantly impact habit measurement: The study found no significant correlation between the duration of food logging and the measured recurrence strength, suggesting that the developed metrics are robust to varying engagement levels.
Key Results
- Minimum data requirement: Approximately 28 days of food diary data are sufficient to estimate recurrent consumption behavior with an average estimation error of less than 0.1. For breakfast alone, 14 days may suffice.
- Recurrence strength distribution: 93% of individuals exhibit non-zero food-item recurrence strength for breakfast, while only a small fraction (27% for breakfast, much lower for other meals) consume recurrent meals regularly.
- Meal occasion differences: Food-item recurrence strength is highest for breakfast (0.65), followed by snacks (0.36), lunch (0.28), and dinner (0.24). Meal recurrence strength follows a similar pattern, with breakfast (0.11) being significantly higher than lunch (0.02), dinner (0.01), and snacks (0.02).
- Weekday vs. Weekend: Food-item-per-meal recurrence strength is significantly higher on weekdays than weekends across all meal occasions, with the largest difference observed for breakfast.
- Correlations: High correlations were observed between food-item recurrence strength and food-item-per-meal recurrence strength, and moderate correlations between recurrence strength across different meal occasions.
Limitations and Future Work
- Demographic Skew: The MyFitnessPal dataset is predominantly female and within the 18-44 age group, limiting the generalizability of findings to other demographics. Future work should involve datasets from diverse populations.
- Self-Reported Data: The data is self-reported and may contain inaccuracies or omissions. While extensive preprocessing was done, this remains a potential limitation.
- Lack of Context: The study did not have access to information about the reasons behind high or low recurrent consumption (e.g., food insecurity, personality, workplace limitations). Future research could explore these drivers.
- Individual Food-Item Independence: The analysis modeled food items independently. Future work could investigate how individuals group or pair food items and repeat these combinations.
- Privacy Concerns: The authors acknowledge that user data privacy is a significant issue in diet-tracking apps.
Summary updated