Beyond calories: evaluating how tailored communication reduces emotional load in diet-coaching

Simone Balloccu, Ehud ReiterACL 2022

TL;DR

This paper introduces a novel approach to diet-coaching apps by augmenting them with personalized, emotionally intelligent communication. By using affective Natural Language Generation (NLG) and persuasive techniques, the authors demonstrate that tailored feedback significantly improves users' emotional state, reducing negative feelings and increasing positive ones, a crucial aspect often overlooked in current diet apps.

What Problem Does This Paper Address?

Dieting is notoriously difficult, and while mobile apps offer a scalable alternative to human coaching, they often fall short. Many apps focus solely on tracking dietary intake and presenting data, neglecting the significant impact of users' emotional states on their eating habits and overall adherence. Traditional diet apps tend to use short, fixed messages, failing to engage users on a deeper, psychological level. This gap leaves a crucial aspect of behavior change unaddressed: how communication can be leveraged not just to inform, but to emotionally support users through the challenges of dieting.

Methods

The core of this research lies in augmenting a standard diet app (MyFitnessPal) with a sophisticated communication layer. This layer utilizes:

  • Affective Natural Language Generation (NLG): This allows the system to generate text that is not only informative but also emotionally attuned, aiming for a positive and supportive tone.
  • Text-Tailoring: Reports are personalized based on user-defined preferences, motivations, and specific dietary goals (e.g., calorie intake, macronutrient balance).
  • Persuasive Communication Techniques: The generated text aims to encourage adherence and positive behavior change.

The system generates weekly "diet reports" that go beyond raw data, offering insights into food analysis, nutrient rankings, and comparisons to previous weeks. These reports are designed to be positive-toned, offering encouragement and framing feedback constructively.

Datasets

The study used data from participants using MyFitnessPal (MFP) for food logging. The primary evaluation metric for emotional state was the PANAS (Positive and Negative Affect Schedule) questionnaire, a validated psychometric tool. Participants completed this weekly to track their emotional well-being.

Key Insights and Ideas

  • Emotional State is Key: The study strongly suggests that emotional well-being is a critical, yet often ignored, factor in the success of diet-coaching. Improving communication can directly impact this.
  • Beyond Data Presentation: Simply presenting charts and numbers isn't enough. Tailored, empathetic language can make a significant difference in user engagement and emotional resilience.
  • The Power of Personalization: Injecting personal elements like nicknames and tailoring feedback to individual motivations makes the communication more impactful.
  • Cognitive Load Matters: The hypothesis that the control group experienced a heavier emotional load due to the cognitive effort of interpreting raw data (charts, numbers) without explanation is compelling. The augmented reports likely reduced this load.

Key Results

  • Dietary Outcomes: The study found mixed results for dietary outcomes. While the report group showed a tendency towards improvement in calories and one nutrient, these differences were not statistically significant. The authors reject Hypothesis 1 (H1) regarding significant dietary improvements.
  • Emotional State (PANAS): This is where the system shone.
  • Positive Affect (PA): The report group showed a statistically significant improvement in PA during the second week (p=0.04).
  • Negative Affect (NA): The report group showed a statistically significant improvement in NA across the entire experiment (p=0.04).
  • Overall, the report group experienced more positive feelings and fewer negative feelings compared to the control group. The control group, conversely, showed a worsening of PA in the second week and a slight worsening of NA.
  • User Feedback: Participants generally provided positive feedback on the reports, appreciating their clarity and tailoring. However, many also requested visual elements like charts, suggesting that a hybrid approach might be most effective.

Limitations and Future Work

  • Dietary Outcome Significance: The lack of statistically significant improvement in dietary outcomes is a limitation. The authors suggest that a longer trial (one month or more) is needed to assess the effectiveness of their communication strategy on diet itself.
  • Ablation Study: The paper did not perform an ablation study to determine which specific elements of the augmented communication (affective NLG, persuasion, tailoring) contributed most to the observed effects. This is a key area for future work.
  • Rigorous Metrics: The authors acknowledge that the definitions of "readability," "accuracy," and "naturalness" in user feedback were not rigorously defined, potentially impacting the reliability of those specific ratings. Future work will focus on more uniform metric definitions.
  • Stress Monitoring: The authors plan to incorporate stress monitoring, as previous work has highlighted its importance in eating habits.
  • Hybrid Approach: Future systems may benefit from combining visual elements (charts) with the advanced textual communication developed in this paper.

Summary updated

BibTeX
@inproceedings{balloccu2022beyond,
  author = {Simone Balloccu and Ehud Reiter},
  title = {{Beyond calories: evaluating how tailored communication reduces emotional load in diet-coaching}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {2022},
}
Figure