Source-linked AI summary
Supporting The Many Lives of Personal Data with Rebite: LLM-Powered Goal-Directed Framing in Food Journaling
Weijun Li, Daniel A. Epstein
TL;DR
Personal informatics systems often struggle when evolving goals make past data’s interpretation obsolete. This paper proposes LLM-powered goal-directed framing in Rebite, which translates goals into metrics and reframes stored meal photos; a one-week study with 21 participants found that reframing clarified priorities and exposed cross-goal overlaps, conflicts, and trade-offs.
Problem
Personal informatics systems often bind data to fixed interpretations, while changing goals require past records to be understood through new metrics and priorities.
Method
The paper introduces goal-directed framing and implements it in Rebite, using LLMs to translate goals into metrics and reinterpret stored meal photos when goals change.
Results
In a one-week deployment with 21 participants managing multiple dietary goals, mapping goals to metrics clarified priorities and reframing past meals exposed overlaps, conflicts, and trade-offs.
Takeaways & Limitations
Goal-directed framing can keep personal data meaningful across changing goals while making users negotiate and refine what those goals mean in practice.
Takeaways & Limitations
The one-week study captured a scheduled first reframing rather than repeated self-initiated changes, and participants’ goals clustered around common dietary themes.
Abstract
from arXiv · showhide
People's health and tracking goals frequently change, but most personal informatics systems struggle to adapt, leading people to abandon their data and start over. We propose goal-directed framing, an approach that repositions goals within personal informatics systems. Instead of fixing the meaning of data at capture time, the approach frames the collected data through the current goal and reframes it whenever the goal changes. We realize this in Rebite, a photo-based food journaling system that uses LLMs to read unstructured meal photos and produce goal-directed feedback. In a one-week deployment with 21 participants managing multiple dietary goals, we find that goal-directed framing shaped how participants engaged with their goals. Translating a goal into metrics helped them see what it meant in practice, confirming existing priorities, surfacing what they overlooked, and revealing where the metrics fell short. When goals changed, seeing past meals reframed under the new goal exposed overlaps and conflicts, prompting participants to negotiate trade-offs and refine priorities. We discuss how goal-directed framing both supports and complicates reflection as goals change, and offer design implications for personal informatics systems to support evolving goals.
1 Introduction
Personal informatics systems often lock data to fixed interpretations even as people’s goals evolve. The paper proposes goal-directed framing, implemented in Rebite, to translate current goals into metrics and reframe past data without re-logging.
- Evolving goals can make existing system metrics no longer relevant, causing people to suspend tracking or switch tools and leave past data behind.
- Most systems either hard-code metrics or allow configuration only for future entries, keeping collected data tied to its original meaning.
- Goal-directed framing translates a stated goal into relevant metrics and interprets collected data through that goal’s lens.
- When goals change, the system reinterprets all past data under the new goal so the same records acquire new meaning without re-logging.
- In a one-week deployment with 21 participants, goal-to-metric mapping clarified priorities and limitations, while reframing exposed overlaps and conflicts that prompted trade-off negotiation.
2 Background and Related Work
Prior work establishes that goals shape personal informatics, but goals and interpretations evolve while systems commonly bind meaning prospectively or at design time. LLMs offer flexible interpretation and personalization, motivating goal-directed framing in food journaling.
- Goals guide what people record and how they reflect, while routines, health needs, and priorities cause goals to be revised, expanded, or abandoned.
- Existing systems struggle with changing goals because they either hard-code a single interpretive frame or apply new configurations only to future data.
- Goal-directed framing adds the missing capability of reapplying a goal to data already collected, keeping past data relevant across goal changes.
- LLMs can generate personalized narratives and goal-oriented feedback, but prior work found users often must repeatedly refine prompts to obtain goal-relevant health answers.
- Food journaling spans nutritional, qualitative, and social purposes, and different conditions require distinct tracking items and representations.
- The interpretation pipeline operationalizes framing by configuring metrics from a selected goal, assessing meal photos, anchoring results, and re-executing on existing logs when goals change.
3 Goal-Directed Framing: An Approach for Personal Informatics Systems with Evolving Goals
Goal-directed framing positions the system to translate evolving intentions into feedback while preserving the underlying data for reinterpretation. The approach uses LLMs to map unstructured inputs and goals into structured interpretations on demand.
- The approach shifts interpretation from fixed system metrics toward the user’s evolving intention.
- The system must translate vague or specific goals into trackable configurations that users can refine as their intent sharpens.
- When goals change, the system reinterprets stored data while keeping the underlying facts stable, distinguishing new readings from changed records.
- Users should be able to provide lightweight, unstructured input while the system extracts goal-directed structure and connects goals to assessments and evidence.
- LLMs address these challenges by mapping goals to interpretations and reinterpreting unstructured logged inputs on demand.
- The approach may extend to exercise, sleep, and mental-health data where preserved raw streams can support reinterpretation as goals change.
4 The Rebite System
Rebite realizes goal-directed framing as a photo-based food journal that converts everyday goals into configurable metrics and layered feedback. Its pipeline preserves meal records while regenerating goal-specific interpretations whenever goals change.
- Rebite lets users state goals in everyday language, translates them into metrics, reinterprets stored meals after goal changes, and presents four-layer feedback.
- Users select a goal, optionally provide personal context, and receive an AI-recommended metric configuration that they can inspect and adjust.
- The goal–metric framework pairs 67 nutritional metrics with 19 goals, assigning priority metrics, supporting metrics, directions, and starting thresholds.
- Users can maintain multiple goals and revisit the same food records through different goal-directed framings.
- Meal photos are processed through configuration, generation, and anchoring stages that produce structured nutritional estimates and goal-relative assessments.
- FoodData Central recomputes database-covered metrics, while food identity, quantities, and unsupported metrics remain model estimates.
- Changing goals creates a new configuration and re-executes generation and anchoring on every stored meal without schema migration.
- The interface keeps feedback structure constant while metrics, colors, narratives, and visualizations change with the active goal.
5 User Study
The mixed-method deployment examined how participants experienced goal-directed feedback and reframing when switching between existing dietary goals. A one-week protocol combined meal logging, interviews, surveys, and qualitative analysis, with findings interpreted against important study boundaries.
- Study Questions: The study asked how users make sense of goal-directed feedback and what reflection emerges when stored data is reframed under a new goal.The investigation was organized around RQ1 and RQ2.
- Protocol: Participants tracked meals under an initial goal for four days, switched to another existing goal mid-study, and continued reviewing feedback over seven days.The scheduled switch examined how first reframing was perceived rather than measuring behavior change.
- Participants: Four participants withdrew, leaving N=21 adults with two or three concurrent goals spanning 14 of the framework’s 19 goals.Weight loss and muscle gain were the most common goals, with 49 goals represented overall.
- Analysis: The study combined a pre/post 11-item survey, semi-structured interviews, daily mini-surveys, and reflexive thematic analysis.Survey trends identified changed reflection scores, while qualitative data helped explain them and distinguish individual cases from broader trends.
- Limitations: The one-week scheduled switch captures how a first reframing lands, not repeated self-initiated changes or effects after novelty fades.The survey also compared Rebite with recalled impressions of prior tools rather than a controlled baseline.
6 Study Results
The results combine survey trends with qualitative evidence to show how goal-directed framing shaped participants’ engagement with their goals.
- Results: Survey results served as a quantitative reference, while qualitative evidence showed how goal-directed framing shaped engagement with goals.The findings were organized around RQ1 and RQ2.
6.1 Statistical Analysis of Pre–Post Survey Scores
Participants reported stronger goal-related reflection with Rebite than with their prior food-journaling tools, while Exploration and Insight changes were not significant after correction.
- Goal-Related Reflection: Five goal-related reflection items were significantly higher with Rebite than participants’ recalled ratings of prior tools, with effect sizes d_z=0.75–1.10.The items concerned awareness, meal-goal fit, meal patterns, and rethinking what achieving a goal means.
- TSRI-Derived Subscales: Exploration increased by ΔM=0.27 but did not reach significance after correction (adjusted p=.078, d_z=0.44).Its unadjusted p value was .057.
- TSRI-Derived Subscales: Insight increased by ΔM=0.41 but did not survive correction for multiple comparisons (adjusted p=.078, d_z=0.48).The unadjusted p value was .039.
- Goal-Related Reflection: Participants reported greater awareness of the active dietary goal with Rebite than prior tools (M_prior=4.00, M_Rebite=4.71, d_z=0.91).This difference was significant, t(20)=4.18, p<.001.
- Goal-Related Reflection: Reports of noticing consistent or inconsistent meal patterns increased by ΔM=1.10, the largest goal-related item change.The difference was significant, t(20)=5.04, p<.001, d_z=1.10.
6.2 Making Sense of Goal-Directed Feedback (RQ1)
Goal-directed feedback translated vague dietary goals into concrete metrics that participants could accept, expand, or challenge. The available metric space also limited reframing when it failed to represent lived experience.
- Making Sense of Goal-Directed Feedback: The system mapped vague goals into concrete metric configurations, enabling participants to confirm priorities, expand goals, and identify mismatches with lived experience.This mapping also provided the basis for later re-translation when goals changed.
- Confirming Existing Understanding: Participants trusted the translation when it surfaced metrics they already cared about, such as iron and B12 for plant-based eating or protein for muscle gain.Trust in the initial translation made participants readier to attend to later re-translation.
- Expanding Vague Goals: The system added overlooked metrics that made vague goals more concrete, including fiber for weight loss and polyphenols for anti-inflammatory eating.Participants could later find these additions useful or irrelevant.
- Expanding Vague Goals: Some participants rejected added metrics, including a spice-related metric and fiber, supporting user adjustment of the configuration.These cases show that system-generated expansions did not always fit participants’ needs.
- Metric-Space Boundaries: Reframing had little to work with when available metrics could not represent goals such as recovery or digestive complexity.Participants described recovery as also involving mood, training load, and rest, while FODMAPs, lactose, and fiber missed one participant’s experience.
6.3 What Happened When Goals Changed (RQ2)
Reframing past meals under changed goals made overlaps, conflicts, and trade-offs visible. Participants used these cross-goal readings to preserve useful strategies, negotiate priorities, or sometimes seek a fresh start.
- Finding overlap between goals.: Reframing showed that the same metrics, foods, and routines could support multiple goals, reducing the sense that participants had to start over.Some participants saw continuity between goals, while others discovered new benefits in previously logged meals.
- Finding conflict between goals.: The same meal could receive contrasting assessments under different goals, making clear that foods had no single stable reading.Salmon was praised for protein under weight loss but also raised sodium concerns under blood-lipid and blood-pressure goals.
- Revising and refining goals.: Conflicting feedback prompted participants to negotiate trade-offs by adjusting portions or switching priorities according to daily context.One participant balanced weight loss and blood-sugar concerns by eating half a bowl of red bean soup, while another prioritized carbohydrates after training but not on rest days.
- Revising and refining goals.: Some participants reordered their goal hierarchy or redefined health after reframing exposed tensions between competing priorities.Participants described placing long-term heart health above weight loss and recognizing that eating fewer calories was not equivalent to eating well.
- Starting fresh.: Not everyone wanted past data reframed: some participants viewed a goal switch as a break with the past and preferred resetting their records.Resetting could mark a fresh-start ritual, protect motivation, or avoid emotionally uncomfortable reinterpretations.
7 Discussion
The discussion characterizes goal-directed framing as a two-step process: mapping goals to metrics creates a shared language, and reframing past data then exposes cross-goal relationships. These relationships can support reflection but also create discomfort and require user control.
- 7.1 Enabling Reframing: Goal-to-metric mapping translated vague goals into a shared language that enabled participants to understand goals concretely and support later reframing.Participants could use the resulting metric interpretation to examine past meals under a changed goal.
- 7.2 Reframing as Reflection: Reframing made overlaps and conflicts between goals visible, helping participants negotiate trade-offs rather than combining competing priorities into a single score.The authors argue that preserving visible conflict gives users material for reasoning and acting on trade-offs themselves.
- 7.2 Reframing as Reflection: Cross-goal relationships emerged sequentially because participants viewed each goal in turn, suggesting value in a direct comparison view during deliberate multi-goal reflection.The study’s everyday interaction favored brief single-goal glances, while explicit trade-off reasoning might benefit from side-by-side comparison.
- 7.2 Reframing as Reflection: Reframing could cause discomfort because reassessing past meals against a new standard sometimes felt like judgment rather than actionable advice.Some participants therefore preferred resetting instead of receiving a broad reassessment of stored data.
- 7.2 User Control: Rebite gave users a largely binary choice between accepting a new assessment and resetting, motivating controls over reframing scope and selective acceptance or rejection.Proposed controls include choosing which data and granularity to reinterpret, deferring or rejecting assessments, and annotating past choices.
7.3 Generalizing Reframing: When Data Lives Many Lives
Keeping data open to reinterpretation can support multiple situated understandings of health, but deferring interpretation also creates motivational and safety risks. Generalization depends on data supporting multiple readings, a knowledge base mapping goals to dimensions, and separable factual estimates and assessments.
- 7.3.1 The same data can mean many things.: Goal-directed framing keeps raw data available for different goal lenses, allowing one meal to support interpretations such as recovery fuel, weight-management risk, or dietary diversity.The approach may resist rigid health narratives by treating health as situational rather than fixed to one frame.
- 7.3.1 The same data can mean many things.: LLM-based reinterpretation may amplify harmful narratives because biased model language can reinforce restrictive dieting or pro-eating-disorder tendencies.Published guidance constrains metric selection only partly, and the framework is authored rather than clinically validated.
- 7.3.2 The cost of deferring interpretation.: Deferring interpretation lowers the barrier to logging and preserves data across changing priorities, but indefinite deferral may erode the motivation for careful recording.Future systems may need to support goal discovery and refinement alongside open-ended data preservation.
- 7.3 Generalizing Reframing: The approach may generalize when raw data supports multiple interpretations, a structured knowledge base maps goals to relevant dimensions, and factual estimates remain separable from assessments.Sleep data is offered as an example domain with multiple dimensions that could be framed through different goal-specific knowledge bases.
8 Conclusion
The paper introduces goal-directed framing as an active lens over preserved data and realizes it in Rebite. Reframing past meals surfaced overlaps and conflicts that prompted trade-off negotiation and priority refinement, although some participants preferred starting fresh.
- 8 Conclusion: Goal-directed framing uses a changing goal to reinterpret preserved data, and Rebite applies this approach to past food-journal entries.The same stored meals can be examined under new dietary priorities without being logged again.
- 8 Conclusion: Reframed meals exposed cross-goal overlaps and conflicts, prompting participants to negotiate trade-offs and refine priorities, while some preferred a fresh start.The conclusion presents these outcomes as support for reflection alongside a boundary on universal reframing.