Skip to main content
Tessa NewsLink
Paste a health news link, or browse

Source study found

Story checked

AI detects daily stress by analyzing routine smartphone behaviors (opens in a new tab)

news-medical.net · 2026-10-14

Short answerEvidenceSource

Short answer

Mostly not supported

Mostly not supported.

One claim goes further than the study. 4 other points were not covered by the paper.

  • 1 supported
  • 1 overstated
  • 4 not covered

Checked against the study summary. The full text wasn't available, so some details couldn't be settled either way.

Share this check

Follow the evidence trail
1
2

NewsLink checks it

Mostly not supported

One claim overstates the study. One of six checks out. Four claims the study doesn't address.

  • 1 supported
  • 1 overstated
  • 4 not covered
Open claim evidence
3

The source study

Leveraging Smartphone Human Interaction Routine Behavior Task Mining and Modeling for Daily Stress Monitoring

Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · 2025
Open source record (opens in a new tab)
Then inspect each claim

Evidence layer

Claim by claim

Each claim gets a verdict. Expand it to see the evidence directly below.

6 claims in this story

Showing all 6 claimsChoose a verdict to focus the list.

Claim 1 of 6OverstatedKAIST researchers developed a technology called SHIRBT-MMF that analyzes repeated smartphone behaviors, including duration, frequency, and text-input patterns, to identify a person's daily stress state.View evidenceHide evidence

Why this verdict

The abstract-level profile supports that the researchers introduced SHIRBT-MMF, a smartphone interaction-routine framework for personalized daily stress monitoring, and that it mines within-app UI interaction routines. However, the headline-style claim presents the system as identifying a person’s daily stress state without qualification, while the abstract evidence is for predictive classification with about 75% average accuracy in a small observational validation. The specific listed inputs—duration, frequency, and text-input patterns—are not verifiable from the abstract profile.

1

Study evidence

SHIRBT-MMF extracts interpretable, task-level features from within-app UI state transitions intended to mitigate covariate shift and feature evolution compared with aggregated app-usage approaches.

“we propose the Smartphone Human Interaction-based Routine Behavior Task Mining, Modeling, and Feature Extraction (SHIRBT-MMF) framework”
2

Study evidence

Personalized SHIRBT-based model achieved an average accuracy of 75% on the in-the-wild datasets.75% average accuracy

“We validated SHIRBT-MMF through one- and four-month in-the-wild datasets with 26 participants”
Claim 2 of 6Not coveredThe system focuses on 'what a person does' rather than which app they use, so activities such as messenger chatting can be recognized across different messaging apps.View evidenceHide evidence

Why this verdict

The paper profile supports the general idea that SHIRBT-MMF extracts task-level within-app interaction features rather than relying on aggregate app-usage statistics or app types. But the more specific story claim that activities such as messenger chatting can be recognized across different messaging apps is not stated in the abstract-level evidence.

1

Study evidence

SHIRBT-MMF extracts interpretable, task-level features from within-app UI state transitions intended to mitigate covariate shift and feature evolution compared with aggregated app-usage approaches.

“we propose the Smartphone Human Interaction-based Routine Behavior Task Mining, Modeling, and Feature Extraction (SHIRBT-MMF) framework”
Claim 3 of 6Not coveredIn evaluations of classifying the same behavior across email, web browser, and messenger apps, the system achieved more than 95% accuracy.View evidenceHide evidence

As statedmore than 95% accuracy

Why this verdict

The abstract profile reports stress-prediction accuracy of about 75% and does not describe a separate evaluation of classifying the same behavior across email, browser, and messenger apps, nor a more-than-95% behavior-classification result. This may exist in the full paper, but it is not verifiable at abstract depth.

1

Study evidence

SHIRBT-MMF extracts interpretable, task-level features from within-app UI state transitions intended to mitigate covariate shift and feature evolution compared with aggregated app-usage approaches.

“we propose the Smartphone Human Interaction-based Routine Behavior Task Mining, Modeling, and Feature Extraction (SHIRBT-MMF) framework”
2

Study evidence

Personalized SHIRBT-based model achieved an average accuracy of 75% on the in-the-wild datasets.75% average accuracy

“We validated SHIRBT-MMF through one- and four-month in-the-wild datasets with 26 participants”
Claim 4 of 6Not coveredThe researchers analyzed real-life smartphone records and self-reported daily stress from 26 university students and office workers in their 20s and 30s, including 16 short-term participants and 10 long-term participants.View evidenceHide evidence

As stated26 participants; 16 short-term participants; 10 long-term participants

Why this verdict

The abstract-level profile supports in-the-wild smartphone datasets with 26 participants over one- and four-month periods. It does not verify the demographic details, the university-student/office-worker composition, self-reported daily stress collection details, or the 16 short-term/10 long-term participant split.

1

Study evidence

Personalized SHIRBT-based model achieved an average accuracy of 75% on the in-the-wild datasets.75% average accuracy

“We validated SHIRBT-MMF through one- and four-month in-the-wild datasets with 26 participants”
Claim 5 of 6Not coveredThe best-performing personalized model achieved average stress-classification accuracy of 75.9% for short-term participants and 73.8% for long-term participants, outperforming a conventional app-usage approach by 5.5 and 3.5 percentage points, respectively.View evidenceHide evidence

As stated75.9% and 73.8% accuracy; 5.5 and 3.5 percentage-point gains

Why this verdict

The abstract profile supports the broad direction: a personalized SHIRBT-based model achieved about 75% average accuracy and outperformed baselines by about 5%. It does not provide the exact subgroup accuracies of 75.9% and 73.8%, the 5.5- and 3.5-percentage-point gains, or the precise definition of the conventional app-usage baseline, so the detailed magnitude claim is not verifiable at this evidence depth.

1

Study evidence

Personalized SHIRBT-based model achieved an average accuracy of 75% on the in-the-wild datasets.75% average accuracy

“We validated SHIRBT-MMF through one- and four-month in-the-wild datasets with 26 participants”
Claim 6 of 6SupportedThe article says the study used associations between behavior and stress and did not demonstrate that any particular behavior causes or reduces stress.View evidenceHide evidence

Why this verdict

The paper profile characterizes the human evaluation as an in-the-wild longitudinal observational predictive-modeling study. Nothing in the abstract-level profile demonstrates causal effects of particular behaviors on stress, so the story’s caveat that the study used associations and did not establish causation is consistent with the supplied evidence.

1

Study evidence

Personalized SHIRBT-based model achieved an average accuracy of 75% on the in-the-wild datasets.75% average accuracy

“We validated SHIRBT-MMF through one- and four-month in-the-wild datasets with 26 participants”
Then look for missing context

Context layer

What the story left out

Important study details the story did not include.

  • The framework uses multi-level sequential pattern mining and LLM-based automated task modeling to extract interpretable task-level features from UI state transitions.

    The story broadly refers to an AI system and repeated behaviors, but it does not mention the paper’s specific methodological components: multi-level sequential pattern mining, LLM-based task modeling, or UI state-transition mining.

    From SHIRBT-MMF framework

  • The paper reports that SHIRBT achieved its performance while using only 3–6% of app types.

    This feature-parsimony result is material in the abstract profile, but the presented story claims do not explicitly report the 3–6% app-type result.

    From In-the-wild longitudinal observational (1-month and 4-month smartphone datasets)

  • The abstract profile lacks detailed evaluation information such as model architectures, train/test protocol, baseline definitions, uncertainty estimates, and statistical significance.

    The story reports accuracy gains but does not mention that, at abstract depth, the evidence does not provide uncertainty, statistical significance, or detailed baseline/evaluation protocol information.

    From In-the-wild longitudinal observational (1-month and 4-month smartphone datasets)

  • The abstract profile notes that privacy and ethical trade-offs of collecting within-app UI state transitions, and dependence on LLMs, are not detailed.

    The story does not mention privacy, ethics, or LLM-dependence limitations related to within-app UI interaction data collection and automated task modeling.

    From SHIRBT-MMF framework

6 things the story did carry across
  • SHIRBT-MMF is a framework for mining fine-grained within-app UI interaction routines/tasks for smartphone-based stress monitoring.
  • The paper’s stated rationale is that task-level features are more stable than aggregate app-usage features and may mitigate covariate shift or feature evolution over time.
  • The in-the-wild validation used one- and four-month smartphone datasets from 26 participants.
  • The reported headline performance in the abstract is about 75% average accuracy and about 5% improvement over baselines.
  • The validation evidence is observational and predictive, not causal evidence that specific behaviors cause, increase, or reduce stress.
  • Important abstract-level limitations include small sample size and uncertain generalizability beyond the validation sample.
Then read the study layer

Study layer

Study at a glance

Scan the study first. Expand only the parts you want to inspect.

Pieces of work

2

Evidence read

study summary

Lead result

human in vivo

1Lead resulthuman in vivoEmpirically validate SHIRBT-MMF for daily stress monitoring using in-the-wild smartphone datasets (1-month and 4-month) from 26 participants, showing improved predictive accuracy vs baselines while using fewer app types and demonstrating longitudinal feature stability/reduced covariate shift.In-the-wild longitudinal observational (1-month and 4-month smartphone datasets)Expand

In plain English

The authors report an in-the-wild longitudinal evaluation (one- and four-month datasets) of the SHIRBT-MMF framework for daily stress monitoring using smartphone interaction data from 26 participants. They claim that a personalized SHIRBT-based model achieved 75% average accuracy, outperforming baseline models by ~5% while using only 3–6% of app types, and that SHIRBT-derived features remain stable over time, reducing covariate shift.

Key findings

  • Personalized SHIRBT-based model achieved an average accuracy of 75% on the in-the-wild datasets.75% average accuracy
  • SHIRBT-based personalized model outperformed baseline models by ~5% while using only 3–6% of app types.≈5% improvement vs baselines; 3–6% of app types used
“We validated SHIRBT-MMF through one- and four-month in-the-wild datasets with 26 participants”
What this piece can’t prove
  • Summary and results are based solely on the abstract; full paper may provide further methodological and statistical details not present here.

2 further details could not be confirmed from the summary.

2in silicoIntroduce SHIRBT-MMF, a framework that mines fine-grained smartphone within-app UI interaction routines (tasks) using multi-level sequential pattern mining and LLM-based automated task modeling to produce interpretable, stable features for stress monitoring under distribution shift/feature evolution.SHIRBT-MMF frameworkExpand

In plain English

The paper introduces SHIRBT-MMF, a methodological framework that mines fine-grained within-app UI interaction routines (tasks) using multi-level sequential pattern mining and large language model (LLM)-based automated task modeling to produce interpretable, stable features for personalized smartphone-based stress monitoring. The framework emphasizes extracting task-level features from UI state transitions rather than aggregate app-usage statistics, aiming to mitigate covariate shift and feature/schema evolution while improving model robustness and interpretability.

Key findings

  • SHIRBT-MMF extracts interpretable, task-level features from within-app UI state transitions intended to mitigate covariate shift and feature evolution compared with aggregated app-usage approaches.
  • In reported in-the-wild validation (1- and 4-month datasets with 26 participants), a SHIRBT-based personalized model achieved an average accuracy of 75%, outperforming baseline models by 5% while using only 3–6% of app types.75% accuracy; +5% vs baseline; uses 3–6% of app types
“we propose the Smartphone Human Interaction-based Routine Behavior Task Mining, Modeling, and Feature Extraction (SHIRBT-MMF) framework”
What this piece can’t prove
  • Abstract does not detail privacy/ethical trade-offs of collecting within-app UI state transitions or dependence on LLMs (e.g., computation, privacy of prompts).

2 further details could not be confirmed from the summary.

Finally, the search trail

Method layer

NewsLink found the paper. Tessa takes you deeper.

NewsLink checks the story. Tessa is where you inspect the paper, authors, evidence, and research context.

Open the paper in Tessa

Leveraging Smartphone Human Interaction Routine Behavior Task Mining and Modeling for Daily Stress Monitoring

Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · 2025

Source record (opens in a new tab)

Why this one

Near certain

NewsLink found the paper. Tessa is where you inspect it deeply.

Papers considered

The selected paper, plus nearby candidates.

Crossref, PubMed, Europe PMC · 16 candidate papers

Selected

Leveraging Smartphone Human Interaction Routine Behavior Task Mining and Modeling for Daily Stress Monitoring

Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies · 2025 · Crossref

Source (opens in a new tab)

And 10 more candidates considered.