Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

mind_call: A Dataset for Mental Health Function Calling with Large Language Models

Fozle Rabbi Shafi, M. Anwar Hossain, Salimur Choudhury

TL;DR

The paper addresses the lack of function-calling datasets for mental health applications that leverage wearable sensor data. It introduces mind_call, a synthetic dataset that maps natural language queries to structured API calls over a standardized health data schema, including an explicit intermediate reasoning field and normalized temporal parameters. A multi-stage LLM-generation pipeline creates diverse, realistic samples, complemented by exploratory data analysis to validate intent grounding and temporal normalization. Mind_call is publicly released on Hugging Face and is designed to enable reliable grounding of natural language queries to wearable health data for safe, interpretable mental health support, facilitating supervised fine-tuning and evaluation of LLM-based agents.

Abstract

Large Language Model (LLM)-based systems increasingly rely on function calling to enable structured and controllable interaction with external data sources, yet existing datasets do not address mental health-oriented access to wearable sensor data. This paper presents a synthetic function-calling dataset designed for mental health assistance grounded in wearable health signals such as sleep, physical activity, cardiovascular measures, stress indicators, and metabolic data. The dataset maps diverse natural language queries to standardized API calls derived from a widely adopted health data schema. Each sample includes a user query, a query category, an explicit reasoning step, a normalized temporal parameter, and a target function. The dataset covers explicit, implicit, behavioral, symptom-based, and metaphorical expressions, which reflect realistic mental health-related user interactions. This resource supports research on intent grounding, temporal reasoning, and reliable function invocation in LLM-based mental health agents and is publicly released to promote reproducibility and future work.

mind_call: A Dataset for Mental Health Function Calling with Large Language Models

TL;DR

The paper addresses the lack of function-calling datasets for mental health applications that leverage wearable sensor data. It introduces mind_call, a synthetic dataset that maps natural language queries to structured API calls over a standardized health data schema, including an explicit intermediate reasoning field and normalized temporal parameters. A multi-stage LLM-generation pipeline creates diverse, realistic samples, complemented by exploratory data analysis to validate intent grounding and temporal normalization. Mind_call is publicly released on Hugging Face and is designed to enable reliable grounding of natural language queries to wearable health data for safe, interpretable mental health support, facilitating supervised fine-tuning and evaluation of LLM-based agents.

Abstract

Large Language Model (LLM)-based systems increasingly rely on function calling to enable structured and controllable interaction with external data sources, yet existing datasets do not address mental health-oriented access to wearable sensor data. This paper presents a synthetic function-calling dataset designed for mental health assistance grounded in wearable health signals such as sleep, physical activity, cardiovascular measures, stress indicators, and metabolic data. The dataset maps diverse natural language queries to standardized API calls derived from a widely adopted health data schema. Each sample includes a user query, a query category, an explicit reasoning step, a normalized temporal parameter, and a target function. The dataset covers explicit, implicit, behavioral, symptom-based, and metaphorical expressions, which reflect realistic mental health-related user interactions. This resource supports research on intent grounding, temporal reasoning, and reliable function invocation in LLM-based mental health agents and is publicly released to promote reproducibility and future work.
Paper Structure (15 sections, 5 equations, 5 figures)

This paper contains 15 sections, 5 equations, 5 figures.

Table of Contents

  1. Introduction
  2. Task Definition
  3. Motivation: Wearable Sensor Data for Mental Health Assistance
  4. Dataset Overview
  5. Health Data Schema and Function Definition
  6. Exploratory Data Analysis of the mind_call Dataset
  7. Query Types and Intent Diversity
  8. Most Frequent Mental Health--Related Terms
  9. Most Common Emotional Terms in Queries
  10. Temporal Expression Distribution and Mapping
  11. Explicit Reasoning Annotation
  12. Dataset Generation Process
  13. Ethical and EDI Considerations
  14. Limitations and Future Works
  15. Conclusion

Figures (5)

  • Figure 1: Distribution of query types in the mind_call dataset. The dataset includes explicit, implicit, behavioral, symptom-based, and metaphorical queries, reflecting diverse user intent expressions.
  • Figure 2: Most frequent mental health--related terms in the mind_call dataset, reflecting common topics and concerns aligned with the thesis scope.
  • Figure 3: Most common emotional terms appearing in user queries, highlighting the affective context of interactions in the mind_call dataset.
  • Figure 4: Distribution of temporal expressions across explicit, relative, and vague forms. All expressions are normalized to a numeric numdays parameter (default = 7 when unspecified).
  • Figure 5: Examples illustrating how user queries are paired with corresponding thinking annotations in the mind_call dataset. The reasoning process explains intent interpretation, function selection, and temporal normalization.