Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Neuro-Symbolic Approach to Monitoring Salt Content in Food

Anuja Tayal, Barbara Di Eugenio, Devika Salunke, Andrew D. Boyd, Carolyn A Dickens, Eulalia P Abril, Olga Garcia-Bedoya, Paula G Allen-Meares

TL;DR

This work targets monitoring salt content in foods for heart failure patients by developing a specialized dialogue system. It creates a template-based conversational dataset from the US Food Data Central and a FoodOn-based ontology to model salt-content inquiries with clarifying questions, then combines a PPTOD-based TOD model with neuro-symbolic rules (NS-PPTOD) to retrieve exact salt values or compute them for non-standard weights. The NS-PPTOD system yields a roughly 20% improvement in joint goal accuracy over fine-tuned PPTOD across dataset sizes, and DST performance rises significantly with the neuro-symbolic integration. In addition to improved accuracy, NS-PPTOD demonstrates markedly lower reading level requirements than ChatGPT, highlighting practical benefits for patient-facing dietary guidance and accessibility in real-world healthcare settings.

Abstract

We propose a dialogue system that enables heart failure patients to inquire about salt content in foods and help them monitor and reduce salt intake. Addressing the lack of specific datasets for food-based salt content inquiries, we develop a template-based conversational dataset. The dataset is structured to ask clarification questions to identify food items and their salt content. Our findings indicate that while fine-tuning transformer-based models on the dataset yields limited performance, the integration of Neuro-Symbolic Rules significantly enhances the system's performance. Our experiments show that by integrating neuro-symbolic rules, our system achieves an improvement in joint goal accuracy of over 20% across different data sizes compared to naively fine-tuning transformer-based models.

A Neuro-Symbolic Approach to Monitoring Salt Content in Food

TL;DR

This work targets monitoring salt content in foods for heart failure patients by developing a specialized dialogue system. It creates a template-based conversational dataset from the US Food Data Central and a FoodOn-based ontology to model salt-content inquiries with clarifying questions, then combines a PPTOD-based TOD model with neuro-symbolic rules (NS-PPTOD) to retrieve exact salt values or compute them for non-standard weights. The NS-PPTOD system yields a roughly 20% improvement in joint goal accuracy over fine-tuned PPTOD across dataset sizes, and DST performance rises significantly with the neuro-symbolic integration. In addition to improved accuracy, NS-PPTOD demonstrates markedly lower reading level requirements than ChatGPT, highlighting practical benefits for patient-facing dietary guidance and accessibility in real-world healthcare settings.

Abstract

We propose a dialogue system that enables heart failure patients to inquire about salt content in foods and help them monitor and reduce salt intake. Addressing the lack of specific datasets for food-based salt content inquiries, we develop a template-based conversational dataset. The dataset is structured to ask clarification questions to identify food items and their salt content. Our findings indicate that while fine-tuning transformer-based models on the dataset yields limited performance, the integration of Neuro-Symbolic Rules significantly enhances the system's performance. Our experiments show that by integrating neuro-symbolic rules, our system achieves an improvement in joint goal accuracy of over 20% across different data sizes compared to naively fine-tuning transformer-based models.
Paper Structure (18 sections, 1 equation, 2 figures, 6 tables)

This paper contains 18 sections, 1 equation, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dataset Creation
  4. Data Source
  5. Ontology Construction
  6. Conversational Dataset Creation
  7. Methodology
  8. Plug and Play Task Oriented Dialog System (PPTOD)
  9. NS-PPTOD
  10. Evaluation
  11. End-to-End Dialog Modeling
  12. DST Module
  13. Analysis
  14. Comparison with ChatGPT
  15. Conclusion and Future Work
  16. ...and 3 more sections

Figures (2)

  • Figure 1: Sample Template Conversation which begins with the user asking about the salt content in food. The system asks clarification questions to determine the precise food item and its salt value.
  • Figure 2: NS-PPTOD Model with Example: This example illustrates that PPTOD alone is not able to identify correct salt value for the food item