Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automated Factual Benchmarking for In-Car Conversational Systems using Large Language Models

Rafael Giebisch, Ken E. Friedl, Lev Sorokin, Andrea Stocco

TL;DR

The paper tackles the challenge of ensuring factual correctness in LLM-based in-car conversational systems. It introduces a black-box LLM-driven benchmarking framework with five prompting methods to assess CarExpert's outputs against an owner’s manual ground truth, supported by a domain-specific dataset. Evaluations across multiple LLMs and temperatures show that GPT-4 with Input-Output prompting achieves strong accuracy and efficiency, suggesting LLM-based testing as a viable tool for validating real-world in-car QA systems. The work highlights practical implications for scalable, automated validation and outlines avenues for synthetic data generation and multilingual expansion.

Abstract

In-car conversational systems bring the promise to improve the in-vehicle user experience. Modern conversational systems are based on Large Language Models (LLMs), which makes them prone to errors such as hallucinations, i.e., inaccurate, fictitious, and therefore factually incorrect information. In this paper, we present an LLM-based methodology for the automatic factual benchmarking of in-car conversational systems. We instantiate our methodology with five LLM-based methods, leveraging ensembling techniques and diverse personae to enhance agreement and minimize hallucinations. We use our methodology to evaluate CarExpert, an in-car retrieval-augmented conversational question answering system, with respect to the factual correctness to a vehicle's manual. We produced a novel dataset specifically created for the in-car domain, and tested our methodology against an expert evaluation. Our results show that the combination of GPT-4 with the Input Output Prompting achieves over 90 per cent factual correctness agreement rate with expert evaluations, other than being the most efficient approach yielding an average response time of 4.5s. Our findings suggest that LLM-based testing constitutes a viable approach for the validation of conversational systems regarding their factual correctness.

Automated Factual Benchmarking for In-Car Conversational Systems using Large Language Models

TL;DR

The paper tackles the challenge of ensuring factual correctness in LLM-based in-car conversational systems. It introduces a black-box LLM-driven benchmarking framework with five prompting methods to assess CarExpert's outputs against an owner’s manual ground truth, supported by a domain-specific dataset. Evaluations across multiple LLMs and temperatures show that GPT-4 with Input-Output prompting achieves strong accuracy and efficiency, suggesting LLM-based testing as a viable tool for validating real-world in-car QA systems. The work highlights practical implications for scalable, automated validation and outlines avenues for synthetic data generation and multilingual expansion.

Abstract

In-car conversational systems bring the promise to improve the in-vehicle user experience. Modern conversational systems are based on Large Language Models (LLMs), which makes them prone to errors such as hallucinations, i.e., inaccurate, fictitious, and therefore factually incorrect information. In this paper, we present an LLM-based methodology for the automatic factual benchmarking of in-car conversational systems. We instantiate our methodology with five LLM-based methods, leveraging ensembling techniques and diverse personae to enhance agreement and minimize hallucinations. We use our methodology to evaluate CarExpert, an in-car retrieval-augmented conversational question answering system, with respect to the factual correctness to a vehicle's manual. We produced a novel dataset specifically created for the in-car domain, and tested our methodology against an expert evaluation. Our results show that the combination of GPT-4 with the Input Output Prompting achieves over 90 per cent factual correctness agreement rate with expert evaluations, other than being the most efficient approach yielding an average response time of 4.5s. Our findings suggest that LLM-based testing constitutes a viable approach for the validation of conversational systems regarding their factual correctness.

Paper Structure

This paper contains 19 sections, 1 equation, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Testing Conversational Systems
  3. Conversational Question Answering Systems
  4. LLM-based Testing Approach
  5. Input-Output Prompting
  6. Chain-of-Thought Prompting
  7. Self-Consistency with Chain-of-Thought
  8. Multi-Persona Self-Collaboration
  9. Round Table Conference
  10. Case Study at BMW
  11. Research Questions
  12. Metrics
  13. Dataset Creation
  14. Configurations
  15. Implementation
  16. ...and 4 more sections

Figures (2)

  • Figure 1: Example dialogues showing both positive (top) and negative (bottom) interactions between a user and CarExpert. In the second response, CarExpert fails to combine documents from the manual to provide a correct response.
  • Figure 2: Overall process of our testing framework for the evaluation of factual correctness of in-car conversational systems.