Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HR-Agent: A Task-Oriented Dialogue (TOD) LLM Agent Tailored for HR Applications

Weijie Xu, Jay Desai, Fanyou Wu, Josef Valvoda, Srinivasan H. Sengamedu

TL;DR

This work presents HR-Agent, an efficient, confidential, and HR-specific LLM-based task-oriented dialogue system tailored for automating repetitive HR processes such as medical claims and access requests.

Abstract

Recent LLM (Large Language Models) advancements benefit many fields such as education and finance, but HR has hundreds of repetitive processes, such as access requests, medical claim filing and time-off submissions, which are unaddressed. We relate these tasks to the LLM agent, which has addressed tasks such as writing assisting and customer support. We present HR-Agent, an efficient, confidential, and HR-specific LLM-based task-oriented dialogue system tailored for automating repetitive HR processes such as medical claims and access requests. Since conversation data is not sent to an LLM during inference, it preserves confidentiality required in HR-related tasks.

HR-Agent: A Task-Oriented Dialogue (TOD) LLM Agent Tailored for HR Applications

TL;DR

This work presents HR-Agent, an efficient, confidential, and HR-specific LLM-based task-oriented dialogue system tailored for automating repetitive HR processes such as medical claims and access requests.

Abstract

Recent LLM (Large Language Models) advancements benefit many fields such as education and finance, but HR has hundreds of repetitive processes, such as access requests, medical claim filing and time-off submissions, which are unaddressed. We relate these tasks to the LLM agent, which has addressed tasks such as writing assisting and customer support. We present HR-Agent, an efficient, confidential, and HR-specific LLM-based task-oriented dialogue system tailored for automating repetitive HR processes such as medical claims and access requests. Since conversation data is not sent to an LLM during inference, it preserves confidentiality required in HR-related tasks.

Paper Structure

This paper contains 24 sections, 5 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods
  4. Baseline Methods
  5. Entity Selection
  6. Entity Extraction
  7. Question Generation
  8. System Evaluation
  9. Prompt
  10. Other considerations
  11. Conclusion
  12. Ethical Statement
  13. Limitations and Risks
  14. Future Work
  15. Synthetic Example
  16. ...and 9 more sections

Figures (5)

  • Figure 1: We systematically compared our method to Claude’s approach zhang2023sgp in terms of response time by collecting 40 conversations across four different categories: time off, medical claims, resume creation, and issue ticket handling. The HR-Agent we propose demonstrates significantly faster response times compared to the Claude-based TOD. In fact, our system achieves a response time of less than 2 seconds in 94 percent of cases, while the Claude-based system accomplishes this in only 4 percent of cases. These results highlight the substantial speed advantage of our HR-Agent over the Claude-based solution.
  • Figure 2: An illustration of the solution. The Entity Selection Model identifies relevant entity. The selected entity is passed to the Entity Extraction model to find the relevant word in the utterance. Based on Schema's memory and previous utterance. The Question Generation model is used to generate the next question. The HR-Agent system then connects to the API to finish the relevant tasks such as drafting email, requesting time off and setting status.
  • Figure 3: Labeler preferences for responses from HR-Agent (A) and HR-MultiWOZ (B).
  • Figure 4: User interface used for the human preference study.
  • Figure 5: MTurk Questions and selected examples