Agentopic: A Generative AI Agent Workflow for Explainable Topic Modeling
Achieves F1 of 0.95 on BBC dataset, matching GPT-4.1 with full reasoning trace.
Agentopic is a novel agent-based workflow for explainable topic modeling that leverages the reasoning capabilities of Large Language Models (LLMs). It uses multiple agents to collaboratively identify, validate, hierarchically group, and explain topics. When seeded with topics from the British Broadcasting Corporation (BBC) dataset, Agentopic achieves an F1-score of 0.95, matching GPT-4.1, improving on LDA (0.93), and close to BERTopic (0.98). By embedding explainability throughout the workflow, Agentopic offers an interpretable alternative to black-box models, making it valuable for applications like finance and healthcare.
- Agentopic uses multiple LLM-based agents (identification, validation, grouping, explanation) for transparent topic modeling.
- On the BBC dataset, it achieves an F1-score of 0.95—matching GPT-4.1 and beating LDA (0.93), close to BERTopic (0.98).
- Unseeded, it auto-generated 2045 topics across six hierarchical levels, enriching the original five-category structure.
Why It Matters
Explainable AI for topic modeling moves from black boxes to traceable reasoning, critical for regulated industries.