MALTopic was published as a research paper at the World AI IoT Congress 2025. This research/concept was later converted into an actual implementation and published as a python library.

MALTopic introduces a multi-agent LLM framework for topic modeling that goes beyond traditional methods by integrating both structured and unstructured survey data. The system uses specialized agents to enrich, extract, and refine topics, resulting in more coherent, diverse, and human-readable themes. This approach makes it especially effective for analyzing complex survey responses, outperforming classic models like LDA and BERTopic in interpretability and contextual relevance.

Why did I build this? Because LLMs are remarkably adept at understanding natural language—something traditional topic modeling frameworks simply can’t match. My own challenges with survey data analysis highlighted this gap and sparked my curiosity. Diving into LLMs, exploring the latest research, and recognizing the untapped potential in this space inspired me to contribute something new. The rapid evolution of LLM technology and the clear need for better solutions drove me to develop MALTopic and share these insights through a research paper.

✨🚀 If you find MALTopic interesting or see ways it could be improved, I’d love for you to get involved! 🤝
Feel free to contribute updates, suggest enhancements, or open issues on the GitHub repo to start a conversation!