From Data to Insights: LLMs for Financial Narration
Date:
Date/Time: Wednesday, 19 November 2025, 5:00-6:00pm
Venue: Seminar Room 21 (COM3-02-60), National University of Singapore
Abstract
How can we bridge the gap between raw data and actionable insights using language models? This talk examines data narration in financial domains, where converting market data into analytical narratives remains a significant challenge. Yajing will first present DATATALES [1], a financial benchmark revealing that current LLMs fall short in performing the complex analysis required for professional financial reporting. She then introduces KAHAN [2], a novel framework that treats LLMs as financial domain experts, employing knowledge-augmented hierarchical reasoning to generate comprehensive market narratives. Attendees will learn about the unique challenges in financial NLP and discover how systematic knowledge integration and multi-level analysis can unlock LLMs’ potential for specialized analytical tasks.
References
[1] DATATALES: A Benchmark for Real-World Intelligent Data Narration. Yajing Yang, Qian Liu, Min-Yen Kan. EMNLP 2024. https://aclanthology.org/2024.emnlp-main.601/
[2] KAHAN: Knowledge-Augmented Hierarchical Analysis and Narration for Financial Data Narration. Yajing Yang, Tony Deng, Min-Yen Kan. EMNLP Findings 2025. https://aclanthology.org/2025.findings-emnlp.1405/
Bio
Yajing Yang is a PhD candidate under the supervision of Prof. Min-Yen Kan at the School of Computing, National University of Singapore, while working as a senior data scientist at Rio Tinto. Her research centers on data-to-text generation with a particular emphasis on financial NLP applications. She investigates how large language models can be enhanced to perform sophisticated analytical reasoning over structured financial data, incorporating domain expertise and hierarchical thinking to produce narratives that support investment decision-making. Her work addresses both theoretical challenges in financial text generation and practical deployment considerations for industry applications.