From Data to Themes: Aspect Mining Meets Agentic AI for Financial Metadata Discovery
International Journal of Intelligent Systems · 2026-01 · 2026 (1)
DOI: 10.1155/int/6821390 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Intelligent Systems, online erschienen am 23.07.2026. Verfasst von Ratnesh Kumar Sharma und 4 weiteren Autor:innen. Die Arbeit nennt 46 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: University of Technology Sydney.
Abstract (Originalsprache)
Thematic analysis (TA) plays a crucial role in uncovering meaningful patterns within unstructured textual data, especially in financial domains. However, manual TA methods often suffer from inefficiency, inconsistency, and difficulty in identifying latent themes; challenges that are exacerbated by the rapid expansion of information systems and big data environments. To address these limitations, this study aims to develop a hybrid, domain‐optimized framework for TA by integrating natural language processing (NLP) and Agentic artificial intelligence (Agentic AI) techniques, with a focus on discovering financial metadata in large‐scale unstructured text. The proposed methodology combines multiple techniques: NLP for data preprocessing, aspect mining for metadata relevance and noise reduction, latent Dirichlet allocation (LDA) for topic modeling, and large language model (LLM) agents for theme identification and validation. Applied to financial‐domain tweets from major Australian cities, temporally diverse data, the framework efficiently identified prevalent and latent themes and reduced end‐to‐end analysis time by 98% (94 h–1 h) compared with manual TA, with only a 7.5% accuracy trade‐off in identifying golden themes. Notably, the best‐performing Agentic AI model exceeded the highest human annotator’s accuracy by 3%; the agentic validation loop achieved 73% accuracy, surpassing the top human score of 70%. Ablation studies further revealed that aspect mining improved thematic precision by 25% relative to manual baselines. This research introduces a novel pathway for automated metadata generation and classification in financial big‐data ecosystems. By enhancing data organization, governance, and usability, the framework supports more informed decision‐making and represents a substantial advancement in the application of AI‐driven TA. These findings underscore the potential of targeted LLM orchestration for scalable, explainable, and cost‐efficient TA in large‐scale financial data environments.
Förderung
- University of Technology Sydney
Zitieren
APA
Sharma, R. K., Bharathy, G., Karimi, F., Sharma, A. K., & Prasad, M. (2026). From Data to Themes: Aspect Mining Meets Agentic AI for Financial Metadata Discovery. International Journal of Intelligent Systems, 2026(1). https://doi.org/10.1155/int/6821390
Vancouver
Sharma RK, Bharathy G, Karimi F, Sharma AK, Prasad M. From Data to Themes: Aspect Mining Meets Agentic AI for Financial Metadata Discovery. International Journal of Intelligent Systems. 2026;2026(1). doi:10.1155/int/6821390
BibTeX
@article{Sharma2026,
author = {Sharma, Ratnesh Kumar and Bharathy, Gnana and Karimi, Faezeh and Sharma, Akhilesh Kumar and Prasad, Mukesh},
title = {From Data to Themes: Aspect Mining Meets Agentic AI for Financial Metadata Discovery},
journal = {International Journal of Intelligent Systems},
year = {2026},
volume = {2026},
number = {1},
doi = {10.1155/int/6821390}
}
EUVIMED