This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
LATTE: Learning Aligned Transactions and Textual Embeddings for Bank Clients
Created by
Haebom
Author
Egor Fadeev, Dzhambulat Mollaev, Aleksei Shestov, Dima Korolev, Omar Zoloev, Ivan Kireev, Andrey Savchenko, Maksim Makarenko
Outline
This paper proposes LATTE, a novel framework for learning customer embeddings in financial applications. Existing large-scale language models (LLMs) suffer from computational overhead in processing long-term event sequences and are difficult to apply to real-world pipelines. LATTE is a contrastive learning framework that aligns raw event embeddings with semantic embeddings extracted from frozen LLMs. It summarizes action features into short prompts, embeds them using LLMs, and performs supervised learning using contrastive loss. This significantly reduces inference costs and input size compared to full-sequence processing using LLMs. Experimental results using real-world financial datasets demonstrate that LATTE outperforms existing state-of-the-art techniques and is deployable even in latency-sensitive environments.
Takeaways, Limitations
•
Takeaways:
◦
We present a novel customer embedding learning method for financial applications that effectively addresses computational cost and latency issues while leveraging LLM.
◦
Validated performance over existing state-of-the-art techniques on real-world financial datasets.
◦
Provides a practical model that can be deployed even in latency-sensitive environments.
•
Limitations:
◦
The performance of the proposed model may be limited to specific financial datasets.
◦
May depend on the performance of the LLM used.
◦
Performance may be affected by the quality of prompt engineering.
◦
Further research is needed on generalization performance across different types of financial events.
