Qifu Technology's Paper Accepted by IJCAI 2025, Using MLLM to Pave New Path in Fintech

Rhea-AI Summary

Qifu Technology (NASDAQ: QFIN) and Beijing Jiaotong University have achieved a significant milestone with their research paper's acceptance at IJCAI 2025, a prestigious AI conference with a 19.3% acceptance rate. The paper introduces TRIDENT, an innovative framework that combines Multimodal Large Language Model (MLLM) embeddings and attribute smoothing for Compositional Zero-Shot Learning. The technology shows promising applications in fintech, particularly in: • Intelligent risk control: Analyzing multimodal data to detect fraud patterns more efficiently • Customer service: Enabling better understanding of complex user inquiries The framework addresses key challenges in AI, including background interference and limited semantic capture, demonstrating state-of-the-art performance across multiple datasets. This achievement reflects Qifu Technology's commitment to AI innovation and its strategic focus on R&D investment and academic collaborations.

Insights

Technology Innovation Expert

Qifu's AI research breakthrough shows innovation potential but offers uncertain timeline for commercial implementation in their fintech operations.

The acceptance of Qifu's research paper at IJCAI 2025 represents a significant technical achievement in a highly competitive field, with only 19.3% of submissions making the cut. The TRIDENT framework they've developed tackles a fundamental AI challenge: identifying novel combinations of attributes and objects using existing knowledge.

The technical innovations are particularly relevant to fintech applications. By combining Multimodal Large Language Model embeddings with specialized modules like Feature Adaptive Aggregation, TRIDENT can filter background noise and extract detailed features from complex data—capabilities that translate directly to two critical fintech operations:

• Enhanced fraud detection through analysis of multimodal data (transaction patterns combined with user profiles)
• More precise understanding of complex customer inquiries, enabling better service automation

What makes this development noteworthy is how it addresses specific limitations in current AI systems: background interference, limited semantic capture, and overconfidence in familiar patterns. These improvements could allow Qifu to identify emerging fraud patterns that traditional models might miss.

However, the path from academic research to deployed systems involves substantial engineering work. The press release doesn't specify implementation timelines or quantify expected operational improvements, creating uncertainty about when these innovations might impact business performance.

SHANGHAI, May 7, 2025 /PRNewswire/ -- China's leading fintech company Qifu Technology ( NASDAQ: QFIN; HKEX: 3660) and Beijing Jiaotong University have jointly achieved a significant milestone as their paper, Leveraging MLLM Embeddings and Attribute Smoothing for Compositional Zero-Shot Learning, has been accepted by the International Joint Conference on Artificial Intelligence（IJCAI）2025, a premier international conference in artificial intelligence.

With an acceptance rate of merely 19.3% — selecting 1,042 out of 5,404 submissions — IJCAI is recognized as one of the most prestigious AI conferences globally and is listed as a Class-A event by the China Computer Federation （CCF）.

The research focuses on Compositional Zero-Shot Learning, aiming to identify novel combinations of attributes and objects using existing knowledge. Overcoming traditional challenges such as background interference, limited semantic capture in word embeddings, and overconfidence in known combinations, the team introduced an innovative framework named TRIDENT. By integrating Multimodal Large Language Model (MLLM) embeddings and attribute smoothing, TRIDENT utilizes modules like Feature Adaptive Aggregation (FAA) to mitigate background noise, learns conditional masks for detailed feature extraction, and leverages MLLM's hidden states to enhance semantic representation. This approach has demonstrated state-of-the-art performance across multiple datasets, offering new solutions for fields including image recognition and content understanding.

In fintech area, TRIDENT shows great potential. In intelligent risk control, it analyzes multimodal data—such as transaction behaviors and user profiles—to detect emerging fraud patterns faster than traditional models, improving assessment accuracy and reducing losses. In customer service, the framework enables more precise understanding of complex user inquiries, providing personalized and efficient support.

This achievement underscores Qifu Technology's commitment to AI innovation. By increasing R&D investment and deepening collaborations with academic institutions, the company aims to drive further advancements in AI applications, contributing to industry growth and social progress.

View original content:https://www.prnewswire.com/news-releases/qifu-technologys-paper-accepted-by-ijcai-2025-using-mllm-to-pave-new-path-in-fintech-302449711.html

SOURCE Qifutech

FAQ

What is the significance of QFIN's IJCAI 2025 paper acceptance?

The acceptance of QFIN's paper at IJCAI 2025 is significant as it was among only 19.3% of submissions accepted (1,042 out of 5,404) at this prestigious Class-A AI conference, demonstrating the company's AI research capabilities.

How does Qifu Technology's TRIDENT framework improve fintech operations?

TRIDENT improves fintech operations by enhancing fraud detection through multimodal data analysis and enabling more precise understanding of customer inquiries for better service delivery.

What are the main features of QFIN's TRIDENT AI framework?

TRIDENT features MLLM embeddings, attribute smoothing, Feature Adaptive Aggregation for background noise reduction, and conditional masks for detailed feature extraction, improving semantic representation in AI applications.

What is the collaboration between Qifu Technology and Beijing Jiaotong University?

Qifu Technology and Beijing Jiaotong University collaborated on developing the TRIDENT framework for Compositional Zero-Shot Learning, resulting in a paper accepted at IJCAI 2025.