研究方向与发表论文

Integrating listwise ranking into pairwise-based image-text retrieval

Zheng Li, Caili Guo, Xin Wang, Zerun Feng, Yanjun Wang

Knowledge-Based Systems, 287: 111431

⭐ SCI检索 · 中科院一区 · 影响因子: 8.8

Image-Text Retrieval (ITR) is essentially a ranking problem. Existing pairwise-based ITR models increase positive pair similarity while decreasing negative pair similarity indiscriminately, ignoring the relevance difference between dissimilar negative pairs. This paper integrates listwise ranking into pairwise-based ITR: a Relevance Score Calculation (RSC) module computes the relevance score of the entire ranked list, and the ranking metric NDCG is transformed into a differentiable listwise loss named Smooth-NDCG (S-NDCG). The approach is plug-and-play for current pairwise-based ITR models, improving performance on ITR benchmarks and providing more user-friendly retrieval results.

基于深度学习的智能问答系统研究

张三, 李四, 王五

计算机学报

本文提出了一种基于深度学习的智能问答系统，通过Transformer模型和注意力机制，实现了对复杂问题的理解和回答。

大数据环境下的实时数据分析框架

张三, 赵六

软件学报

针对大数据环境下的实时数据分析需求，本文设计并实现了一个基于流处理技术的数据分析框架。

Cloud-Native Microservices Architecture for Enterprise Applications

San Zhang, Si Li

IEEE International Conference on Cloud Computing

This paper presents a cloud-native microservices architecture designed for enterprise applications, focusing on scalability and maintainability.