Yueqi's Homepage
Email: yueqi (at) berkeley.edu | Tel: (+86)13260053441
[Homepage] [GitHub] [GScholar] [Twitter|X]
(Logo credits: UC Berkeley | Photo credits: Online)
Embedded Files
About Me
About Me
My name is Yueqi Wang. Welcome to my site!
My current research interests mainly include Social Computing, Multimodality, Information Retrieval and User Preference Modeling/Recommender Systems. I graduated my Master's of Arts from UC Berkeley advised by Prof. Zachary Pardos and supported by the UC Berkeley GSE Fellowship, where I mainly worked on social impact of neural nets on human cognition/learning. Past works and publications could be viewed below. Before attending UC Berkeley, I received my B.E. of Computer Science at University of Electronic Science and Technology of China (UESTC). I was also honored to receive the 2018 Outstanding Student Award of the School of Computer Science and Engineering at UESTC.
News
News
(Nov, 2024) ❀ Our work of Auto-Encoding vs. Auto-Regression for self-attentive sequential recommendation as a reality check, Your Causal Self-Attentive Recommender Hosts a Lonely Neighborhood, is accepted to ACM WSDM 2025 (Oral), see everyone in Germany, March 2025!
(Oct, 2024) ❀ Our work of fMRLRec, Train Once, Deploy Anywhere: Matryoshka Representation Learning for Multimodal Recommendation, is accepted to Findings of ACL EMNLP 2024, see everyone in Miami, November 2024!
(2024) ❀ Our work of LRURec, Linear Recurrent Units for Sequential Recommendation, is accepted to ACM WSDM 2024, see everyone in Mexico, March 2024!
Selected Projects
Selected Projects
Towards more efficient and effective LLM/Multimodality:
Current results: EMNLP 2024 Findings
We aim to develop parameter-efficient training/evaluation mechanisms for LLM/Multimodal information systems (e.g. recommendation systems) ready for deployment; Specifically with SOTA performance, we present a novel full-scale, Matryoshka-based training mechanism that allows variously sized full models (instead of learned embeddings as in traditional MRL) to be trained in a single session.
A reality check on uni/bi-directional attention for user preference modeling and recommendation
Exploring the theoretical and empirical implications of uni/bi-directional self-attention mechanisms for sequential user preference modeling and movie/product recommendation. Sparsity analysis and low-rank approximation of attention matrices are explained with extensive, empirical experiments.
Faster while more personalized sequential recommenders:
We develop parallelizable linear recurrent models that achieve both parallel training (like transformer) and incremental inference (like RNN) for sequential recommendation. We work on comparisons of such models with self-attention for user preference modeling. We also discuss models' interpretability regarding long/short arrange dependency, sparse recommendation scenarios and beyond.
Knowledge Tracing - Sequential modeling in human learning:
EDM 2022. Ongoing work.
Finding the role of (1) sequence ordering and (2) chronology modeling in neural representations of human learning signal (e.g. student test-taking data)---how well do strongly sequential-based models (e.g. LSTM) model human learning versus contextual-based models (e.g. self-attention) with relatively weak positional encoding? Models' representation, performance, and interpretability are extensively discussed.
[PDF]
Publications
Publications
Your Causal Self-Attentive Recommender Hosts a Lonely Neighborhood. (2025) [Oral]
Yueqi Wang, Zhankui He, Zhenrui Yue, Julian McAuley, Dong Wang
Proceedings of the 18th ACM International Conference on Web Search and Data Mining (WSDM).
Train Once, Deploy Anywhere: Matryoshka Representation Learning for Multimodal Recommendation. (2024)
Yueqi Wang, Zhenrui Yue, Huimin Zeng, Dong Wang, Julian McAuley
Findings of the Association for Computational Linguistics: EMNLP 2024 (EMNLP Findings).
Linear Recurrent Units for Sequential Recommendation. (2024)
Zhenrui Yue, Yueqi Wang, Zhankui He, Huimin Zeng, Julian McAuley, Dong Wang.
Proceedings of the 17th ACM International Conference on Web Search and Data Mining (WSDM).
Preference-Optimized Retrieval and Ranking for Efficient Multimodal Recommendation. (2025) [Oral]
Zhenrui Yue, Huimin Zeng, Yueqi Wang, Julian McAuley, Dong Wang.
Proceedings of the 31th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (SIGKDD).
Learning from Mistakes: Contrastive Decoding for Retrieval Augmented Generation. (Under review)
Zhenrui Yue, Huimin Zeng, Yueqi Wang, Yaokun Liu, Fengran Mo, Yang Zhang, Na Wei, Dong Wang
Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL).
Multimodal Cross-Correlation Learning for Fair Vision-Language Models in Federated Learning. (Under review)
Huimin Zeng, Zhenrui Yue, Yueqi Wang, Yang Zhang, Dong Wang
Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL).
Does Chronology Matter? Sequential vs Contextual Approaches to Knowledge Tracing. (2022)
Yueqi Wang, Zachary Pardos
Proceedings of the 15th International Conference on Educational Data Mining (EDM), 601–605.
[PDF]
BertKT: A Purely Attention-Based, Bidirectional Deep Learning Architecture for Knowledge Tracing. (2020)
Yueqi Wang, Zachary Pardos
Proceedings of the Workshop on AI-based Multimodal Analytics for Understanding Human Learning in Real-world Educational Contexts (AIMA4EDU) @ the 29th International Joint Conference on Artificial Intelligence (IJCAI).
Experiences
Experiences
University of California, San Diego | San Diego, CA | Aug 2023 - Now | (Research Intern)
With Professor Julian McAuley.
University of California, Berkeley | Berkeley, CA | Aug 2022 - Dec 2022 | Paid position
Lead Graduate Student Instructor (Lead GSI) for DATA 144/EDUC 244 Data Mining and Analytics.
University of California, Berkeley | Berkeley, CA | Jan 2022 - Aug 2022 | Paid position
Graduate Student Researcher (GSR) with Prof. Zachary Pardos, Knowledge Tracing project.
University of electronic Science and Technology of China (UESTC) | Chengdu, China | July 2020 - June 2021 | Paid position
Research Assistant
Google Sites
Report abuse