Yunshan Ma

News archive

Updates & milestones.

Research announcements, awards, and academic activities.

One paper accepted by Multimedia Modeling (MMM) 2027 of using Language Modeling for Bundle Construction!
BundleLM: Bundle Construction Using Language Model.[pdf come soon]
One paper gets accepted by TOIS of adapting MLLM for session recommendation, one paper gets accepted by SIGIR-AP of building a multimodal fashion understanding benchmark FashionStylist!
Distilling Transitional Patterns to Multimodal Large Language Models for Session-based Recommendation.[pdf]
FashionStylist: An Expert Knowledge-enhanced Multimodal Dataset for Fashion Understanding.[pdf] [dataset and code]
One paper gets accepted by CIKM on tokenization of GenRec, one paper gets accepted by EMNLP of reusing LLM reasoning for financial report generation!
Personalized Context-Conditioned Tokenization for Generative Recommendation.[pdf] [code]
AnalysisBank: An Expert Analysis Pattern Library for Financial Report Generation.[pdf] [code]
One paper gets accepted by TKDE about continual pre-training for recommendation, one demo paper gets accepted by ACMMM of applying visual analogy retrieval to financial K-line analysis!
Large Language Model Empowered Recommendation Meets All-domain Continual Pre-Training.[pdf]
FinAlogy: A Visual Analogy Retrieval System for Financial K-Line Analysis.[pdf coming soon][code] [video]
Our paper, Frozen LVLMs for Micro-Video Recommendation: A Systematic Study of Feature Extraction and Fusion, receives the best paper award of ICMR 2026!
One paper gets accepted by Transactions on Information Systems about using LLMs for personalized fashion outfit recommendation!
CFALR: Collaborative Filtering-Augmented Large Language Model for Personalized Fashion Outfit Recommendation.[pdf]
One paper accepted by KDD'26 about multi-behabior recommendation!
Dynamic Spectral Denoising with Global-Context Attention for Multi-Behavior Recommendation. [pdf][code]
One paper gets accepted by Transactions on Information Systems about promoting long-tail items via product bundling!
Promoting Long-tail Items through Debiased Product Bundling.[pdf]
One paper gets accepted by ICMR'26 about video LLM for micro-video recommendation, congratulations to Huatuan!
Frozen LVLMs for Micro-Video Recommendation: A Systematic Study of Feature Extraction and Fusion.[pdf] [code]
Two papers are accepted by ACL'26! One accepted by the main conference about document QA, and another is accepted as findings on open-ended event forecasting. Congruatulations to Yixin and Chang He!
MAB-DQA: Addressing Query Aspect Importance in Document Question Answering with Multi-Armed Bandits.[pdf] [code]
Scattered Hypothesis Generation for Open-Ended Event Forecasting.[pdf] [code]
One paper gets accepted by Neurocomputing about tackling the heterogeneity and heterophily issues of graphs, congratulations to Kangkang!
Addressing graph heterogeneity and heterophily from a spectral perspective.[pdf][code]
Two papers are accepted by SIGIR'26! One on the research track about fashion outfit image generation, and another on the resource track about travel itinerary modification. Congruatulations to Mingzhe and Zhuoxuan!
Dual-Diffusional Generative Fashion Recommendation.[pdf coming soon]
iTIMO: An LLM-empowered Synthesis Dataset for Travel Itinerary Modification.[pdf][dataset and code]
One paper gets accepted by Transactions on Information Systems about using graph RAG for website owner identification, congratulations to Tu Cheng!
MetaRAG: Identifying Website Owner using Meta-Path-Guided Dynamic Graph Retrieval-Augmented Generation.[pdf]
One paper gets accepted by Knowledge-Based Systems about constructing attack graph using multimodal data, congratulations to Yongheng!
MM-AttacKG: A Multimodal Approach to Attack Graph Construction with Large Language Models.[pdf]
Our paper, Integrating Symbolic and Waveform Music into Large Language Models, receives the best paper award of Multimdeia Modeling (MMM) 2026!
Two papers are accepted by ICLR'26! On the topics of product bundling and stock forecasting, respectively. Congruatulations to Tu Teng, Li Ai, Kelvin, and Jan!
Discrete Diffusion for Bundle Construction.[pdf][code]
Reasoning on Time-Series for Financial Technical Analysis.[pdf][code]
One short paper, about geo-political event forecasting using LLM agents, gets accepted by The Web Conference (WWW) 2026, congratulations to Haoxuan and Chang He!
ThinkTank-ME: A Multi-Expert Framework for Middle East Event Forecasting.[pdf][code]
Our demo paper, VibeMus: Proactive Agentic System for Music Personalization, receives the best demo paper award of ACM Multimdeia Asia 2025!
Our paper, Reasoning on Time-Series for Financial Analysis, wins the best paper award of the Workshop on Rethinking Financial Time-Series in ACM ICAIF 2025.
One paper accepted by AAAI 2026, congratulations to Jiajie!
Potent but Stealthy: Rethink Profile Pollution against Sequential Recommendation via Bi-level Constrained Reinforcement Paradigm.[pdf][code]
One demo paper accepted by AAAI 2025, big congratulations to Jared Chan! Notably, Jared is a year-2 undergradute student in SMU SCIS, and this work was done during a part-time summer research project.
PortfolioPilot: An Agentic Platform for Financial Portfolio Management Algorithm Development and Evaluation.[pdf][code]
One paper accepted by Multimedia Modeling (MMM) 2025, congratulations to Tu Teng and Xiaohao!
Integrating Symbolic and Waveform Music into Large Language Models.[pdf][website]
One paper accepted by The 2nd Workshop on LLMs and Generative AI for Finance (co-host on ICAIF'25), congratulations to Kelvin!
Temporal Relational Reasoning of Large Language Models for Detecting Stock Portfolio Crashes.[pdf]
One paper accepted by the Workshop on Rethinking Financial Time-Series, congratulations to Kelvin!
Reasoning on Time-Series for Financial Analysis. [pdf][code]
One demo paper accepted by ACM Multimedia Asia 2025, congratulations to Zhiliang and Tu Teng!
VibeMus: Proactive Agentic System for Music Personalization.[pdf][code]
One paper accepted by TKDD, congratulations to Tu Cheng!
Website Owner Identification through Multi-level Contrastive Representation Learning.[pdf][code]
Our workshop proposal, titled as "International Workshop on Multimodal Generative Search and Recommendation", gets accepted by CIKM 2025. We look forward to submissions relevant to broad topics on Multimodal Generative Search and Recommendation. More details can be seen on the website.
Our demo gets accepted by ACM Multimedia'25 Demo Track, congratulations to Jin Peng, Yilin, and Mingzhe!
GenWardrobe: A Fully Generative System for Travel Fashion Wardrobe Construction.[pdf][code]
One paper accepted by KDD'25, congratulations to Yingzhi and Xiaohao!
LLM2Rec: Large Language Models Are Powerful Embedding Models for Sequential Recommendation.[pdf][code]
One paper accepted by SIGIR'25, congratulations to Mingzhe!
FashionDPO:Fine-tune Fashion Outfit Generation Model using Direct Preference Optimization.[pdf][code]
Our ICLR'25 paper, "Alphaedit: Null-space constrained model editing for language models", wins the outstanding paper award!
Our workshop proposal, titled as "FinIR 2025: The 2nd Workshop on Financial Information Retrieval in the Era of Generative AI", gets accepted by SIGIR 2025. We look forward to submissions relevant to broad topics on Finance and Information Retrieval. More details can be seen on the website.