MixFormer: Co-Scaling Up Dense and Sequence in Industrial Recommenders

Xu Huang, Hao Zhang, Zhifang Fan, Yunwen Huang, Zhuoxing Wei, Zheng Chai, Jinan Ni, Yuchao Zheng, Qiwei Chen

Project

• MixFormer is a unified Transformer architecture for recommender systems that jointly models sequential behaviors and feature interactions within a single backbone.
• It enables effective co-scaling of dense capacity and sequence length, with a user-item decoupling strategy for efficiency.
• Large-scale online A/B tests on Douyin and Douyin Lite show consistent improvements in user engagement metrics including active days and in-app usage duration.

HyperGate: Hierarchical Perceptive Gating Network for Multi-domain Multi-task Recommendation

Xu Huang*, Xiaolong Chen*, Yichao Wang, Weiwen Liu, Yang Yang, Xingmei Wang, Defu Lian and Ruiming Tang

Project

• A hierarchical gating network for multi-domain and multi-task recommendation.
• We propose a contrastive domain and task representation augumentation module to extract domain and task embeddings, and a hierarchical gating network to construct a domain- and task-perceptive parameter-sharing network from the bottom up.
• The paper will be public soon.

ToolACE: Winning the Points of LLM Function Calling

Weiwen Liu*, Xu Huang*, Xingshan Zeng*, Xinlong Hao, Shuai Yu, Dexun Li, Shuai Wang, Weinan Gan, et al.

Project

• A data synthetic framework tailored for enhancing tool-using ability of LLMs.
• Model; Dataset

WESE: Weak Exploration to Strong Exploitation for LLM Agents

Xu Huang, Weiwen Liu, Xiaolong Chen, Xingmei Wang, Defu Lian, Yasheng Wang, Ruiming Tang, Enhong Chen

Project

• A novel prompt-based method for LLM agents, leveraging weaker agent for exploration and stronger agent for exploitation.

Understanding the planning of LLM agents: A survey

Xu Huang, Weiwen Liu, Xiaolong Chen, Xingmei Wang, Hao Wang, Defu Lian, Yasheng Wang, Ruiming Tang, Enhong Chen

Project

• A survey about the planning ability of LLM agents.

A Data-Centric Multi-Objective Learning Framework for Responsible Recommendation Systems

Xu Huang, Jianxun Lian, Hao Wang, Defu Lian, Xing Xie

Project

• A data-centric framework for multi-objective learning in recommendation systems.
• Code

Recommender AI Agent: Integrating Large Language Models for Interactive Recommendations

Xu Huang, Jianxun Lian, Yuxuan Lei, Jing Yao, Defu Lian, Xing Xie

Project

• A framework to build an interactive recommendation agent with LLM
• Code

RecStudio: Towards a Highly-Modularized Recommender System

Defu Lian(mentor), Xu Huang, Xiaolong Chen, Jin Chen, Xingmei Wang, Yankai Wang, Haoran Jin, Rui Fan, Zheng Liu, Le Wu, Enhong Chen

Project

• A modularized recommender system library RecStudio was developped and released to public
• Project homepage

Cooperative Retriever and Ranker in Deep Recommenders

Xu Huang, Defu Lian, Jin Chen, Zheng Liu, Xing Xie, Enhong Chen

Project

• An adaptive hard negative sampler was introduced to enhance the ranker
• An sampling based KL divergence was proposed to enhance the retriever

RecExplainer: Aligning Large Language Models for Explaining Recommendation Models

Yuxuan Lei, Jianxun Lian, Jing Yao, Xu Huang, Defu Lian, Xing Xie

Project

• A novel approach to leverage LLMs as surrogate models for explaining black-box recommender models.

Learning-Efficient Yet Generalizable Collaborative Filtering for Item Recommendation

Yuanhao Pu, Xiaolong Chen, Xu Huang, Jin Chen, Defu Lian, Enhong Chen

Project

• This work proposes a new square loss RG2 for recommendation based on the approximation of the softmax loss with Taylor expansion.
• We have studied the theoretical properties of the proposed loss in terms of generalization and consistency.

RecAI: Leveraging Large Language Models for Next-GenerationRecommender Systems

Jianxun Lian, Yuxuan Lei, Xu Huang, Jing Yao, Wei Xu, Xing Xie

Project

• A practical toolkit designed to augmentor even revolutionize recommender systems with the advanced capabilities of Large Language Models.
• Demo Paper

When large language models meet personalization: Perspectives of challenges and opportunities

Jin Chen, Zheng Liu, Xu Huang, Chenwang Wu, Qi Liu, Gangwei Jiang, Yuanhao Pu, Yuxuan Lei, Xiaolong Chen, Xingmei Wang, Defu Lian, Enhong Chen

Project

• A survey to summarize the combination of large language models and personlization systems

Fast variational autoencoder with inverted multi-index for collaborative filtering

Jin Chen, Defu Lian, Binbin Jin, Xu Huang, Kai Zhang, Enhong Chen

Project

• An adaptive negative sampler was proposed to training VAE efficiently