Dan Zhang

I am a third-year phd candidate in the Knowledge Engineering Group (KEG) at Tsinghua, supervised by Jie Tang. I recieved my Master’s Degree from School of Software, Tsinghua University in 2021, advised by Ping Luo. My research focus on scientific language model and graph representaion learning. I am now a visiting student researcher at Caltech in the Computing + Mathematical Sciences (CMS) Department, currently working with Yisong Yue.

张丹  /  zd18 [MASK] tsinghua [MASK] org [MASK] cn  /  Bio  /  Google Scholar  /  GitHub

profile photo
Honors and Awards
ReST-MCTS*: LLM Self-Training via Process Reward Guided Tree Search
Dan Zhang*, Sining Zhoubian*, Yisong Yue, Yuxiao Dong, and Jie Tang
In submission, 2024
arXiv / code / model / policy data / PRM data

ReST-MCTS* is a reinforced self-training approach, based on integrating process reward guidance with tree search MCTS* for collecting higher-quality reasoning traces as well as per-step value to train policy and reward models.

SciGLM: Training Scientific Language Models with Self-Reflective Instruction Annotation and Tuning
Dan Zhang, Ziniu Hu, Sining Zhoubian, Zhengxiao Du, Kaiyu Yang, Zihan Wang, Yisong Yue, Yuxiao Dong, and Jie Tang
In submission, 2024
arXiv / code & data / model

SciGLM is a suite of scientific language models able to conduct college-level scientific reasoning. Central to our approach is a novel self-reflective instruction annotation framework to address the data scarcity challenge in the science domain. Applying this framework, we curated SciInstruct, a diverse and high-quality dataset encompassing physics, chemistry, math, and formal proofs.

AutoRE: Document-Level Relation Extraction with Large Language Models
Lilong Xue*, Dan Zhang*, Yuxiao Dong, and Jie Tang
ACL, 2024
arXiv / code / platform

AutoRE is an end-to-end DocRE model that adopts a novel RE extraction paradigm named RHF (Relation-Head-Facts). AutoRE does not rely on the assumption of known relation options, making it more reflective of real-world scenarios.

OAG-Bench: A Human-Curated Benchmark for Academic Graph Mining
Fanjin Zhang, Shijie Shi, Yifan Zhu, Bo Chen, Yukuo Cen, Jifan Yu, Yelin Chen, Lulu Wang, Qingfei Zhao, Yuqing Cheng, Tianyi Han, Yuwei An, Dan Zhang, Weng Lam Tam, Kun Cao, Yunhe Pang, Xinyu Guan, Huihui Yuan, Jian Song, Xiaoyan Li, Yuxiao Dong, and Jie Tang
KDD, 2024
arXiv / code & data / OAG-Challenge @ KDD Cup 2024

We present OAG-Bench, a comprehensive, multi-aspect, and fine-grained human-curated benchmark based on the Open Academic Graph (OAG). OAG-Bench covers 10 tasks, 20 datasets, 70+ baselines, and 120+ experimental results to date.

RecDCL: Dual Contrastive Learning for Recommendation
Dan Zhang, Yangliao Geng, Wenwen Gong, Zhongang Qi, Zhiyu Chen, Xing Tang, Ying Shan, Yuxiao Dong, and Jie Tang
WWW, 2024, Oral
arXiv / code & data / slides_pdf

RecDCL is a dual contrastive learning recommendation framework. In this work, we investigate how to employ both batch-wise CL (BCL) and feature-wise CL (FCL) for recommendation. We theoretically analyze the relation between BCL and FCL, and find that combining BCL and FCL helps eliminate redundant solutions but never misses an optimal solution.

ApeGNN: Node-Wise Adaptive Aggregation in GNNs for Recommendation
Dan Zhang, Yifan Zhu, Yuxiao Dong, Yuandong Wang, Wenzheng Feng, Evgeny Kharlamov, and Jie Tang
WWW, 2023
PDF / code & data / slides_pdf

ApeGNN develops a node-wise adaptive diffusion mechanism for information aggregation, in which each node is enabled to adaptively decide its diffusion weights based on the local structure (e.g., degree).

DropConn: Dropout Connection Based Random GNNs for Molecular Property Prediction
Dan Zhang, Wenzheng Feng, Yuandong Wang, Zhongang Qi, Ying Shan, and Jie Tang
TKDE, 2024
PDF / code & data

DropConn is an adaptive data augmentation strategy, which better leverages edge features and assigns weights for chemical bonds to emphasize their importance, and generates robust representations for molecule graph property prediction.

WinGNN: Dynamic Graph Neural Networks with Random Gradient Aggregation Window
Yifan Zhu, Fangpeng Cong, Dan Zhang, Wenwen Gong, Qika Lin, Wenzheng Feng, Yuxiao Dong, and Jie Tang
KDD, 2023
PDF / code & data

WinGNN models dynamic graphs by combining a simple graph neural network model with meta-learning strategies and implementing a time-encoder-free dynamic graph coding process through a stochastic gradient aggregation mechanism.

Detecting Social Bot on the Fly using Contrastive Learning
Ming Zhou, Dan Zhang, Yuandong Wang, Yangliao Geng, and Jie Tang
CIKM, 2023

CBD is characterized by a two-stage model learning strategy: a contrastive pre-training stage to mine generalization patterns from massive unlabeled social graphs, followed by a semi-supervised fine-tuning stage to model task-specific knowledge latent in social graphs with a few annotations.

SIRAN: Semi-Supervised Social Bot Detection with Initial Residual Relation Attention Networks
Ming Zhou, Wenzheng Feng, Yifan Zhu, Dan Zhang, Yuxiao Dong, and Jie Tang
ECML-PKDD, 2023, Best Student Paper

We analyze human-bot networks and propose SIRAN, which combines relation attention with initial residual connection to reduce and prevent the noise aggregated from neighbors to improve the capability of distinguishing different kinds of nodes on social graphs with heterophily. Then we use a consistency loss to boost the detection performance of the model for limited annotated data.

  • Teaching Assistant, Advanced Machine Learning, 2023 Fall, Tsinghua University
  • Teaching Assistant, Programing and Training, 2023 Summer, Tsinghua University

Website template credit: Jon Barron