Haonan Wang

Haonan Wang

Johns Hopkins University

  • Currently Now:
    I am currently a graduate research assistant volunteer at the Johns Hopkins University, working withProf. Ziang Xiao and Postdoc Fellow. Jie Gao and I have also actively engaged in NLP discussions with Prof. Jason Eisner(Faculty advisor) at the Center for Language and Speech Processing (CLSP). At the same time, I received my Master’s degree in Computer Science Engineering at Johns Hopkins University (JHU) in the 2025 Spring.

  • In my Undergraduate:
    Before joining JHU, I received my B.S. degree in Information and Computer Science, and Mathematics from Liaoning Technology University. During my undergrad study, I spent time at the Institute of Mathematics and Systems Science and the Institute of Machine Learning and Data Mining, where I was advised by Prof. Wei Liu and Prof.Yu Zhang works in Multivariate Statistical Analysis, Machine Learning, Reinforcement Learning, Informatics Theory, Numerical Analysis, and Math Modeling and Optimization Algorithms.
  • Another activity:
    I am also a Co-founder of SpringTeng AI, a startup and community based in the China National University Science Park. This community focuses on developing AI data analysis systems, Software Patents Apps for industry technologies, and collaborating with companies such as EACON Driverless and the Chinese Ming Group.

🤔 Research interest

My research interest focuses on the Natural Language Processing, Human-Centered Artificial Intelligence (HCI-AI), and Machine learning. My work is motivated by fundamental questions RQ1: How to explore the human-centered Social Value Alignment for Computational Social Science datasets, with the goal of understanding and simulating multi-perpective diversity of human behavior** and RQ2: How to build text-world modeling for evaluating and enhancing the human-interactive reasoning and simulation ability of LLMs Agent Specifically, including:

  • 1. Human-Centered Social Value Alignment for Computational Social Science in Subjective coding: How to achieve the bidirectional Human-AI Alignment to bridge the socio-technical gap in building LLMs for an effective data annotation/collection pipeline and LLMs for mining/analysis more enriched information in NLP subjective data tasks.
  • 2. Advancing LLMs and Agents for Dynamic World Simulation and Reasoning: How to evaluate and improve the capabilities of LLMs and agents to understand and interact with humans in the interaction process, like simulating the text-based world.
  • 3. Machine learning: How to understand data statistics dynamics using math modeling and developing efficient machine learning algorithms for large-scale data analysis and intelligent pattern recognition.

🔥 News

  • 2024.12: 🎉 Graduate with my Master in Computer Science degree and become a Student Research Assistant with ISLE lab at Johns Hopkins University and looking for a good Ph.D opportunity in 2026.

🤔 Ongoing Projects

👇 Select Publications

📕Large Language Models(LLMs) and Human-Centered AI

TBD
sym

PerspectiveCoder-LM: A LLM-based Multi-agent System for Large-scale Corpus Inductive Text Coding Analysis
Haonan Wang,Jie Gao, Kristina Gligorić, Ziang Xiao.

Code

EMNLP 2025
sym

From Noise to Nuance: Enriching Subjective Data Interpretation through Qualitative Analysis
Ruyuan Wan,Haonan Wang,Ting-Hao Kenneth Huang,Jie Gao.

The 4th HCI + NLP Workshop at EMNLP, 2025.

  • Abstract: Subjective data annotation (SDA) in NLP should treat annotator disagreement as a valuable source of insight rather than noise. Drawing on qualitative data analysis (QDA), we compare the two methodologies and highlight differences in human roles, workflows, and evaluation. We propose five recommendations to make SDA more interpretive and demonstrate these ideas through a reinforcement learning from human feedback (RLHF) case study, advocating for a more interdisciplinary approach to understanding subjective data.
EMNLP 2025
sym

ByteSized32Refactored: Towards an Extensible Interactive Text Games Corpus for LLM World Modeling and Evaluation
Haonan Wang, Junfeng Sun, Xindi Yuan, Ruoyao Wang, Ziang Xiao.

The 5th WordPlay Workshop at EMNLP, 2025. Code Slides Poster-recording

  • Abstract: ByteSized32Refactored presents a modular and extensible redesign of the original ByteSized32 interactive text games corpus to advance research on LLM world modeling and evaluation. By introducing a unified foundation library that abstracts common game logic into reusable base classes, the total codebase is reduced by half while enabling easier expansion to new environments. Experiments with GPT-4o and GPT-5 demonstrate that the refactored structure improves code quality and extensibility while revealing new challenges for hierarchical reasoning in LLMs. This work establishes a scalable platform for studying interactive reasoning, adaptability, and generalization in language models.
Plos one
sym

Unsupervised Feature Selection Algorithm Based on L2,p-norm Feature Reconstruction
Wei liu, Miao Zhong, Guangwei Liu, Haonan Wang, Qian Ning.

Plos one. Code

  • Abstract: NFRFS (an unsupervised feature selection algorithm based on L2, p-norm feature reconstruction) proposes a more robust feature selection method, aiming to address the problem that traditional algorithms are sensitive to noise and outliers. The L2, p-norm is used to enhance robustness, the integrative adaptive graph learning is employed to maintain the local structure of the data, and the inner product sparse regularization is utilized to select lower-level reference features. Experimental results on 14 benchmark datasets show that NRFFS significantly matches the existing state-of-the-art methods in Bluetooth performance, demonstrating its effectiveness and practicality in high-dimensional data processing.
Plos one
sym

IM3HRL: Model-assisted Intrinsically Motivated Modular Hierarchical Reinforcement Learning
Wei Liu, Jiaxiang Wang, Guangwei Liu, Haonan Wang.

under review.

  • Abstract: IM3HRL (Model-assisted Internal Reinforcement Hierarchical Reinforcement Learning) is an efficient and robust new reinforcement learning framework specifically designed to address the exploration and sample efficiency issues in complex goal-conditioned tasks. It hierarchically decomposes tasks by combining model-assisted internal guidance (MHA), uses internal guidance (CLP) to direct the agent to prioritize exploration, and employs a model-assisted relabeling strategy (FGRS) to generate reinforcement targets and enhance learning efficiency. Experiments have shown that the learning speed of IM3HRL is at least 15% faster than that of a single reinforcement method, and it demonstrates strong robustness against forgetting and perturbations.

📝 All Publications

  • [Dissertation-Project] – Master in Computer Science “Unveiling Statistical Relationships Among Popular LLM Benchmarks: A Quantitative Framework”
    Haonan Wang,Ziang Xiao.
    Johns Hopkins University,2024.[pdf],

  • [Dissertation-Project] – B.S in Information, Computer Science, and Math “Research on the application of human action recognition based on LSTM-CNN”
    Haonan Wang,Wei Liu.
    Liaoning Technology University,2023.[pdf],

  • [Journal] – Advances in AI and ML “Research on geometric figure classification algorithm based on Deep Learning”
    Ruiyang Wang,Haonan Wang,Junfeng Sun,Mingjia Zhao,Meng Liu.
    Advances in Artificial Intelligence and Machine Learning, 2022. [pdf],

  • [Journal] – Scientific Journal “Research status and future prospects of machine learning algorithm in big data analysis”
    Haonan Wang.
    Scientific Journal of Intelligent Systems Research,2021. [pdf],

🖥 Software

Software Patents :

—🙏Thanks to the software application development collaboration with all Professors and Graduate Assistants from the SpringTeng AI and Liaoning Technical University, Liaoning University, and Qinghua University, being funded through our collaboration with Chinese central State-owned Shenhua Ming Group Ltd and Zijin Ming Group Ltd enterprises.
[P.1] Haonan Wang, Mingjia, Zhao, et al. (2022). Artificial intelligence robot programming interactive control system. PRC Software Copyright Patent, Patent No. 2022SR1053901.
[P.2] Haonan Wang, Junfeng, Sun, et al. (2022). Image recognition processing operation platform. PRC Software Copyright Patent, Patent No. 2022SR1052419.
[P.3] Haonan Wang, Chang, Liu, et al. (2022). Artificial Intelligence Community Security Equipment Monitoring System. PRC Software Copyright Patent, Patent No. 2022SR1052492.
[P.4] Haonan Wang, Meng, Liu, et al. (2022). A network behavior analysis system based on machine learning. PRC Software Copyright Patent, Patent No. 2022SR1049807.
[P.5] Haonan Wang, Chi, Li, et al. (2022). Autonomous Driving Intelligent Dispatching Center Management System. PRC Software Copyright Patent, Patent No. 2022SR1052526.
[P.6] Haonan Wang, Ruiyang, Wang, et al. (2022). Unmanned shortest path planning system. PRC Software Copyright Patent, Patent No.2022SR0935020.
[P.7] Haonan Wang, Junfeng Sun, et al. (2022). Data operation analysis and collection system based on machine learning. PRC Software Copyright Patent, Patent No. 2022SR1052428.
[P.8] Jiawei, Zhang, Pengyu, Cai, Haonan Wang, et al. (2021). Staff check-in face recognition system.PRC Software Copyright. PRC Software Copyright Patent, Patent No. 2021SR0699354.

Competition Awards(Mathematical Modeling and Computer Science Design):

• 1st Place, 12th Mathor Cup College Mathematical Modeling Challenge 2022
National-level award in China
• 1st Place, Liaoning Mathematical Modeling Contest 2022
Provincial-level award in China
• 1st Place, 7th Shuwei Mathematical Modeling Challenge for College Students 2022
National-level award in China
• 1st Place, 12th Mathor Cup College Mathematical Modeling Challenge 2022
National-level award in China
• 3rd Place, Liaoning Province “Shuo Ri Cup” College Student Computer Design 2022
Provincial-level award in China
• 3rd Place, Northeast Three Provinces Mathematical Modeling Competition 2022
Provincial-level award in China
• 2nd Place, American Mathematical Contest in Modeling 2022
International award
• 3rd Place, 14th National Undergraduate Computer Design Competition 2021
National-level award in China
• 3rd Place, 11th Mathor Cup University Mathematical Modeling Challenge 2021
National-level award in China
• 2nd Place, National College Students “Hua Shu Cup” Mathematical Modeling 2021
National-level award in China
• 1st Place, Liaoning Province “Shuo Ri Cup” College Student Computer Design 2021
Provincial-level award in China,
• 1st Place, Liaoning AgricuLNTUral Economic Modeling Competition 2021
Provincial-level award in China
• 1st Place, Outstanding Scholarship of the Faculty of Science, LNTU 2021
School-level award in China
• 1st Place, Career Planning Competition of the Faculty of Science, LNTU 2021
School-level award in China
• Academic Achievement Award of the School of Science, LNTU 2021
School-level award in China
• 2nd Place, Liaoning Mathematical Modeling Contest 2021
Provincial-level award for China

🎖 Honors and Awards

  • Oct 2022 China National Scholarship (Top 1%)
    Ministry of Education of the People’s Republic of China
    Awarded the highly competitive China National Scholarship, granted to the top 1% of students nationwide in recognition of outstanding academic performance, research excellence, and comprehensive personal development
  • May 2023 Outstanding Student Scholarship, Special Prize(Top 1%)
    Liaoning Province Governiment
  • May 2019,2020,2021,2022 Outstanding Student Scholarship, First Prize(Top 10%)
    Liaoning Technology University

📖 Educations

  • 2023.08 - 2025.1, Master degree, Master of Science in Engineering in Computer Science (MSE), Johns Hopkins University, Baltimore, Maryland, USA.
    • Johns Hopkins University
  • 2019.09 - 2023.07, B.S. degree, Information and Computer Science, and Mathematics, Liaoning Technical University, Fuxin, Liaoning, China.
    • Liaoning Technical University

💻 Internships

🤝 Collaborators

I deeply appreciate and am grateful to my collaborators and institutions for their continuous support and opportunities to grow in both academic and professional domains.