Welcome

Welcome to Haijiang LIU’s homepage! I am a PhD student focusing on Natural Language Processing with an emphasis on cross-cultural and multilingual aspects. My research interests lie at the intersection of Knowledge Graphs, Large Language Models, and Multilingual NLP.

🔬 Research Interests

• Cross-cultural NLP

  • Cultural-aware language understanding
  • Cross-cultural dialogue systems
  • Cultural bias detection and mitigation

• Knowledge Graphs

  • Multilingual knowledge representation
  • Knowledge graph construction and completion
  • Cultural knowledge integration

• Large Language Models

  • Cross-lingual transfer learning
  • Cultural adaptation in LLMs
  • Multilingual prompt engineering

• Multilingual NLP

  • Low-resource language processing
  • Cross-lingual semantic alignment
  • Language-agnostic representations

📝 Selected Publications

1. Towards realistic evaluation of cultural value alignment in large language models: Diversity enhancement for survey response simulation, Haijiang Liu, Yong Cao, Xun Wu, Chen Qiu, Jinguang Gu, Maofu Liu, Daniel Hershcovich, Information Processing & Management, Volume 62, Issue 4, 2025, 104099, ISSN 0306-4573, https://doi.org/10.1016/j.ipm.2025.104099.

   • The paper introduces a diversity-enhancement framework with a memory simulation mechanism to assess the alignment of large language models (LLMs) with cultural values. This improves the reliability of evaluations through realistic survey experiments and comprehensive metrics. Among eleven evaluated models, the Mistral and Llama-3 series demonstrate superior cultural value alignment, with Mistral-series models excelling in understanding U.S. and Chinese contexts.

2. Specializing Large Language Models to Simulate Survey Response Distributions for Global Populations. Yong Cao, Haijiang Liu, Arnav Arora, Isabelle Augenstein, Paul Röttger, Daniel Hershcovich. 2025. 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics.

   • The paper introduces a novel fine-tuning method for large language models (LLMs) to specialize them in simulating group-level survey response distributions, significantly outperforming zero-shot classifiers and other techniques on unseen questions, countries, and surveys. While challenges remain, this approach demonstrates the potential of specialized LLMs to advance accurate survey simulation, offering valuable insights for social science research.

3. Decoupled contrastive learning for multilingual multimodal medical pre-trained model, Qiyuan Li, Chen Qiu, Haijiang Liu, Jinguang Gu, Dan Luo, Neurocomputing, Volume 633, 2025, 129809, ISSN 0925-2312, https://doi.org/10.1016/j.neucom.2025.129809.

   • The paper introduces a novel multilingual multimodal medical pre-training model that addresses challenges in healthcare by expanding the MIMIC-CXR report dataset to 20 languages and developing a label disambiguation technique for improved semantic similarity and intermodality interactions. The proposed model significantly enhances medical image classification and multilingual image–text retrieval, outperforming baselines by up to 13.78% and 12.6%, respectively.

🛠️ Projects

Project 1

Constructing knowledge graphs of culture-loaded words and their demonstration application in the cross-cultural field

This project aims to solve the problem of insufficient research on the cross-cultural attributes of culture-loaded words and limited cross-cultural knowledge graph representation capabilities by constructing a multilingual knowledge graph of culture-loaded words. The research adopts a four-stage approach, including ontology construction, knowledge graph construction, fusion expansion, and educational application. Ultimately it forms a fine-grained, cross-cultural knowledge graph to help China-foreign cooperative education & cross-cultural exchanges and improve the teaching efficiency & cultural understanding ability of international students.

• Project belongs: Philosophy and Social Sciences
• Relation: Participate (Leader - Qiaoling Xiao)
• Status: in progress

Project 2

Theoretical and practical paths for Hubei provincial universities to carry out high-level applied technology exchanges and cooperation with African countries in subdivided fields

This project aims to systematically elaborate on the relevant theories of carrying out technology transfer and innovation cooperation between China and Africa to serve the country’s overall diplomatic situation; to analyze and study the effective paths for leveraging the resource advantages of more than 120 provincial universities in Hubei to carry out exchanges and cooperation between China and Africa in the field of specialized and subdivided applied technologies, and to form a “Research Report on the Theory and Implementation Paths of Hubei Provincial Universities to Carry out High-level Application Technology Exchanges and Cooperation with African Countries in Subdivided Fields”.

• Project belongs: The Soft Science Plan Research Project of Hubei Province
• Relation: Participate (Leader - Qiaoling Xiao)
• Status: finished

🎓 Education

• Ph.D. in Control Systems and Engineering (Expected 2028)

  • Wuhan University of Science and Technology
  • Advisor: Prof. Jinguang GU
  • Team: Ontoweb

• B.S. in Software Engineering (2023)

  • Wuhan University of Science and Technology

🤝 Collaborations and Service

• Reviewer
  • IP&M etc.
• Workshop Organization
  • AI+Humanities Seminar and 2024 Knowledge Organization Academic Exchange Forum
• Student Volunteer
  • DIKS 2023
  • CCF Wuhan Member Activity Center 2024 Annual Meeting

💻 Technical Skills

• Programming Languages: Python, Java, etc.
• ML/DL Frameworks: PyTorch, Transformers, etc.
• Tools & Technologies: Git, Docker, etc.
• Languages: Chinese, English, Spanish (still learning)

📫 Contact

• 📧 Email: alecliu@ontoweb.wust.edu.cn
• 📚 Google Scholar: Hajiang Liu

🔍 Current Research Focus

I am currently working on:

1. Developing culturally aware language models through cognition augmentation, finetuning, and KG reasoning.
2. Creating multicultural knowledge graphs to assist understanding and communication.
3. Exploring cross-culture LLMs applications in reality, i.e. education, conflict resolution, and media de-polarization.

Feel free to reach out if you’re interested in collaboration or have questions about my research!

---

Last updated: [2025-03-19]