AI-Ready Research Workflows in Computational Social Science: Lessons on Building a Shared Language for Interdisciplinary Collaboration
Model-Driven Engineering (MDE) enables scalable SSH workflows, reducing GPU hours from 9000 to 92 for classifying 220 million records.
Key Findings
Methodology
The approach employs MDE to create shared semantic models, abstracting HPC complexities into configurable actions and axes. Agile practices foster interdisciplinary collaboration, translating research questions into engineering specifications. The workflow automates data extraction, feature generation, and large-scale inference, validated through iterative testing on OpenAlex data. Techniques include model reduction, parallel inference, and metadata management, significantly lowering computational costs while maintaining accuracy. The framework emphasizes organizational and technical innovation, enabling sustainable AI integration in SSH.
Key Results
- Classifying 220 million documents on MareNostrum 5 reduced GPU hours from 9000 to 92, achieving a hundredfold efficiency gain.
- Parallel inference and model reduction maintained classification accuracy (>85%), with 9 billion total inferences executed.
- Shared vocabulary and automation improved interdisciplinary communication, standardizing workflows across teams.
Significance
This work addresses critical barriers in SSH AI adoption, providing a scalable, reproducible, and collaborative workflow. It advances the integration of HPC and AI, fostering data-driven social science research. The framework supports cross-institutional efforts, promoting sustainable practices and setting a new standard for large-scale SSH analysis.
Technical Contribution
Introducing a formalized, model-driven architecture that encapsulates HPC operations into reusable, configurable actions. Combining agile collaboration with automation, the approach enhances workflow modularity, reproducibility, and scalability. It bridges the gap between complex HPC environments and interdisciplinary research needs, offering a paradigm shift in SSH AI methodologies.
Novelty
First comprehensive implementation of MDE in SSH for large-scale literature classification, integrating agile workflows with semantic modeling. It uniquely abstracts HPC complexity into shared, configurable components, enabling efficient, reproducible, and collaborative research at scale.
Limitations
- Validation relies heavily on manual human evaluation, limiting full automation and scalability.
- Metadata standards and FAIR compliance are still evolving, hindering seamless data sharing.
- High hardware costs for large-scale inference remain a challenge, requiring further optimization.
Future Work
Future efforts will focus on developing automated validation pipelines, standardizing FAIR metadata, and reducing hardware costs. Cross-institutional collaborations are essential to build robust, scalable, and sustainable SSH AI workflows, ultimately enabling broader adoption and innovation.
AI Executive Summary
Artificial intelligence (AI) has begun transforming social sciences and humanities, yet significant technical barriers hinder widespread adoption. Traditional workflows struggle with the scale, complexity, and reproducibility of modern AI methods, especially on HPC infrastructure. This paper introduces a novel workflow built on Model-Driven Engineering (MDE), designed to abstract HPC complexities into shared, configurable components. By integrating agile practices, the team fostered effective interdisciplinary collaboration, translating research questions into engineering specifications. The workflow automates large-scale bibliometric analysis on the OpenAlex database, classifying 220 million records with a reduction of GPU hours from 9000 to just 92, demonstrating remarkable efficiency. This approach not only accelerates computation but also improves communication among diverse team members, establishing a scalable, reproducible framework for SSH research. The work emphasizes the importance of organizational and technical innovations, including shared vocabularies, metadata standards, and validation protocols, to ensure long-term sustainability. Despite these advances, challenges remain in automating validation, standardizing metadata, and reducing hardware costs. The authors advocate for cross-institutional efforts to develop comprehensive tooling and standards, ensuring AI-ready SSH workflows can scale sustainably. This research marks a significant step toward integrating AI into social sciences at an unprecedented scale, promising new insights and collaborative models for future research.
Deep Analysis
Background
近年来，AI技术在社会科学和人文学科中逐渐崭露头角，代表性研究包括Kozlowski等的词嵌入分析和Berganzo-Besga的考古图像识别。这些工作推动了文化变迁、社会行为分析等方向，但仍受限于数据处理效率、验证机制和跨机构协作的瓶颈。传统研究流程多为手工操作，缺乏标准化工具和自动化平台，限制了大规模AI应用的推广。随着数据量的激增和模型的复杂化，亟需构建高效、可复用的工作流体系，以实现跨学科的深度合作与创新。
Core Problem
SSH研究面临的核心问题是如何在保证科学严谨的基础上，有效处理海量异构数据，特别是在高性能计算环境中实现自动化、可复现的研究流程。现有工具多为单一任务定制，缺乏跨项目、跨机构的标准化流程，导致重复劳动和沟通障碍，严重制约了大规模AI方法的落地应用。此外，验证机制不足、元数据不统一，也影响了研究的可信度和数据的可复用性。
Innovation
本研究的创新点包括：1）引入模型驱动工程（MDE）抽象HPC操作，将复杂的流程封装为共享的行动和配置轴，提升流程的可配置性和维护性；2）结合敏捷实践，促进跨学科团队的高效协作，将研究需求转化为工程行动，形成可复用的自动化流程；3）标准化大规模文献分类的自动化流程，显著降低计算成本，增强流程的可验证性，为SSH领域提供了新型的技术范式。这些创新突破了传统单一任务定制的局限，为未来大规模、多机构合作奠定基础。
Methodology
- �� 需求转化：将研究问题（如文献分类）转化为具体工程需求，形成用户故事。• 模型构建：与领域专家合作，建立包含数据提取、特征生成、模型推断的共享语义模型。• 自动化流程：利用COMPSs等工具封装任务，支持大规模并行推理，优化资源利用。• 迭代优化：多轮测试验证模型缩减策略和推理参数，确保效率与准确性。• 跨学科协作：采用敏捷实践，定期回顾流程，调整策略。• 共享语汇：建立可配置的行动和配置轴，支持不同研究场景。• 元数据管理：设计符合FAIR原则的元数据体系，支持数据发现和复用。
Experiments
采用OpenAlex数据库，分类220百万文献，验证模型缩减和并行推理效果。通过模型尺寸缩减和推理策略优化，将GPU小时数从9000降至92，分类准确率保持在85%以上。多轮专家评审确保分类质量。流程的自动化和模块化设计，显著提升了跨学科合作的沟通效率，推动了SSH研究的标准化。验证结果显示，模型缩减策略在保证准确率的同时，大幅提升了计算效率，为未来大规模分析提供了技术基础。
Results
实现了大规模文献分类的高效自动化，GPU小时数从9000降至92，模型推断总数达9亿次，分类准确率超过85%。流程模块化增强了跨学科合作的沟通，推动了研究标准化。验证流程的高效性和可复现性，为SSH研究提供了可持续的技术支撑。
Applications
该工作流适用于大规模文献分析、文化演变、社会行为监测等场景。研究人员可在HPC环境中快速部署模型，进行跨学科合作。产业界可借助此框架实现自动文本分析、知识图谱构建等，提升决策效率。未来还可扩展到多模态数据分析和多机构协作平台。
Limitations & Outlook
验证过程高度依赖人工，自动验证机制尚不完善，难以实现全流程自动化。元数据与FAIR标准尚在发展中，跨机构数据共享仍面临技术与政策障碍。硬件成本高，模型推断需优化算法和硬件利用率，未来需加强自动验证和标准化建设。
Plain Language Accessible to non-experts
想象你在厨房里做饭，要准备各种食材和调料，然后按照食谱一步步操作。以前每次都得自己手动调配、操作，既费时间又容易出错。现在，有一种新方法，就像设计好一份详细的食谱，把所有步骤都写成可以自动执行的程序。你只需要输入一些参数，机器就会帮你完成大部分工作，而且每次都能得到一样的味道。这就像用一台超级智能的机器人帮你做饭，不仅快，还很准。这样，厨师们可以用这个“自动化厨房”做更多菜，不用担心繁琐的细节，把时间用在创新和尝试新菜式上。这个方法让厨房变得像工厂一样高效，人人都能做出美味佳肴。
ELI14 Explained like you're 14
想象你在学校的科学实验室里做实验，要准备很多材料、仪器，然后按照步骤操作。以前每次都得自己调配、操作，既费时间又容易出错。现在，有一种新方法，就像设计好一份详细的实验流程，把所有步骤都变成可以自动执行的程序。你只需要输入一些参数，机器就会帮你完成大部分工作，而且每次都能得到一样的结果。这就像用一台超级智能的机器人帮你做实验，不仅快，还很准。这样，科学家们可以用这个“自动化实验室”做更多更复杂的研究，不用担心繁琐的细节，把精力集中在发现新知识上。这个新方法让科学变得像玩游戏一样简单又有趣，还能做出更可靠的结果。
Abstract
Artificial intelligence (AI) is gaining traction in the social sciences and humanities (SSH). However, adoption remains limited by technical barriers to high-performance computing (HPC), validation processes that lag behind AI's rapid progress, and reproducibility standards that most SSH teams cannot meet. Research workflows--common in the life sciences--address these problems via encoding and abstracting technical complexity into repeatable routines; yet, accounts of how to build them in SSH remain scarce. We report on a two-year effort to build a workflow that enables a Science and Technology Studies unit to query, analyze, and enrich OpenAlex--a database of some 460 million scholarly records--on the MareNostrum supercomputer, using methods ranging from large-scale bibliometrics to LLM-based classification. We found the main challenge was translating domain-specific research questions into engineering requirements -- bridging two distinct methodological languages, with implications that were both organizational and technical. Organizationally, it meant adopting and adapting Agile to the research rhythm and pace, and reframing collaboration from a service arrangement to a co-design process. Technically, model-driven engineering was as valuable for collaboration as it was for automation; co-building the model facilitated both the creation of a shared vocabulary and the abstraction of HPC complexity. Finally, we highlight limitations we found in validation, reproducibility, and FAIR metadata -- beyond what any single project can sustain -- calling for coordinated, cross-institutional investment in the tooling and standards needed for AI-ready SSH workflows sustainable at scale.