This is a project which is currently making use of HPC facilities at Newcastle University. It is active.
For further information about this project, please contact:
The Trust Research Methodology Database (TReMeDa) project aims to create a comprehensive, curated database of secondary quantitative replication data related to the study of social trust. The project contributes to improving quantitative research pedagogy in sociology and related social science disciplines while supporting the broader research reproducibility agenda by providing structured metadata and replication resources for empirical studies.
The project involves constructing a large-scale corpus of approximately 75,000+ academic publications (2000–2025) collected from sources including Web of Science, Scopus, and OpenAlex. Advanced natural language processing (NLP) and large language model (LLM)-based methods are applied to identify research relevance, classify publications according to thematic areas and methodological approaches, and extract detailed information about datasets, analytical methods, statistical techniques, and replication potential from full-text academic articles.
HPC resources are required to support the computationally intensive large-scale document processing and machine learning pipeline. The workflow involves preprocessing and analysing a large academic corpus, running transformer-based language models for classification and information extraction, and performing model evaluation and light fine-tuning of LLMs for domain-specific methodological understanding. Access to HPC infrastructure will enable efficient parallel processing, accelerated model inference, and scalable experimentation that would be infeasible on standard computing environments.
The expected outcome is a structured and searchable research methodology database that enhances transparency, facilitates replication studies, and provides researchers and educators with accessible resources for understanding quantitative research practices in social trust research.
The project will use a Python-based natural language processing (NLP) and machine learning workflow for large-scale analysis of academic literature. The software stack will include tools for scholarly document processing, text extraction, data preprocessing, machine learning, and large language model (LLM)-based analysis. Relevant frameworks include PyTorch, Hugging Face Transformers, and other modern NLP libraries for model inference, evaluation, experimentation, and potential domain adaptation.
The workflow will involve CPU-based document processing and information extraction from large academic corpora, alongside GPU-accelerated workloads for deep learning model inference, fine-tuning, and experimentation with different transformer-based language models. Techniques such as parameter-efficient fine-tuning, retrieval-augmented generation (RAG), and embedding-based semantic search may be explored to improve domain-specific information extraction and classification performance.
HPC resources will primarily support scalable batch processing of a large corpus of research publications (~75,000+ documents), including parallel text processing, model inference, and iterative machine learning experiments. Computational environments will be managed through reproducible container-based workflows using technologies such as Apptainer/Singularity or Docker-compatible containers, ensuring consistency across development and HPC execution environments.
Typical workloads will include CPU-intensive preprocessing pipelines, single- or multi-GPU accelerated model inference, training/fine-tuning experiments, and evaluation workflows requiring scalable compute resources.