Ensuring Reproducibility in OntoClue – Vector-Based Document Similarity for Biomedical Literature – Using Docker

Accurately assessing document-to-document similarity in biomedical research is a challenging task due to the vast amount of literature and the complexity of semantic relationships between documents. OntoClue addresses this problem by providing a comprehensive framework that enhances document recommendations using various vector-based approaches. Built around the RELISH corpus—an expert-curated biomedical database—OntoClue integrates powerful NLP models such as Word2Vec, Doc2Vec, fastText, SciBERT, BioBERT, and SPECTER, along with hybrid methods incorporating Named Entity Recognition (NER) for better semantic understanding. The framework evaluates similarity through precision and cumulative gain metrics and streamlines processes like dataset downloading and model training. What sets OntoClue apart is its Docker-based containerization, which ensures reproducibility, version control, and smooth dependency management across platforms. By standardizing the environment, Docker eliminates inconsistencies often encountered in research, ensuring that results are consistent and reliable, even down to decimal precision.