Performance of the Annif Model in Enhancing Metadata Quality: A Bibliometric Perspective Across Multidisciplinary Research Outputs
The rapid growth of scholarly publications across multiple disciplines has made high-quality metadata essential for effective information retrieval and resource discovery. Traditional human indexing, while accurate, is time-consuming, labour-intensive, and often inconsistent, highlighting the need for scalable automated solutions. This study evaluates the performance of the Annif model in enhancing metadata quality across four major academic domains: Library and Information Science (LIS), Computer Science, Health Sciences, and Social Sciences. Using a sample of 5,000 research outputs, Annif-generated keywords were compared with human-assigned metadata to assess improvements in keyword completeness, thematic specificity, vocabulary consistency, and retrieval performance. Evaluation metrics included precision, recall, F1-score, and user-oriented measures such as search satisfaction and ease of discovery. The findings indicate that Annif significantly enhances metadata quality, increasing keyword coverage by 38–57%, improving retrieval precision by 12–20%, and reducing indexing time from 18 hours per 1,000 records to 32 minutes. Error analysis revealed common challenges, including concept drift and over-generalisation, emphasising the continued need for domain oversight. Overall, the study demonstrates that Annif offers a scalable, efficient, and effective approach to metadata enhancement, supporting improved discoverability and knowledge organisation in multidisciplinary research repositories.