In this article we present a hybrid approach to recommending the Universal Decimal Classification (UDC) codes for unclassified documents. By recommending UDC codes to librarians, we can provide the decision support as part of a semi-automatic cataloguing process. As current approaches to recommending UDC codes are scarce and limited to certain fields of expertise, our motivation was to create a hybrid approach which covers all fields of expertise within the UDC hierarchy. The cascade hybrid approach combines the BM25 ranking function with a multi-label classifier based on a BERT (Bidirectional Encoder Representations from Transformers) deep neural network architecture. Additionally, lists of the most commonly used UDC codes within a document’s origin organization are used as a final content-based filtering method. The BM25 ranking function is used to create an initial list of recommendations. The first cascade step re-ranks the initial list of recommendations using the list of recommendations produced by the multi-label BERT-based classifier. The second cascade step re-ranks the resulting recommendation list from the first cascade step using a list of most commonly used UDC codes within the document’s organization. Finally, post-processing steps are applied to obtain the final list of recommended UDC codes. We present in detail the UDC structure, the used text corpus of documents and the functioning of our hybrid recommendation approach. We perform the analysis of the generated recommendation lists for the corpus from the Slovenian Open-Access Infrastructure using the metrics hit-ratio, normalized discounted cumulative gain, mean reciprocal rank and mean average precision. Our hybrid recommendation approach improves reference scores of individual methods for these metrics.

This paper presents a hybrid document recommender system intended for use in digital libraries and institutional repositories that are part of the Slovenian Open Access Infrastructure. The recommender system provides recommendations of similar documents across different digital libraries and institutional repositories with the aim to connect researchers and improve collaboration efforts. The hybrid recommender system makes use of document processing techniques, document metadata, and the similarity ranking function BM25 to provide content-based recommendations as a primary method. It also uses collaborative-filtering methods as a secondary method in a cascade hybrid recommendation technique. We also provide a real-world data feedback collection analysis for our hybrid recommender system on an academic digital repository in order to be able to identify suitable time-frames for direct feedback collection during the year.

Manual assignment of Universal Decimal Classification (UDC) codes is time-consuming and inconsistent as digital library collections expand. This study evaluates 17 large language models (LLMs) as UDC classification recommender systems, including ChatGPT variants (GPT-3.5, GPT-4o, and o1-mini), Claude models (3-Haiku and 3.5-Haiku), Gemini series (1.0-Pro, 1.5-Flash, and 2.0-Flash), and Llama, Gemma, Mixtral, and DeepSeek architectures. Models were evaluated zero-shot on 900 English and Slovenian academic theses manually classified by professional librarians. Classification prompts utilized the RISEN framework, with evaluation using Levenshtein and Jaro–Winkler similarity, and a novel adjusted hierarchical similarity metric capturing UDC’s faceted structure. Proprietary systems consistently outperformed open-weight alternatives by 5–10% across metrics. GPT-4o achieved the highest hierarchical alignment, while open-weight models showed progressive improvements but remained behind commercial systems. Performance was comparable between languages, demonstrating robust multilingual capabilities. The results indicate that LLM-powered recommender systems can enhance library classification workflows. Future research incorporating fine-tuning and retrieval-augmented approaches may enable fully automated, high-precision UDC assignment systems.

The OpenScience Slovenia metadata dataset contains metadata entries for Slovenian public domain academic documents which include undergraduate and postgraduate theses, research and professional articles, along with other academic document types. The data within the dataset was collected as a part of the establishment of the Slovenian Open-Access Infrastructure which defined a unified document collection process and cataloguing for universities in Slovenia within the infrastructure repositories. The data was collected from several already established but separate library systems in Slovenia and merged into a single metadata scheme using metadata deduplication and merging techniques. It consists of text and numerical fields, representing attributes that describe documents. These attributes include document titles, keywords, abstracts, typologies, authors, issue years and other identifiers such as URL and UDC. The potential of this dataset lies especially in text mining and text classification tasks and can also be used in development or benchmarking of content-based recommender systems on real-world data.

The paper presents the legal, organisational and technical perspectives regarding the implementation of the Slovenian national open access infrastructure for electronic theses and dissertations as well as for research publications. The infrastructure consists of four institutional repositories and a national portal that aggregates content from the university repositories and other Slovenian archives in order to provide a common search engine, recommendation of similar publications, and similar text detection. We have developed the software which is integrated with the universities' information and authentication systems and with the COBISS.SI. During the project the necessary legal background was defined and processes for mandatory submission of electronic theses and dissertations as well as of research publications were designed. The processes for data exchange between the institutional repositories and the national portal, and the processes for similar text detection and recommendation system were established. Bilingual web and mobile applications, a recommendation system and the interface suitable for persons with disabilities are provided to the users from around the world. The repositories are an effective promotion tool for universities and their researchers. It is expected that they will improve the recognition of Slovenian universities in the world. The complex national open access infrastructure with similar text detection support and integration with other systems will enable the storage of almost eighty percent of peer-reviewed scientific papers, annually published by Slovenian researchers. The majority of electronic theses and dissertations yearly produced at the Slovenian higher education institutions will also be accessible.

The article explores the potential of PDFGear Copilot, a chatbot-based PDF editing tool, in assisting with the analysis of historical documents. We evaluated the chatbot's performance on a document relating to the Slovenian War of Independence. We included 25 factual and 5 interpretative questions to address its formal characteristics and content details, assess its capacity for in-depth interpretation and contextualized critical analysis, and evaluate the chatbot’s language use and robustness. The chatbot exhibited some ability to answer factual questions, even though its performance varied. It demonstrated proficiency in navigating document structure, named entity recognition, and extracting basic document information. However, performance declined significantly in tasks such as document type identification, content details, and tasks requiring deeper text analysis. For interpretative questions, the chatbot's performance was notably inadequate, failing to link cause-and-effect relationships and provide the depth and nuance required for historical inquiries.