A Knowledge-Driven Dual-Path Framework for Sparse and Cold-Start Recommendation via Progressive Interaction Densification and Optimization

This dataset contains anonymized digital library borrowing records and metadata collected to support research in recommender systems, collaborative filtering, sparsity reduction, and cold-start recommendation. The dataset was used in the experimental evaluation presented in the paper: A Knowledge-Driven Dual-Path Framework for Sparse and Cold-Start Recommendation via Progressive Interaction Densification and Optimization The data represent implicit user–item interactions derived from borrowing activities in a digital library environment. In addition to interaction records, the dataset includes metadata attributes that enable the development and evaluation of hybrid and metadata-aware recommendation approaches. The dataset exhibits extreme sparsity characteristics and contains both warm-start users and real cold-start users with no interaction history, making it suitable for benchmarking recommendation models under sparse and cold-start conditions. Contents The dataset includes: Anonymized user identifiers Book/item identifiers (BIBID) Borrowing indicators User metadata attributes Item metadata attributes Interaction records for recommendation experiments Research Applications This dataset may be used for research in: Recommender systems Collaborative filtering Cold-start recommendation Sparse matrix learning Implicit feedback modeling Metadata-aware recommendation Digital library analytics Machine learning optimization Data Privacy and Ethics All personally identifiable information has been removed or anonymized prior to publication. Membership identifiers were masked to preserve user confidentiality and comply with research ethics requirements.