Of a Recommender Subsystem for a Web Application for Tv Series Viewing Based on Machine Learning Methods

The rapid growth of digital media platforms has intensified the need for efficient recommendation technologies capable of assisting users in navigating large volumes of audiovisual content. In environments where users are exposed to extensive catalogs of TV series, the role of recommendation systems becomes crucial for enhancing user engagement and improving navigation efficiency. This paper presents the development of a recommendation subsystem for a web application designed for watching TV series, with the Apriori algorithm selected as the central method due to its interpretability, simplicity, and strong performance on medium-scale transactional datasets. The study addresses the challenge of constructing a transparent recommendation mechanism that identifies stable patterns in user behaviour without relying on resource-intensive neural architectures. The proposed methodology utilizes a transactional representation of the users’ viewing history, where each transaction corresponds to a set of series watched by a specific user. The Apriori algorithm is applied to extract frequent itemsets and generate associative rules based on support, confidence, and lift metrics. Mathematical formalization ensures a robust and transparent mechanism for detecting consistent associations, forming the basis for personalized recommendations. To support the algorithm, the system incorporates a comprehensive data preprocessing pipeline involving data cleaning, normalization, and transformation of viewing logs into structured transactions. The analysis also considers the impact of minimum support and confidence thresholds on the quality and interpretability of the resulting rules. The architecture of the developed recommendation subsystem includes modules for data collection, preprocessing, analytical processing, recommendation generation, and monitoring. This modular design ensures flexibility, enabling independent updates of individual components and adaptation to evolving user behaviours. The implementation demonstrates that Apriori-based rules can be efficiently integrated into the web application’s workflow, forming recommendation lists grounded in observable user tendencies rather than statistical artifacts. Evaluation metrics—including precision, NDCG, diversity, novelty, and serendipity—indicate that the system provides relevant and varied recommendations while maintaining interpretability. The results confirm that classical associative methods remain highly valuable for recommendation tasks in small- and medium-scale platforms, offering transparency and low computational overhead compared to deep learning approaches. The developed subsystem provides a practical foundation for further improvement, including hybridization with content-based features, incorporation of contextual signals, and integration of online learning techniques for adaptive rule updating. This research demonstrates that Apriori-driven recommendation models can serve as a reliable, interpretable, and efficient solution for modern media applications requiring structured knowledge extraction from transactional data.