The Implementation of Supervised Learning Recommender System to Enhance Reading Interest of Visitors in Regional Libraries

Author

Irfan Dwiguna Sumitra, S.Kom., M.Kom. Ph.D

Abstrak

In the era of digitization, regional libraries face challenges in sustaining the reading interest of their visitors. One solution that could be implemented to enhance the reading interest of visitors in regional libraries is by applying the supervised learning recommender system method based on the profiling of previous book borrowers. This study aims to implement this method on the borrowing history of books by visitors to regional libraries in order to recommend books that align with their reading interests, thereby increasing the number of borrowed books, facilitating visitors in finding the books they need, and providing the best service such as user experience to the visitors. The method used in this research is item-based collaborative filtering with the K-Nearest Neighbor (K-NN) algorithm and cosine similarity distance to find similar book titles based on the titles previously borrowed by book borrowers. The visitor’s borrowing history data in this study consists of Bibliographic Data from book borrowers in 2022 at regional libraries. The data underwent preprocessing to enhance accuracy, resulting in quantitative data comprising attributes such as borrower ID, book ID, and result. The model evaluation method used a confusion matrix on the training data compared to the testing data to determine the accuracy, precision, recall, and f-measure values generated by the created K-Nearest Neighbor (K-NN) model. The result of this study is a recommender system model that shows the top 10 book titles based on the borrowing history of previous book titles entered by the visitor. The performance measurements of the recommendation model yield the following results: the model’s accuracy stands at 68.53%, signifying its ability to accurately classify all samples. Precision is recorded at 99.79%, indicating that the model tends to have minimal false positives (FP). The recall rate is 68.61%, suggesting that the model excels at detecting positive samples compared to negative ones. The F-measure registers at 81.32%, showcasing a fairly effective trade-off between precision and recall. The implementation of this model can be used as a reference for regional libraries in providing book recommendations favored by visitors and assisting visitors in selecting books aligned with their interests based on the profiling of previous borrowers, thus increasing the number of books borrowed.