Design and Development of an Intelligent Web-Based Recommendation System Using Collaborative Filtering

Abstract

Recommender systems reduce information overload by predicting items that are likely to match an individual user’s preferences. This methodology paper presents the design and development of an intelligent web-based recommendation system centered on collaborative filtering (CF). The proposed system captures explicit ratings and implicit interaction signals through a browser interface, converts them into a sparse user-item matrix, computes similarity among users or items, selects nearest neighbors, estimates unknown ratings, and returns a ranked Top-N list through a recommendation service. The design emphasizes modular web integration, explainable neighborhood-based prediction, continuous feedback, and evaluation with mean absolute error (MAE) and root mean square error (RMSE). A controlled synthetic prototype experiment using 300 users, 250 items, 10,520 observed 1–5 ratings, an 80:20 train-test split, and k=20 neighbors was used to demonstrate the evaluation procedure. User-based CF achieved an MAE of 0.638 and RMSE of 0.797, while item-based CF achieved 0.643 and 0.801 respectively, both improving on a global-mean baseline. The methodology provides a practical foundation for web commerce, digital libraries, learning portals, and media platforms while retaining compatibility with later extensions such as matrix factorization, context-aware filtering, and hybrid recommendation.