Abstract
Abstract: Recommender systems are widely deployed in e-commerce, yet their evaluation is still dominated by prediction-accuracy metrics that say little about the value customers actually perceive. This study examines Collaborative Filtering (CF) as a driver of customer value and loyalty in e-commerce management. Four CF variants (user-based and item-based, with cosine similarity and Pearson correlation) were compared with a popularity baseline using 10-fold cross validation, followed by a six-month A/B experiment and a survey of 400 customers analysed with validity, reliability, classical assumption, difference, multiple regression, and bootstrap mediation tests in Python. Item-based CF with Pearson correlation performed best (MAE 0.687; RMSE 0.856), reducing RMSE by 7.0% relative to the baseline and quadrupling catalogue coverage. CF recommendations increased click-through rate by 29.9% and conversion rate by 41.7%, and shortened product search time by 13.6%. CF significantly raised functional and emotional value but not economic value; the three value dimensions jointly explained loyalty (adjusted R² = 0.224), and customer value fully mediated the effect of CF on loyalty. Performance deteriorated for cold-start users and at very high sparsity. CF therefore acts as a strategic management instrument whose business value flows through search efficiency and emotional experience. Keyword: collaborative filtering; Teknologi informasi, customer value; e-commerce; customer loyalty; recommender system. Abstrak: Sistem rekomendasi telah banyak diterapkan dalam e-commerce, tetapi evaluasinya masih didominasi metrik akurasi prediksi yang belum menggambarkan nilai yang dirasakan pelanggan. Penelitian ini mengkaji Collaborative Filtering (CF) sebagai penggerak customer value dan loyalitas pelanggan dalam manajemen e-commerce. Empat varian CF (user-based dan item-based dengan cosine similarity dan Pearson correlation) dibandingkan dengan baseline produk populer menggunakan 10-fold cross validation, dilanjutkan eksperimen A/B selama enam bulan dan survei terhadap 400 pelanggan yang dianalisis dengan uji validitas, reliabilitas, asumsi klasik, uji beda, regresi linier berganda, dan mediasi bootstrap menggunakan Python. Item-based CF dengan Pearson correlation memberikan kinerja terbaik (MAE 0,687; RMSE 0,856), menurunkan RMSE 7,0% dibanding baseline dan memperluas cakupan katalog empat kali lipat. Rekomendasi CF meningkatkan CTR sebesar 29,9% dan conversion rate sebesar 41,7%, serta mempersingkat waktu pencarian produk 13,6%. CF meningkatkan nilai fungsional dan emosional secara signifikan, tetapi tidak nilai ekonomi; ketiga dimensi nilai secara simultan menjelaskan loyalitas (adjusted R2 = 0,224), dan customer value memediasi penuh pengaruh CF terhadap loyalitas. Kinerja CF menurun pada pengguna cold start dan pada tingkat sparsity sangat tinggi. CF merupakan instrumen manajemen strategis yang nilai bisnisnya bekerja melalui efisiensi pencarian dan pengalaman emosional pelanggan. Kata kunci: collaborative filtering; customer value; e-commerce; loyalitas pelanggan; sistem rekomendasi.
