Pembangunan Sistem Rekomendasi Produk Berbasis Collaborative Filtering Pada Lingkungan Terdistribusi

Athallah, Muhammad Arsy (2026) Pembangunan Sistem Rekomendasi Produk Berbasis Collaborative Filtering Pada Lingkungan Terdistribusi. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5027221048-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Pertumbuhan pesat platform e-commerce menyebabkan peningkatan signifikan dalam jumlah produk dan interaksi pengguna dalam bentuk rating dan ulasan, sehingga menimbulkan permasalahan information overload di mana pengguna mengalami kesulitan dalam menemukan produk yang benar-benar sesuai dengan preferensi dan kebutuhan mereka. Untuk mengatasi permasalahan tersebut, sistem rekomendasi hadir sebagai solusi yang memanfaatkan data historis interaksi pengguna guna menghasilkan saran produk yang dipersonalisasi secara otomatis. Penelitian ini mengembangkan dan membandingkan sistem rekomendasi produk menggunakan tiga pendekatan, yaitu Collaborative Filtering berbasis Alternating Least Squares (ALS-CF), Content-Based Filtering berbasis Pre-trained Language Model (PLM-CBF) menggunakan Sentence Transformer all-MiniLM-L6-v2, serta GRAM-CF (Graph-based Recommendation with Attentive Memory) yang mengintegrasikan PLM embedding dengan Bayesian Personalized Ranking untuk optimasi, dengan seluruh sistem dibangun pada lingkungan terdistribusi menggunakan MongoDB Sharded Cluster untuk menjamin ketersediaan dan skalabilitas data. Dataset yang digunakan adalah Amazon Product Reviews 2023 kategori Video Games dari HuggingFace yang mencakup ~1.243.553 interaksi, 312.572 pengguna unik, dan 51.756 produk unik dengan tingkat sparsitas 99,97%. Evaluasi dilakukan menggunakan metrik Root Mean Square Error (RMSE) untuk prediksi rating serta Precision@K, Recall@K, NDCG@K, dan Hit Rate@K untuk evaluasi rekomendasi Top-N pada nilai K = {5, 10, 20}. Hasil penelitian menunjukkan bahwa model ALS-CF memberikan performa terbaik dengan RMSE 2,3457 dan NDCG@10 sebesar 0,0629, diikuti PLM-CBF (NDCG@10 = 0,0078) dan GRAM-CF (NDCG@10 = 0,0053). Sistem juga dievaluasi secara subjektif melalui User Acceptance Testing (UAT) menggunakan System Usability Scale (SUS) terhadap antarmuka pengguna Suggestify, dengan sparsitas data yang sangat tinggi (99,97%) menjadi tantangan utama yang mempengaruhi performa seluruh metode yang diuji.
=================================================================================================================================
The rapid growth of e-commerce platforms has caused a significant increase in the number of products and user interactions in the form of ratings and reviews, giving rise to the problem of information overload where users struggle to find products that truly match their preferences and needs. To address this, recommendation systems have emerged as a solution that leverages historical user interaction data to automatically generate personalized product suggestions. This study develops and compares a product recommendation system using three approaches, namely Collaborative Filtering based on Alternating Least Squares (ALS-CF), Content-Based Filtering based on a Pre-trained Language Model (PLM-CBF) using Sentence Transformer all-MiniLM-L6-v2, and GRAM-CF (Graph-based Recommendation with Attentive Memory) integrating PLM embeddings with Bayesian Personalized Ranking for optimization, with the entire system built in a distributed environment using MongoDB Sharded Cluster to ensure data availability and scalability. The dataset used is the Amazon Product Reviews 2023 Video Games category from HuggingFace, comprising ~1,243,553 interactions, 312,572 unique users, and 51,756 unique products with a sparsity of 99.97%. Evaluation was conducted using Root Mean Square Error (RMSE) for rating prediction and Precision@K, Recall@K, NDCG@K, and Hit Rate@K for Top-N recommendation evaluation at K = {5, 10, 20}. Results show that ALS-CF achieved the best performance with RMSE of 2.3649 and NDCG@10 of 0,0629, followed by PLM-CBF (NDCG@10 = 0.0078) and GRAM-CF (NDCG@10 = 0.0053). The system was also evaluated subjectively through User Acceptance Testing (UAT) using the System Usability Scale (SUS) on the Suggestify user interface, with very high data sparsity (99.97%) identified as the primary challenge affecting the performance of all methods tested.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Bayesian Personalized Ranking, content-based filtering, collaborative filtering, GRAM-CF, MongoDB Sharded Cluster, pre-trained language model, sistem rekomendasi, Bayesian Personalized Ranking, collaborative filtering, content-based filtering, GRAM-CF, MongoDB Sharded Cluster, pre-trained language model, recommendation system
Subjects:	T Technology > T Technology (General)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information Technology > 59201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Arsy Athallah
Date Deposited:	01 Aug 2026 06:56
Last Modified:	01 Aug 2026 06:56
URI:	http://repository.its.ac.id/id/eprint/141702

Actions (login required)

View Item