Integrasi Fitur Item dan Fitur Pengguna untuk Rekomendasi E-Commerce Menggunakan TimeSVD dan Contractive Autoencoder

Tertiabudi, Vania Aileen (2026) Integrasi Fitur Item dan Fitur Pengguna untuk Rekomendasi E-Commerce Menggunakan TimeSVD dan Contractive Autoencoder. Other thesis, Institut Tekonologi Sepuluh Nopember.

Text 5026221108-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Perkembangan e-commerce mendorong perlunya sistem rekomendasi yang mampu menyesuaikan diri dengan perubahan preferensi pengguna. Sistem rekomendasi berbasis collaborative filtering seperti TimeSVD telah mengakomodasi dinamika temporal, namun masih menghadapi tantangan dalam mengolah fitur tambahan pengguna dan item secara stabil. Untuk itu, tugas akhir ini mengusulkan integrasi TimeSVD dengan Contractive Autoencoder (CAE) guna menghasilkan representasi fitur yang lebih tangguh terhadap gangguan minor dalam data. Metodologi mencakup tiga tahapan utama yaitu ekstraksi fitur menggunakan Contractive Autoencoder, Model TimeSVD dengan Contractive Autoencoder, dan Uji Model Prediksi menggunakan metrik RMSE dan MAE. Tugas akhir ini ini memanfaatkan fitur temporal dan informasi laten dari ulasan pengguna pada Amazon Beauty and Personal Care Dataset. Hasil yang diberikan dari model TimeSVD dengan Contractive Autoencoder memberikan nilai MAE yang paling kecil dibandingkan model lainnya.
================================================================================================================================
The development of e-commerce drives the need for recommendation systems that are able to adapt to changes in user preferences. Collaborative filtering-based recommendation systems such as TimeSVD have accommodated temporal dynamics, but still face challenges in processing additional user and item features stably. Therefore, this thesis proposes the integration of TimeSVD with Contractive Autoencoder (CAE) to produce feature representations that are more robust to minor disturbances in the data. The methodology includes three main stages: feature extraction using Contractive Autoencoder, TimeSVD Model with Contractive Autoencoder, and Prediction Model Testing using RMSE and MAE metrics. This thesis utilizes temporal features and latent information from user reviews on the Amazon Beauty and Personal Care Dataset. The results provided by the TimeSVD model with Contractive Autoencoder provide the smallest MAE value compared to other models.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Sistem Rekomendasi, TimeSVD, Contractive Autoencoder, Fitur Item, Fitur Pengguna Recommendation System, TimeSVD, Contractive Autoencoder, Item Features, User Features
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Information Technology > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Vania Aileen Tertiabudi
Date Deposited:	29 Jan 2026 07:26
Last Modified:	29 Jan 2026 07:26
URI:	http://repository.its.ac.id/id/eprint/131021

Actions (login required)

View Item