PENGEMBANGAN MODEL REKOMENDASI DESTINASI WISATA DENGAN LIGHTGCN DAN CONTENT-BASED FILTERING

Risyad Rafi, - (2025) PENGEMBANGAN MODEL REKOMENDASI DESTINASI WISATA DENGAN LIGHTGCN DAN CONTENT-BASED FILTERING. S1 thesis, Universitas Pendidikan Indonesia.

Abstract

Sistem rekomendasi destinasi wisata tradisional masih menghadapi tantangan serius berupa cold-start pada pengguna baru yang minim histori interaksi dan rendahnya relevansi hasil rekomendasi, sehingga menurunkan akurasi prediksi. Untuk menjawab permasalahan tersebut, penelitian ini mengembangkan pendekatan hybrid yang mengintegrasikan Light Graph Convolutional Network (LightGCN) dan Content-Based Filtering (CBF) dengan optimasi dynamic learning rate berbasis inspirasi algoritma hybrid FOX-TSA. LightGCN dimanfaatkan untuk mempelajari representasi graf interaksi pengguna–destinasi, sedangkan CBF digunakan untuk membangun rekomendasi berbasis atribut eksplisit item pada kasus cold-start. Penelitian ini menggunakan metode Design Science Research Methodology (DSRM) melalui tahapan identifikasi masalah, perancangan, pengembangan, demonstrasi, dan evaluasi. Hasil eksperimen menunjukkan konfigurasi terbaik pada bobot fusi α = 0,7 dan Top-K = 25, dengan performa Recall@25 sebesar 0,3292, Accuracy@25 sebesar 0,9300, dan Coverage@25 sebesar 0,9542. Penerapan strategi FOX-TSA terbukti meningkatkan Recall sekitar 24,8% dibanding fixed learning rate, sekaligus memperluas cakupan rekomendasi hingga mendekati sempurna. Kesimpulannya, integrasi LightGCN dan CBF dengan optimasi FOX-TSA mampu menghasilkan sistem rekomendasi yang akurat, relevan, dan adaptif, serta efektif mengatasi problem cold-start dalam konteks pariwisata digital. ---------- Traditional tourism destination recommender systems still face critical challenges such as the cold-start problem for new users with limited interaction history and the low relevance of recommendations, which reduce overall prediction accuracy. To address these issues, this study develops a hybrid approach that integrates Light Graph Convolutional Network (LightGCN) and Content-Based Filtering (CBF) with a dynamic learning rate optimization inspired by the hybrid FOX-TSA algorithm. LightGCN is employed to learn graph representations of user–destination interactions, while CBF leverages explicit item attributes to provide recommendations in cold-start cases. The research adopts the Design Science Research Methodology (DSRM) through the stages of problem identification, design, development, demonstration, and evaluation. Experimental results indicate the optimal configuration at fusion weight α = 0.7 and Top-K = 25, with system performance achieving Recall@25 of 0.3292, Accuracy@25 of 0.9300, and Coverage@25 of 0.9542. The FOX-TSA strategy improved Recall by approximately 24.8% compared to a fixed learning rate and expanded recommendation coverage to near-complete levels. In conclusion, integrating LightGCN and CBF with FOX-TSA optimization produces accurate, relevant, and adaptive recommendations, effectively overcoming the cold-start problem in digital tourism contexts.

	Text S_RPL_2101739_Title.pdf Download (650kB)
	Text S_RPL_2101739_Chapter1.pdf Download (252kB)
	Text S_RPL_2101739_Chapter2.pdf Restricted to Staf Perpustakaan Download (507kB)
	Text S_RPL_2101739_Chapter3.pdf Download (749kB)
	Text S_RPL_2101739_Chapter4.pdf Restricted to Staf Perpustakaan Download (493kB)
	Text S_RPL_2101739_Chapter5.pdf Download (210kB)
	Text S_RPL_2101739_Appendix.pdf Restricted to Staf Perpustakaan Download (359kB)

Item Type:	Thesis (S1)
Uncontrolled Keywords:	Sistem Rekomendasi, LightGCN, Content-Based Filtering, FOX-TSA, Hybrid Filtering, Pariwisata Digital, ecommender System, LightGCN, Content-Based Filtering, FOX-TSA, Hybrid Filtering, Digital Tourism.
Subjects:	L Education > L Education (General) Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software T Technology > T Technology (General)
Divisions:	UPI Kampus cibiru > S1 Rekayasa Perangkaat Lunak
Depositing User:	Risyad Rafi
Date Deposited:	19 Sep 2025 08:19
Last Modified:	19 Sep 2025 08:19
URI:	http://repository.upi.edu/id/eprint/137692

Actions (login required)

View Item