Perancangan Sistem Rekomendasi Author Dengan Perlindungan Privasi Query Menggunakan Homomorphic Inference

Supriyanto, Ricardo (2026) Perancangan Sistem Rekomendasi Author Dengan Perlindungan Privasi Query Menggunakan Homomorphic Inference. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221218-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Sistem rekomendasi author membantu peneliti menemukan kolaborator berdasarkan kesesuaian topik. Namun sistem konvensional memproses query plaintext di server, membuka risiko kebocoran ide riset sensitif. Penelitian ini merancang sistem yang melindungi privasi query melalui homomorphic inference, di mana embedding query dienkripsi di client dan server menghitung kesesuaian pada data terenkripsi tanpa mengetahui isi query. Representasi author menggunakan multi-centroid dari Findme-Scholar dengan embedding all-MiniLM-L6-v2. Penelitian mengevaluasi empat skenario homomorphic inference dan lima skema homomorphic encryption (TFHE, CKKS, BGV, BFV, Paillier) terhadap 2.620 author ITS dengan 21 kombinasi query, mengukur akurasi melalui Overlap@5 dan NDCG@5 terhadap baseline plaintext. Hasil menunjukkan bounded-pool mencapai Overlap@5 68 - 85% (NDCG 0,63 - 0,92) dalam 3 - 40 detik, dot product penuh mencapai 100% dengan variasi waktu 83 – 4.120 detik antar skema (tercepat BFV, terlambat Paillier), sedangkan surrogate skala penuh gagal (0% akurasi) dengan ledakan waktu kompilasi. Penelitian menyimpulkan homomorphic inference layak untuk menjaga privasi query pada rekomendasi author, dengan bounded-pool menjadi pilihan paling praktis mengingat trade-off akurasi-efisiensi.
=========================================================================================================================================
Author recommendation systems help researchers find collaborators based on research topic similarity. However, conventional systems process queries in plaintext on the server, risking exposure of sensitive unpublished research ideas. This study designs an author recommendation system protecting query privacy through homomorphic inference, where query embeddings are encrypted on the client and the server computes similarity on encrypted data without accessing query content. Author representations use the multi-centroid approach from Findme-Scholar with all-MiniLM-L6-v2 embeddings. The study evaluates four homomorphic inference scenarios and compares five homomorphic encryption schemes (TFHE, CKKS, BGV, BFV, Paillier) over 2,620 ITS authors using 21 query combinations, measuring accuracy through Overlap@5 and NDCG@5 against a plaintext baseline. Results show bounded-pool achieves Overlap@5 of 68–85% (NDCG 0.63–0.92) in 3–40 seconds, full dotproduct reaches 100% accuracy with execution times varying from 83 seconds (BFV) to 4,120 seconds (Paillier), while full-scale surrogate fails (0% accuracy) with compilation time explosion. This study concludes homomorphic inference is viable for preserving query privacy in author recommendation systems, with the bounded-pool approach as the most practical choice given the accuracy-efficiency trade-off.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	sistem rekomendasi author, privasi query, homomorphic inference, homomorphic encryption, multi-centroid, author recommendation system, query privacy, homomorphic inference, homomorphic encryption, multi-centroid
Subjects:	Q Science > QA Mathematics > QA278.55 Cluster analysis Q Science > QA Mathematics > QA76.9.A25 Computer security. Digital forensic. Data encryption (Computer science) Q Science > QA Mathematics > QA76.9.C55 Client/server computing Q Science > QA Mathematics > QA76.9.D33 Data compression (Computer science) Q Science > QA Mathematics > QA76.9.I58 Recommender systems (Information filtering) Q Science > QA Mathematics > QA9.58 Algorithms
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Ricardo Supriyanto
Date Deposited:	25 Jul 2026 06:03
Last Modified:	25 Jul 2026 06:03
URI:	http://repository.its.ac.id/id/eprint/137800

Actions (login required)

View Item