Rekomendasi Peneliti terhadap Topik Riset Berdasarkan Rekam Jejak Publikasi Menggunakan Transformer-based Models dan Similarity-Distance Metrics

Wati, Pelangi Masita (2025) Rekomendasi Peneliti terhadap Topik Riset Berdasarkan Rekam Jejak Publikasi Menggunakan Transformer-based Models dan Similarity-Distance Metrics. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221051-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (42MB) | Request a copy

Abstract

Sistem rekomendasi merupakan salah satu penerapan machine learning yang memanfaatkan data untuk memprediksi atau menyarankan objek, konten, atau layanan yang relevan bagi pengguna. Dalam konteks akademik, sistem serupa telah digunakan untuk mencari peneliti yang relevan. Namun, banyak solusi yang ada masih terbatas pada pendekatan berbasis pencocokan kata kunci atau statistik sederhana, yang kurang mampu menangkap makna semantik serta konteks dari suatu topik penelitian. Oleh karena itu, penelitian ini bertujuan untuk mengembangkan sistem rekomendasi peneliti berbasis rekam jejak publikasi menggunakan model bahasa berbasis transformer dan similarity-distance metrics. Sistem ini dirancang untuk memberikan rekomendasi nama-nama peneliti yang relevan terhadap topik input pengguna. Eksperimen dilakukan dengan menguji kombinasi beberapa parameter seperti variasi penggunaan data, model bahasa pre-trained yang telah di-finetuning, serta metrik untuk menemukan konfigurasi yang paling optimal terhadap ground truth cases. Dataset yang digunakan merupakan data tekstual yang diperoleh dari platform manajemen informasi riset institusi. Untuk mengevaluasi performa sistem rekomendasi yang dikembangkan, penelitian ini menggunakan evaluasi Top-K metric dengan Normalized Discounted Cumulative Gain (NDCG) dan Mean Average Precision (MAP). Hasil eksperimen menunjukkan bahwa kombinasi optimal diraih oleh MPNet dan DistilBERT serta Minkowski dan Kullback-Leibler (KL). Dengan kombinasi terbaik, MPNet dan Minkowski meraih skor 0,2827 untuk NDCG@5 dan 0,2253 untuk MAP@5.
=====================================================================================================================================
Recommendation systems are one application of machine learning that utilizes data to predict or suggest objects, content, or services that are relevant to users. In an academic context, similar systems have been used to find relevant researchers. However, many existing solutions are still limited to keyword-based or simple statistical approaches, which are less capable of capturing the semantic meaning and context of a research topic. Therefore, this study aims to develop a researcher recommendation system based on publication records using a transformer-based language model and similarity-distance metrics. This system is designed to provide recommendations for researcher names that are relevant to the user's input topic. Experiments were conducted by testing combinations of several parameters, such as variations in data usage, pre-trained language models that had been fine-tuned, and metrics to find the most optimal configuration for ground truth cases. The dataset used was textual data obtained from an institutional research information management platform. To evaluate the performance of the developed recommendation system, this study used Top-K metric evaluation with Normalized Discounted Cumulative Gain (NDCG) and Mean Average Precision (MAP). The experimental results show that the optimal combination is achieved by MPNet and DistilBERT as well as Minkowski and Kullback-Leibler (KL). With the best combination, MPNet and Minkowski achieved a score of 0,2827 for NDCG@5 and 0,2253 for MAP@5.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Rekam Jejak Publikasi, Similarity Distance Metrics, Sistem Rekomendasi, Transformer Model
Subjects:	T Technology > T Technology (General) > T58.6 Management information systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Pelangi Masita Wati
Date Deposited:	08 Jan 2026 00:46
Last Modified:	09 Jan 2026 03:11
URI:	http://repository.its.ac.id/id/eprint/129352

Actions (login required)

View Item