Rekomendasi Peneliti Berdasarkan Rekam Jejak Topik Publikasi Menggunakan Representasi Kontekstual Bert Dan Similaritas

Nirwasita, Gavrila (2026) Rekomendasi Peneliti Berdasarkan Rekam Jejak Topik Publikasi Menggunakan Representasi Kontekstual Bert Dan Similaritas. Other thesis, InstitutTeknologi Sepuluh Nopember.

Text 5025221227-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB)

Abstract

Fenomena ledakan informasi di era digital menimbulkan tantangan yang substansial dalam pencarian pakar yang tepat, di mana metode pencarian yang konvensional yang mengandalkan pencocokan kata kunci (keyword matching) memiliki kelemahan yang fundamental. Pendekatan ini gagal dalam mengatasi masalah vocabulary mismatch dan tidak mampu menangkap makna tekstual, sehingga sistem mengekstrak kata kunci dalam abstrak dengan keahlian inti yang sebenarnya. Untuk mengatasi permasalahan tersebut, penelitian ini mengusulkan pengembangan sistem rekomendasi peneliti berdasarkan rekam jejak publikasi yang memanfaatkan teknologi Natural Language Processing (NLP). Tiga pre-trained model berbasis BERT dievaluasi untuk mengubah data 850 profil peneliti dari ITS Scholar menjadi representasi vektor kaya konteks, yang kemudian dicocokkan dengan query pengguna menggunakan algoritma Cosine Similarity. Berdasarkan pengujian menggunakan metrik precision, recall, F1-Score, dan NDCG, model IndoSentence-BERT menghasilkan kinerja pemeringkatan dengan skor NDCG@10 mencapai 0.8703 pada bidang Kecerdasan Artifisial. Sistem ini memberikan solusi objektif untuk mempermudah mahasiswa menemukan dosen pembimbing yang relevan serta mendorong kolaborasi riset yang lebih efektif.
=======================================================================================================================================
The phenomenon of information overload in the digital era creates significant challenges in identifying appropriate academic experts. Conventional search methods relying on keyword matching possess fundamental weaknesses, primarily failing to address vocabulary mismatch and unable to capture textual meaning. Consequently, systems struggle to distinguish between mere keyword mentions in abstracts and actual core expertise. To address these issues, this research proposes a researcher recommendation system based on publication track records utilizing Natural Language Processing (NLP) technology Three BERT-based pre-trained models were evaluated to transform 850 researcher profiles sourced from ITS Scholar into context-rich vector representations, which were then matched with user queries using the Cosine Similarity algorithm. Based on evaluations using Precision, Recall, F1-Score, and NDCG metrics, the IndoSentence-BERT model proved to deliver the supurior performance, achieving a NDCG@10 score of 0.8703 in the Artificial Intelligence domain. This system provides an objective solution to facilitate students in finding relevant thesis advisors and foster more effective research collaboration.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Sistem Rekomendasi, Natural Language Processing (NLP), BERT, Cosine Similarity, Rekam Jejak Publikasi
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76 Computer software Q Science > QA Mathematics > QA76.6 Computer programming. Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science) Q Science > QA Mathematics > QA76.9.I58 Recommender systems (Information filtering)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Gavrila Nirwasita
Date Deposited:	24 Jul 2026 01:39
Last Modified:	24 Jul 2026 01:39
URI:	http://repository.its.ac.id/id/eprint/136854

Actions (login required)

View Item