Rekomendasi Kolaborasi Penelitian Antardomain Menggunakan Metode Cross-Domain Topic Learning Berbasis Frase

zuraida, vit (2018) Rekomendasi Kolaborasi Penelitian Antardomain Menggunakan Metode Cross-Domain Topic Learning Berbasis Frase. Masters thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 05111650010021-Master_Thesis.pdf]
Preview
Text
05111650010021-Master_Thesis.pdf - Accepted Version

Download (2MB) | Preview

Abstract

Rekomendasi kolaborasi antardomain memiliki permasalahan terkait kelangkaan topik kolaborasi antardomain yang relevan sehingga berpengaruh terhadap akurasi rekomendasi. Sistem rekomendasi dibangun berdasarkan dokumen penelitian yang pernah dipublikasikan, baik judul, abstrak, bibliografi, maupun isi secara keseluruhan. Oleh karena itu, proses ekstraksi topik riset seorang peneliti merupakan tahapan penting. Model topik berbasis bag-of-words belum dapat merepresentasikan topik dengan baik sebab urutan kata pada dokumen tidak diperhitungkan. Penelitian ini mengusulkan sistem rekomendasi kolaborasi penelitian antardomain menggunakan metode Cross-Domain Topic Learning (CTL) Berbasis Frase yang memperhatikan urutan kata. CTL dengan frase juga mempertimbangkan kelangkaan topik kolaborasi antardomain.
Sistem rekomendasi kolaborasi yang diusulkan terdiri dari tiga fase utama. Fase pertama adalah transformasi dokumen dari format bag-of-words menjadi bag-of-phrases. Fase kedua adalah pemodelan topik terhadap frase yang sudah dibentuk. Hasilnya adalah distribusi probabilitas keterkaitan peneliti dengan topik. Nilai probabilitas tersebut selanjutnya dijadikan input dalam fase Random Walk with Restart yang menghasilkan rekomendasi kolaborator.
Uji coba dilakukan pada domain visualization dan data mining dari dataset penelitian AMiner untuk maksimal tiga kata dalam frase. Hasil uji coba menunjukkan bahwa rekomedasi yang dihasilkan CTL Berbasis Frase lebih baik daripada CTL berbasis kata tunggal (bag-of-words). Terdapat peningkatan nilai precision sebesar ±10% pada 10 rekomendasi teratas dan ±5% pada 20 rekomendasi teratas untuk kebenaran hasil rekomendasi.
=========================================================
Cross-domain collaboration have several specific issues including the rarity of relevant cross-domain collaboration topics that could affect the accuracy of the recommendation. Recommendation systems are built based on published research documents, including titles, abstracts, bibliographies, or the entire content of the documents. Therefore, the process of extracting research topics from a researcher is an important step. Topic modeling based on bag-of-words are not able to represent the topic effectively because the order of words in the document is not considered. This research proposes cros-domain research collaboration recommendation system using Phrase-Based Cross-Domain Topic Learning (CTL) method that considers word order. Phrase-Based CTL also considers the rarity of cross-domain collaboration topics.
Phrase-Based CTL consists of three main phases. The first phase is the transformation of documents from the bag-of-words format into bag-of-phrases. The second phase is the topic modeling of the established phrases. The result is probability distribution of the researcher's relevance to each topic. The probability distribution is then used as input in the Random Walk with Restart phase resulting in collaborator ranking.
Experiments were conducted on the domain visualization and data mining of the AMiner research dataset for a maximum of three words in a phrase. Experimental result shows that the recommendations produced by Phrase-Based CTL are better than CTL based on bag-of-words. There is ± 10% improvement of precision value in the top 10 recommendations and ± 5% improvement in the top 20 recommendations

Item Type: Thesis (Masters)
Uncontrolled Keywords: model topik, rekomendasi kolaborasi antardomain, random walk
Subjects: T Technology > T Technology (General) > T57.5 Data Processing
Divisions: Faculty of Information Technology > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User: Vit Zuraida
Date Deposited: 19 Jun 2021 12:22
Last Modified: 19 Jun 2021 12:22
URI: http://repository.its.ac.id/id/eprint/58146

Actions (login required)

View Item View Item