TOPIC MODELING PADA JURNAL PINTER MENGGUNAKAN METODE LATENT DIRICHLET ALLOCATION

DAVIN ARAZI, . (2026) TOPIC MODELING PADA JURNAL PINTER MENGGUNAKAN METODE LATENT DIRICHLET ALLOCATION. Sarjana thesis, UNIVERSITAS NEGERI JAKARTA.

	Text COVER.pdf Download (757kB)
	Text BAB I.pdf Download (870kB)
	Text BAB II.pdf Restricted to Registered users only Download (985kB) | Request a copy
	Text BAB III.pdf Restricted to Registered users only Download (737kB) | Request a copy
	Text BAB IV.pdf Restricted to Registered users only Download (4MB) | Request a copy
	Text BAB V.pdf Restricted to Registered users only Download (493kB) | Request a copy
	Text DAFTAR PUSTAKA.pdf Download (468kB)
	Text lampiran.pdf Restricted to Registered users only Download (1MB) | Request a copy

Abstract

Meningkatnya volume publikasi ilmiah dalam Jurnal PINTER menyebabkan kesulitan dalam mengidentifikasi distribusi tema penelitian secara manual, sementara belum terdapat mekanisme otomatis berbasis komputasi yang diadopsi untuk pemetaan topik pada jurnal tersebut. Kondisi ini mendorong penerapan pendekatan text mining berbasis topic modeling menggunakan metode Latent Dirichlet Allocation (LDA) yang mampu mengidentifikasi struktur tematik tersembunyi dalam korpus dokumen secara probabilistik, sekaligus membandingkan tiga metode representasi fitur yaitu TF-IDF, FastText, dan kombinasi TF-IDF+FastText untuk mengetahui pendekatan yang menghasilkan kualitas topik terbaik. Proses penelitian dilakukan melalui tahapan Knowledge Discovery in Databases (KDD) yang meliputi selection, preprocessing, transformation, data mining, dan evaluation, dengan menggunakan dataset 160 artikel full text dari Jurnal PINTER Volume 1 Nomor 1 hingga Volume 8 Nomor 2. Tahap preprocessing mencakup data cleansing, case folding, tokenization, stemming, dan stopword removal menggunakan pustaka Sastrawi. Pengujian dilakukan pada dua skenario data, yaitu berbasis volume dan berbasis jurnal, dengan variasi jumlah topik K = 3, K = 4, dan K = 5. Evaluasi model menggunakan coherence score (C_v) untuk mengukur keterkaitan semantik antar kata dalam topik dan perplexity untuk mengukur kemampuan generalisasi model. Hasil penelitian menunjukkan bahwa metode TF-IDF menghasilkan nilai perplexity tertinggi pada seluruh konfigurasi (rata-rata 233,990 berbasis volume), mengindikasikan keterbatasan representasi berbasis frekuensi dalam memodelkan distribusi probabilistik kata. Metode FastText menunjukkan performa yang lebih seimbang dengan titik optimal pada K = 4 berbasis volume (coherence = 0,7379; perplexity = 38,701). Kombinasi TF-IDF+FastText terbukti menghasilkan performa terbaik secara konsisten, dengan model optimal pada K = 5 berbasis volume (coherence = 0,6709; perplexity = 16,793) dan K = 4 berbasis jurnal (coherence = 0,6203; perplexity = 40,060). Berdasarkan topik yang terbentuk, tema riset dominan dalam Jurnal PINTER mencakup jaringan komputer dan analisis kualitas layanan, media pembelajaran berbasis teknologi, pengembangan sistem dan aplikasi, serta evaluasi dan pengukuran dalam Pendidikan teknik informatika. Kata Kunci: topic modeling, Latent Dirichlet Allocation, TF-IDF, FastText, coherence score, perplexity, Jurnal PINTER ***** The growing volume of scientific publications in PINTER Journal has made it increasingly difficult to manually identify the distribution of research themes, while no automated computational mechanism has been adopted for topic mapping in the journal. This condition prompted the application of a text mining approach based on topic modeling using the Latent Dirichlet Allocation (LDA) method, which is capable of probabilistically identifying hidden thematic structures within a document corpus, while simultaneously comparing three feature representation methods TF-IDF, FastText, and the TF-IDF+FastText combination to determine the approach that yields the best topic quality. The research was conducted through the Knowledge Discovery in Databases (KDD) stages comprising selection, preprocessing, transformation, data mining, and evaluation, using a dataset of 160 full-text articles from PINTER Journal Volume 1 Number 1 through Volume 8 Number 2. The preprocessing stage included data cleansing, case folding, tokenization, stemming, and stopword removal using the Sastrawi library. Experiments were conducted under two data scenarios volume-based and journal based with topic count variations of K = 3, K = 4, and K = 5. Model evaluation employed coherence score (C_v) to measure semantic relatedness among words within a topic, and perplexity to assess the model's generalization capability. Results show that TF-IDF produced the highest perplexity values across all configurations (average 233.990 in the volume-based scenario), indicating the limitations of frequency-based representation in modeling probabilistic word distributions. FastText demonstrated more balanced performance with an optimal point at K = 4 in the volume-based scenario (coherence = 0.7379; perplexity = 38.701). The TF-IDF+FastText combination consistently produced the best performance, with the optimal model at K = 5 in the volume-based scenario (coherence = 0.6709; perplexity = 16.793) and K = 4 in the journal-based scenario (coherence = 0.6203; perplexity = 40.060). Based on the identified topics, the dominant research themes in PINTER Journal include computer networks and quality of service analysis, technology-based learning media, system and application development, and evaluation and measurement in informatics and computer engineering education. Keywords: topic modeling, Latent Dirichlet Allocation, TF-IDF, FastText, coherence score, perplexity, PINTER Journal

Item Type:	Thesis (Sarjana)
Additional Information:	1). Dr. Widodo, S.Kom., M.Kom. ; 2). Neng Ayu Herawati, S.Pd., M.T.
Subjects:	Sains > Sains, Ilmu Pengetahuan Alam Sains > Statistika Teknologi dan Ilmu Terapan > Teknik Komputer
Divisions:	FT > S1 Pendidikan Teknik Informatika Komputer
Depositing User:	Users 35035 not found.
Date Deposited:	10 Aug 2026 07:38
Last Modified:	10 Aug 2026 07:38
URI:	http://repository.unj.ac.id/id/eprint/69139

Actions (login required)

View Item