Dinata, Rafi Ammar (2025) PENERAPAN METODE WORD2VEC DAN RANDOM FOREST UNTUK ANALISIS SENTIMEN ULASAN PENGGUNA APLIKASI LIVIN’ BY MANDIRI. Skripsi thesis, UPN Veteran Yogyakarta.
![[thumbnail of 1_SKRIPSI FULLTEXT_RAFI AMMAR DINATA_123210145.pdf]](http://eprints.upnyk.ac.id/style/images/fileicons/text.png) |
Text
1_SKRIPSI FULLTEXT_RAFI AMMAR DINATA_123210145.pdf Restricted to Repository staff only Download (4MB) |
|
Text
2_COVER_123210145_RAFI AMMAR DINATA.pdf Download (139kB) |
|
Text
3_ABSTRAK_123210145_RAFI AMMAR DINATA.pdf Download (179kB) |
|
Text
4_PENGESAHAN PENGUJI_RAFI AMMAR DINATA_123210145.pdf Download (466kB) |
|
Text
5_PENGESAHAN PEMBIMBING_RAFI AMMAR DINATA_123210145.pdf Download (427kB) |
|
Text
6_DAFTAR ISI_123210145_RAFI AMMAR DINATA.pdf Download (154kB) |
|
Text
7_DAFTAR PUSTAKA_123210145_RAFI AMMAR DINATA.pdf Download (169kB) |
Abstract
ABSTRAK
Kemajuan teknologi memberikan pengaruh signifikan terhadap berbagai sektor
kehidupan, termasuk sektor perbankan. Dalam menghadapi era digital, institusi perbankan
berupaya meningkatkan kualitas layanan dan produk melalui inovasi yang membedakan diri
dari para kompetitornya. Analisis sentimen terhadap ulasan pengguna aplikasi mobile
banking menjadi sangat penting untuk mengevaluasi kualitas layanan dan kepuasan
pengguna.
Penelitian ini bertujuan untuk mengkaji efektivitas penerapan metode Word2Vec dan
algoritma Random Forest dalam mengklasifikasikan sentimen ulasan pengguna aplikasi
Livin’ by Mandiri. Data yang digunakan sebanyak 60.000 ulasan berbahasa Indonesia yang
diperoleh melalui teknik web scraping dari Google Play Store. Proses preprocessing
dilakukan melalui tahapan cleansing, case folding, tokenization, word normalization,
stemming, negation handling dan stopword removal. Selanjutnya, data diubah menjadi
representasi vector menggunakan dua arsitektur Word2Vec, yakni Continuous Bag of Words
(CBOW) dan Skip-Gram, dengan berbagai kombinasi parameter seperti vector size, window,
dan epochs. Representasi vector tersebut menjadi input bagi model Random Forest untuk
klasifikasi sentimen ke dalam kelas positif dan negatif.
Hasil penelitian menunjukkan bahwa kombinasi Word2Vec dan Random Forest
secara umum mampu menghasilkan performa klasifikasi yang baik. Arsitektur Skip-Gram
menunjukkan performa yang sedikit lebih tinggi dibandingkan CBOW, dengan akurasi
terbaik dengan accuracy 87.31%, f1-score 87.28% dan k-fold cross validation 86.98% pada
kombinasi vector size 200, window 15, dan epoch 5. Sedangkan CBOW mencapai akurasi
terbaik dengan accuracy 86.70%, f1-score 86.68% dan k-fold cross validation 86.79% pada
kombinasi vector size 300, window 5, dan epoch 10. Skip-Gram terbukti lebih unggul dalam
menangkap konteks semantik terutama pada ulasan pendek yang mengandung negasi atau
opini implisit. Secara keseluruhan, pemilihan parameter Word2Vec memiliki pengaruh
signifikan terhadap kinerja model klasifikasi. Vector size yang besar meningkatkan kualitas
representasi pada Skip-Gram, tetapi dapat memperburuk performa pada CBOW jika terlalu
kompleks. Window kecil cenderung memberikan hasil terbaik karena sesuai dengan konteks
pendek dalam ulasan aplikasi. Sementara itu, epoch yang terlalu tinggi berisiko
menyebabkan overfitting, khususnya pada Skip-Gram. Dengan demikian, efektivitas metode
Word2Vec dan Random Forest dalam analisis sentimen sangat dipengaruhi oleh kombinasi
parameter yang digunakan, yang perlu disesuaikan dengan karakteristik dataset secara
spesifik.
Kata Kunci: Analisis Sentimen, Word2Vec, Random Forest, Livin’ by Mandiri, Ulasan
Pengguna
vii
ABSTRACT
Technological advances have had a significant impact on various sectors of life,
including the banking sector. In the digital age, banking institutions are striving to improve
the quality of their services and products through innovations that differentiate them from
their competitors. Sentiment analysis of user reviews of mobile banking applications has
become very important for evaluating service quality and user satisfaction.
This study aims to examine the effectiveness of applying the Word2Vec method and
Random Forest algorithm in classifying the sentiment of user reviews of the Livin' by Mandiri
app. The data used consists of 60,000 Indonesian-language reviews obtained through web
scraping from the Google Play Store. The preprocessing process was carried out through
the following stages: cleansing, case folding, tokenization, word normalization, stemming,
negation handling, and stopword removal. Next, the data was converted into vector
representations using two Word2Vec architectures, namely Continuous Bag of Words
(CBOW) and Skip-Gram, with various parameter combinations such as vector size, window,
and epochs. These vector representations were used as input for the Random Forest model
to classify sentiment into positive and negative classes.
The results of the study indicate that the combination of Word2Vec and Random
Forest generally produces good classification performance. The Skip-Gram architecture
showed slightly higher performance than CBOW, with the best accuracy of 87.31%, f1-score
of 87.28%, and k-fold cross validation of 86.98% in the combination of vector size 200,
window 15, and epoch 5. Meanwhile, CBOW achieved the best accuracy with 86.70%, f1
score 86.68%, and k-fold cross validation 86.79% in the combination of vector size 300,
window 5, and epoch 10. Skip-Gram proved to be superior in capturing semantic context,
especially in short reviews containing negation or implicit opinions. Overall, the selection
of Word2Vec parameters has a significant impact on the performance of the classification
model. A large vector size improves the quality of representation in Skip-Gram, but can
worsen performance in CBOW if it is too complex. A small window tends to yield the best
results because it aligns with the short context in app reviews. Meanwhile, an excessively
high epoch risks causing overfitting, especially in Skip-Gram. Thus, the effectiveness of the
Word2Vec and Random Forest methods in sentiment analysis is greatly influenced by the
combination of parameters used, which must be tailored to the specific characteristics of the
dataset.
Keywords: Sentiment Analysis, Word2Vec, Random Forest, Livin’ by Mandiri, User Reviews
viii
| Item Type: | Tugas Akhir (Skripsi) |
|---|---|
| Uncontrolled Keywords: | Sentiment Analysis, Word2Vec, Random Forest, Livin’ by Mandiri, User Reviews |
| Subjek: | Z Bibliography. Library Science. Information Resources > ZA Information resources |
| Divisions: | Fakultas Teknik Industri > (S1) Informatika |
| Depositing User: | Eko Yuli |
| Date Deposited: | 17 Oct 2025 01:43 |
| Last Modified: | 17 Oct 2025 01:43 |
| URI: | http://eprints.upnyk.ac.id/id/eprint/44510 |
Actions (login required)
 |
View Item |