Variabel numerik adalah variabel yang memiliki nilai berupa angka dan dapat dilakukan operasi matematis. Disini ditemukan 8 variabel yang dapat dikategorikan sebagai variabel numerik seperti yang ada pada tabel berikut:
| Variabel | Deskripsi |
|---|---|
| Student_ID | ID unik untuk setiap siswa |
| Age | Usia siswa (tahun) |
| Study_Hours_per_Week | Jumlah jam belajar per minggu |
| Attendance | Persentase kehadiran siswa (%) |
| Family_Income_BDT | Pendapatan keluarga (dalam BDT) |
| Previous_GPA | Nilai GPA sebelumnya |
| SSC_Result | Nilai ujian SSC |
| HSC_Result | Nilai ujian HSC |
Variabel numerik adalah variabel yang berisi kategori atau tabel tertentu. Disini terdapat 6 variabel yang dapat dikategorikan sebagai variabel numerik seperti yang ada pada tabel berikut:
| Variabel | Deskripsi | Contoh Nilai |
|---|---|---|
| Gender | Jenis kelamin siswa | Male, Female |
| District | Wilayah tempat tinggal siswa | Rangpur, Dhaka, dll |
| School_Type | Jenis sekolah | Private, Public |
| Parent_Education | Tingkat pendidikan orang tua | Graduate, Undergraduate, dll |
| Internet_Access | Akses internet di rumah | Yes, No |
| Private_Tuition | Mengikuti les privat atau tidak | Yes, No |
Disini statistik dasar dari dataset utamanya variabel numerik akan diuraikan satu demi satu.
🔢 Ukuran Pemusatan
📉 Ukuran Penyebaran
📌 Nilai Ekstrem
🔢 Ukuran Pemusatan
📉 Ukuran Penyebaran
📌 Nilai Ekstrem
🔢 Ukuran Pemusatan
📉 Ukuran Penyebaran
📌 Nilai Ekstrem
🔢 Ukuran Pemusatan
📉 Ukuran Penyebaran
📌 Nilai Ekstrem
🔢 Ukuran Pemusatan
📉 Ukuran Penyebaran
📌 Nilai Ekstrem
🔢 Ukuran Pemusatan
📉 Ukuran Penyebaran
📌 Nilai Ekstrem
This project is a Final Project that implements a TextCNN to perform sentiment analysis on TikTok comments in Indonesian language. The system is designed to classify comments into two categories: Cyberbullying (insult/embarrass content comments) and Non-Cyberbullying (normal/clear content comments).
Institution: Institut Teknologi Sumatera (ITERA)
Study Program: Informatics Engineering
Author: Nikola Arinanda
Year: 2026
This project uses TextCNN architecture to analyze YouTube comments sentiment. Initially, the data will go through preprocessing stages such as data division (80:20), case folding, text cleaning, augmentation (AEDA, random swap character, random delete character), tokenization and stopword removal to prepare the data. The next step is model training using k-fold cross-validation as many as 5 fold to ensure robustness and good generalization. Final step is model evaluation using confussion matrix such as accuracy, precision, recall and F1 score.
tugas-akhir-main/
├── dataset/
│ ├── k_fold.json # k-fold cross-validation dictionary
│ └── cyberbullying.csv # Original dataset
├── code/
│ ├── datareader.py # Data loader and preprocessing
│ ├── model.py # Model architecture
│ └── train.py # Main script for model training
├── model_outputs/
│ ├── run_YYYYMMDD_HHMMSS/
│ │ ├── fold_1_model.pth # Model output
│ │ ├── fold_2_model.pth # ...
│ │ ├── fold_3_model.pth
│ │ ├── fold_4_model.pth
│ │ ├── fold_5_model.pth
│ │ └── ...
│ └── ...
└── report/
│ └── thesis.pdf # Documentation and reports
└── requirements.txt # Python dependencies
This project requires:
All dependencies are listed in the requirements.txt file. Main libraries:
| Library | Version | Purpose |
|----------------|----------|--------------------------------------|
| torch | >=2.0.0 | Deep learning framework |
| pandas | >=1.5.0 | Data manipulation |
| numpy | >=1.23.0 | Numerical computing |
| matplotlib | >=3.7.0 | Data visualization |
| seaborn | >=0.12.0 | Statistical visualization |
| scikit-learn | >=1.2.0 | Machine learning utilities |
| transformers | >=4.30.0 | NLP models (IndoBERT, etc.) |
| nltk | >=3.8.0 | Text preprocessing |
| tqdm | >=4.65.0 | Progress bar |
| wandb | >=0.15.0 | Experiment tracking |
For the complete list, see requirements.txt
git clone https://github.com/nikolaarinanda/tugas-akhir.git
cd tugas-akhir
It is highly recommended to use a virtual environment to avoid dependency conflicts. Using venv (built-in python):
# Linux/Mac
python3 -m venv venv
source venv/bin/activate
# Windows
python -m venv venv
venv\Scripts\activate
Using conda:
conda create -n youtube-sentiment python=3.9
conda activate youtube-sentiment
# Upgrade pip to the latest version
pip install --upgrade pip
# Install all requirements
pip install -r requirements.txt
Note for PyTorch with GPU: If you want to use GPU, install the CUDA-specific version of PyTorch:
# For CUDA 11.8
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
# For CUDA 12.1
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
The dataset consists of TikTOk comments in Indonesian language which comes from this research with labels:
The dataset cyberbullying.csv has the following columns:
| Column | Type | Description |
|---|---|---|
| sentiment | Integer | label (-1 cyberbullying, 1 non-cyberbullying) |
| comment | String | Comment content |
py train.py
Result:
python train.py \
--max_length 128 \
--dropout 0.3 \
--batch_size 50 \
--optimizer_name Muon \
--embed_dim 100 \
--conv_filters 50 \
--kernel_size 3 4 \
--epochs 100 \
--lr 5e-4 \
| Argument | Type | Default | Description |
|---|---|---|---|
| –seed | int | 01012001 | Random seed for reproducibility |
| –dataset_path | str | ’../dataset/cyberbullying.csv’ | Path to dataset file |
| –max_length | int | 128 | Maximum sequence length |
| –tokenizer | str | ‘indobenchmark/indobert-base-p1’ | Tokenizer name |
| –dropout | float | 0.5 | Dropout rate |
| –batch_size | int | 50 | Batch size for embedding |
| –embed_dim | int | 100 | Embedding dimension for CNN |
| –num_classes | int | 2 | Number of classes |
| –conv_filters | int | 50 | Number of filters for CNN |
| –kernel_size | int | [3, 4] | Kernel sizes for CNN |
| –n_folds | int | 5 | Fold number for cross-validation |
| –epochs | int | 100 | Number of epochs |
| –lr | float | 52-4 | Learning rate |
| –output_model | flag | True | Save model after training |
| –output_dir | str | ‘model_outputs’ | Directory to save model outputs |
| –use_wandb | flag | False | Enable Weights & Biases logging |
| –wandb_group | str | ‘Light TextCNN’ | Create group for Weights & Biases runs |
| –wandb_note | str | ‘Light TextCNN Note’ | Add Weights & Biases notes |
| –patience | int | 5 | Patience for early stopping (epochs to wait after no improvement) |
model_outputs/
├── run_YYYYMMDD_HHMMSS/
│ ├── fold_1_model.pth # Model output
│ ├── fold_2_model.pth # ...
│ ├── fold_3_model.pth
│ └── fold_4_model.pth
│ └── fold_5_model.pth
│ └── ...
└── ...
The model produces the following metrics:
Augmentation techniques are applied during training to improve robustness:
Example:
Original: "makannya segentong buset"
AEDA: "makannya segentong buset!"
Random swap Character: "makannya segetnong buset"
Random delete Character: "makanya segentong buset"
If you encounter issues CUDA out of memory:
# Reduce batch size
python train.py --batch_size 8
# Reduce embedding dimension
python train.py --embedding_dim 64
If you encounter issues module not found:
# Make sure virtual environment is activated
# Reinstall dependencies
pip install -r requirements.txt --force-reinstall
If you encounter issues dataset file not found:
If you encounter issues downloading IndoBERT:
# Manual download
python -c "from transformers import AutoTokenizer; AutoTokenizer.from_pretrained('indobenchmark/indobert-base-p1')"
If you use or adapt this code/model in your research or publication, please use one of the following citation formats:
@thesis{arinanda2026cyberbullying,
title={Sentiment Analysis of Cyberbullying Comments on TikTok Social Media Using TextCNN Architecture},
author={Nikola Arinanda},
year={2026},
school={Institut Teknologi Sumatera (ITERA)},
type={Final Project},
address={Lampung, Indonesia}
}
Arinanda, N. (2026). Sentiment Analysis of Cyberbullying Comments on TikTok Social Media Using TextCNN Architecture [Final Project]. Institut Teknologi Sumatera (ITERA).
Arinanda, Nikola. “Sentiment Analysis of Cyberbullying Comments on TikTok Social Media Using TextCNN Architecture.” Final Project, Institut Teknologi Sumatera (ITERA), 2026.
Arinanda, Nikola. “Sentiment Analysis of Cyberbullying Comments on TikTok Social Media Using TextCNN Architecture.” Final Project, Institut Teknologi Sumatera (ITERA), 2026.
N. Arinanda, “Sentiment Analysis of Cyberbullying Comments on TikTok Social Media Using TextCNN Architecture”, Final Project, Institut Teknologi Sumatera (ITERA), 2026.
For questions or issues about this project, please:
Last Update: April 2026 Status: Active Development
]]>