metadata

tags:
  - indobert
  - sentiment-analysis
  - text-classification
  - social-media
  - indonesian
  - django
  - fine-tuned
  - academic-evaluation
model-index:
  - name: IndoBERT Sentiment Classifier For University Review - University XYZ
    results:
      - task:
          type: text-classification
          name: Sentiment Analysis
        dataset:
          name: Online Lecture Sentiment Dataset
          type: custom-dataset
        metrics:
          - type: accuracy
            value: 0.89
            name: Accuracy
          - type: f1
            value: 0.88
            name: F1 Score
          - type: precision
            value: 0.87
            name: Precision
          - type: recall
            value: 0.89
            name: Recall

IndoBERT Sentiment Classifier for Social Media Posts – Universitas XYZ

This model is a fine-tuned IndoBERT transformer for performing sentiment analysis on Indonesian social media text (Twitter) related to university services. It classifies input text into positive, neutral, or negative sentiment categories.

🧠 Model Description

The model is built upon indobert-base-p2, a BERT-based transformer pre-trained on over 220 million Indonesian words. The fine-tuning process was done on 7500 samples containing balanced sentiment labels related to online academic services.

Label classes: Positive, Neutral, Negative
Preprocessing: Case folding, punctuation removal, stopword removal, stemming, tokenization (using IndoBERT tokenizer)

✅ Intended Use

Analyzing Indonesian tweets about universities
Sentiment-driven dashboards for academic service quality
NLP applications in education sector

⚠️ Limitations

Domain-specific to university-related sentiment
May not generalize well to informal or slang-heavy text
Sarcasm or mixed-sentiment detection is not supported
Doesn’t handle toxicity or hate speech detection

📊 Dataset

Source: Custom crawled tweets via keywords and hashtags (e.g. #telkomuniversity, Universitas XYZ)
Size: 7500 samples
Split: 70% train, 10% validation, 20% test
Labels: 2500 positive, 2500 neutral, 2500 negative
Language: Indonesian

⚙️ Training Procedure

Hyperparameters

Learning rate: 5e-5
Batch size: 8
Epochs: 3
Optimizer: Adam (β1=0.9, β2=0.999, ε=1e-8)
Scheduler: Linear
Seed: 42

Framework Versions

Transformers: 4.24.0
PyTorch: 1.13.0
Tokenizers: 0.13.2

📈 Evaluation Metrics

Metric	Score
Accuracy	89%
F1 Score	88%
Precision	87%
Recall	89%

💻 Deployment Context

This model was integrated into a Django-based sentiment dashboard application with:

A custom Twitter crawler
Real-time sentiment classification
Wordclouds and sentiment breakdowns by time period
Admin tools for filtering, deleting, and exporting data

📄 Citation

If you use this model or its components, please cite:

@article{wijaya2023indobert,
  author    = {Kurniadi Ahmad Wijaya and Ade Romadhony and Donni Richasdy},
  title     = {Implementasi Model IndoBERT pada Dashboard Sentimen Media Sosial (Studi Kasus Universitas XYZ)},
  journal   = {eProceedings of Engineering},
  volume    = {10},
  number    = {4},
  year      = {2023},
  month     = {September},
  url       = {https://openlibrary.telkomuniversity.ac.id},
}