How lingadeu/sentiment-model-experiment works
Project type and purpose
• This is a data-science / machine-learning research project that builds and compares several sentiment-analysis models (conventional ML, LSTM, and fine-tuned IndoBERT) on product reviews from Tokopedia.
• The goal is to automate extraction of binary sentiment (positive/negative) and to compare performance across vectorizer+ML, RNN, and transformer-based approaches.
File structure
• Top level:
• data/ — raw and cleaned CSV files
• notebook/ — series of Jupyter notebooks for analysis and experiments (01_analysis, 02_tfidf-and-ml-models, 03_tfidf-and-lstm, 04_embeddings-and-IndoBERT)
• model/ — serialized artifacts from best conventional ML experiments (.pkl) and notes about the HuggingFace-hosted IndoBERT model
• img/ — supporting images (e.g., header.png)
• README.md and .gitignore
• No web frontend or server code—this is a modeling pipeline rather than a deployed service.
README for additional insights
• README outlines three experimental paths:
a. Quantitative text analysis & exploration
b. Conventional ML with TF-IDF + SVM
c. Deep-learning LSTM
d. Transformer fine-tuning (IndoBERT) stored on HuggingFace
• It describes key results, dependencies (e.g., scikit-learn, TensorFlow/PyTorch, HuggingFace Transformers), and folder organization.