Sentiment Analysis of Tokopedia App Reviews by Fajar SatriaSentiment Analysis of Tokopedia App Reviews by Fajar Satria

Sentiment Analysis of Tokopedia App Reviews

Fajar Satria

Data Scientist

pandas

Python

TensorFlow

Electronics

Like this project

Posted Jun 7, 2025

Performed sentiment analysis on Tokopedia app reviews using machine learning and deep learning.

Likes

Views

Clients

Tokopedia

Sentiment Analysis of Tokopedia App Reviews📱

This project performs web scraping of Tokopedia app reviews from the Google Play Store and builds sentiment analysis models using various machine learning and deep learning approaches.

📁 Project Structure

scraping.ipynb

Notebook for scraping review data from Google Play Store using google-play-scraper.

sentiment_model.ipynb

Notebook containing:

Data exploration

Automatic sentiment labeling based on ratings

Text preprocessing

Implementation and evaluation of 3 model schemes:

TF-IDF + SVM

Word2Vec + Random Forest

LSTM (Deep Learning)

🛠️ Technologies & Libraries

Python

pandas, numpy, re, string

nltk (stopwords)

scikit-learn

gensim (Word2Vec)

tensorflow / Keras (LSTM)

google_play_scraper

🔍 Steps

1. Scraping Tokopedia Reviews

Using google_play_scraper to fetch up to 10,000 user reviews for the Tokopedia app from Google Play Store (Indonesia localization).

from google_play_scraper import reviews
Output is saved in the file data_tokopedia_reviews.csv.

2. Preprocessing & Labeling

Remove unnecessary characters, numbers, punctuation, and stopwords.

Sentiment labels are assigned based on ratings:

⭐ 1–2 → Negative

⭐ 3 → Neutral

⭐ 4–5 → Positive

3. Built Model Schemes

✅ Scheme 1: TF-IDF + SVM TF-IDF vectorization

Linear Support Vector Classifier (SVM)

Evaluation: accuracy & classification report (sklearn)

🌳 Scheme 2: Word2Vec + Random Forest Word embedding using Word2Vec (gensim)

Each review represented as the average of word vectors

Classification model: Random Forest

🧠 Scheme 3: LSTM (Deep Learning) Tokenization and sequence padding

Architecture: LSTM + Bidirectional + Dropout

Final activation: Softmax for 3 classes

📈 Results

TF-IDF + SVM Accuracy    : 0.989
Word2Vec + RF Accuracy   : 0.989
LSTM Accuracy            : 0.9905
Results may vary depending on dataset & random state.

🧪 Inference

# TF-IDF + SVM
"This product is very disappointing and does not match the description" → Prediction: NEGATIVE

# Word2Vec + RF
"fast and friendly service" → Prediction: POSITIVE

# LSTM
"This product is really cool" → Prediction: POSITIVE

📌 Notes

The dataset is dynamic and depends on the latest reviews.

LSTM is suitable for large-scale data but requires longer training time.

Word2Vec is good for generating reusable embeddings for other projects.

📂 Output

data_tokopedia_reviews.csv — Dataset from scraping

Inference models are available at the end of each scheme notebook

📄 License

This project is created for educational and research purposes. Data is publicly sourced from Google Play Store. Use responsibly.