This project aims to classify the sentiment (negative, neutral, positive) of 8 Ball Pool game reviews on the Play Store. By analyzing the reviews, we can gain insights into the overall user sentiment towards the game and identify areas for improvement. The classification model will help categorize the reviews accurately, enabling developers to make data-driven decisions to enhance the gaming experience.
This project utilizes a dataset of over 90,000 reviews scraped from the Play Store. After performing extensive data cleaning and addressing the issue of unbalanced data, the final dataset consists of a total of 24,000 reviews. This refined dataset will be used for sentiment classification, allowing for accurate analysis of user sentiment towards the 8 Ball Pool game. You can see the code used to scrape the data in this notebook
The project uses the following libraries:
- Scraping the reviews from the Play Store using the
google-play-scraperlibrary - Performing natural language processing tasks with the
nltklibrary - Performing sentiment analysis using the
textbloblibrary - Handling numerical operations with the
numpylibrary - Manipulating and analyzing data using the
pandaslibrary - Visualizing data using the
matplotliblibrary - Creating visualizations with the
seabornlibrary - Generating word clouds using the
wordcloudlibrary - Implementing machine learning models with the
scikit-learnlibrary - Building deep learning models with the
tensorflowlibrary
The sentiment analysis process involves several key steps, including data preprocessing, labelling, imbalance data handling, feature extraction, and model training. The dataset is preprocessed to remove noise, and tokenize the text. The text data is then transformed into feature vectors using techniques such as TF-IDF and Tokenizer. These features are fed into machine learning and deep learning models such as Logistic Regression, LSTM, and Bi-LSTM for sentiment classification.
As you can see in the notebook, the sentiment classification models were evaluated using different methodologies and achieved the following accuracy scores:
- Experiment 1 (LR + 90-10): 89.38%
- Experiment 2 (LSTM + 80-20): 91.73%
- Experiment 3 (Bi-LSTM + 70-30): 91.97%
These results demonstrate the effectiveness of the models in accurately classifying the sentiment of 8 Ball Pool game reviews. The highest accuracy was achieved by Experiment 3, which utilized a Bi-LSTM model with a 70-30 train-test split.
The project also includes a Streamlit web app that allows users to input their own review and receive a sentiment prediction. The app is can be accessed here.