Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/aydan-moon/text_classification_imdb


https://github.com/aydan-moon/text_classification_imdb

Last synced: 19 days ago
JSON representation

Awesome Lists containing this project

README 

        
# Text Classification for IMDB Movie Reviews



### Table of Contents

1. [Project Overview](#project-overview)

2. [Project Objectives](#project-objectives)

3. [Dataset Description](#dataset-description)

4. [Project Workflow](#project-workflow)

   - [Data Cleaning and Preprocessing](#1-data-cleaning-and-preprocessing)

   - [Feature Extraction with TF-IDF](#2-feature-extraction-with-tf-idf)

   - [Model Training](#3-model-training)

   - [Model Evaluation](#4-model-evaluation)

5. [Results and Insights](#results-and-insights)

6. [Usage](#usage)

   - [Prerequisites](#prerequisites)

   - [Running the Project](#running-the-project)

   - [Example Code Snippet](#example-code-snippet)

7. [Project Structure](#project-structure)

8. [Future Improvements](#future-improvements)

9. [Contact](#contact)



---



### Project Overview

This project implements a text classification system to categorize IMDB movie reviews as either *positive* or *negative*. The focus is on creating a model optimized for limited computational resources, using classical machine learning models and a simple neural network. This project compares model performance to identify the most effective approach.



### Project Objectives

- Use the IMDB Movie Reviews dataset for binary sentiment classification.

- Implement data cleaning and preprocessing techniques.

- Create TF-IDF vectorization for feature extraction.

- Train and compare multiple classical ML models (Naive Bayes, Logistic Regression, SVM).

- Implement a neural network using Keras for sentiment classification.

- Perform hyperparameter tuning and k-fold cross-validation.

- Evaluate model performance using confusion matrices and ROC curves.



---



## Dataset Description

**Dataset**: [IMDB Movie Reviews](https://www.kaggle.com/datasets/lakshmi25npathi/imdb-dataset-of-50k-movie-reviews), a binary sentiment dataset with 50,000 reviews labeled as positive or negative.  

- **Positive Reviews**: Score ≥ 7 out of 10

- **Negative Reviews**: Score ≤ 4 out of 10



---



## Project Workflow



### 1. Data Cleaning and Preprocessing

- Removed HTML tags, non-alphabetic characters, and stop words.

- Applied lemmatization to convert words to their base forms.

- Generated word clouds to visualize the most common words in positive and negative reviews.



### 2. Feature Extraction with TF-IDF

- Used TF-IDF vectorization to transform text into a feature matrix.

- Limited features to the top 5,000 words to optimize for memory and processing efficiency.



### 3. Model Training

- Trained and evaluated the following models:

  - **Naive Bayes**

  - **Logistic Regression**

  - **Support Vector Machine (SVM)**

  - **Simple Neural Network (Keras)**



### 4. Model Evaluation

- Evaluated models using accuracy, precision, recall, F1-score, and AUC.

- Created confusion matrices and ROC curves for detailed performance insights.

- Performed hyperparameter tuning using GridSearchCV and k-fold cross-validation.



---



## Results and Insights

| Model               | Accuracy | Precision | Recall | F1-Score | AUC   |

|---------------------|----------|-----------|--------|----------|-------|

| Naive Bayes         | 0.8469   | 0.8449    | 0.8527 | 0.8488   | 0.9233|

| Logistic Regression | 0.8798   | 0.8686    | 0.8972 | 0.8827   | 0.9524|

| SVM                 | 0.8845   | 0.8719    | 0.9036 | 0.8874   | 0.9555|

| Neural Network      | 0.8716   | 0.8566    | 0.8950 | 0.8754   | 0.9427|



**Best Model**: SVM demonstrated the highest scores across precision, recall, F1-Score, and AUC, making it the most effective model for this dataset.



### Key Insights

- **SVM** provided the best overall performance, particularly in handling complex text patterns.

- **Logistic Regression** and the **Neural Network** also performed well and achieved good balance across metrics.

- **Naive Bayes** showed strong performance but was slightly behind in comparison with the other models.



---



## Usage



### Prerequisites

- Python 3.7+

- Install the required libraries:

  ```bash

  pip install pandas numpy matplotlib seaborn spacy scikit-learn tensorflow tabulate wordcloud



## Running the Project



**1. Load the Dataset**

- Place the IMDB dataset CSV file in the project folder, and load it using Pandas:

  ```bash 

  import pandas as pd

  df = pd.read_csv('IMDB Dataset.csv')



**2. Clean and Preprocess Text Data**

- Use the clean_review function to preprocess each review by removing unnecessary characters and lemmatizing words.



**3. Train Models**

- Apply TF-IDF vectorization, then train models including Naive Bayes, Logistic Regression, and SVM.



**3. Evaluate and Compare Models**

- Generate confusion matrices, ROC curves, and compare performance metrics.



## Example Code Snippet



- ### Example: Training SVM Model

  ```bash

  from sklearn.svm import SVC

  svm = SVC()

  svm.fit(X_train, y_train)

  svm_pred = svm.predict(X_test)



  from sklearn.metrics import accuracy_score

  print("SVM Accuracy:", accuracy_score(y_test, svm_pred))



## Project Structure

- nlp_02.py: Main code file with data loading, preprocessing, model training, and evaluation.

- README.md: Project documentation.

- svm_model.pkl: Saved SVM model file.

- IMDB Dataset.csv: IMDB dataset file used for training and evaluation.



## Future Improvements

- Experiment with more advanced NLP models, such as Transformer-based models (e.g., BERT) for improved accuracy.

- Implement additional data cleaning steps, like removing rare words or applying further regularization.

- Try ensemble methods combining multiple models for potentially better results.



## Contact

For questions or collaboration, please reach out at 📧 [email protected].