Naive Bayes Classifier for Sentiment Analysis

This project involves the development of a Naive Bayes Classifier (NBC) for sentiment analysis using a dataset that includes reviews along with their corresponding star ratings. The dataset comprises two files, "train.csv" and "test.csv," which will be supplied for this project. In this context, a review with a 5-star rating is categorized as expressing a positive sentiment, whereas all other ratings are treated as indicating a negative sentiment.

Description

Feature Selection:

• Preprocess "train.csv" dataset.
• Select the top 1000 most frequently occurring words as features for the Naive Bayes Classifier model.
• Exclude all other words not in the top 1000 by frequency.
• Print a list of the selected features, typically the top 20 to 50 words.

Model Training and Evaluation:

• Utilize both "train.csv" and "test.csv" datasets.
• Train and evaluate the Naive Bayes Classifier.
• Apply Laplace Smoothing during parameter estimation.
• For an attribute Xi with k values, incorporate Laplace correction by adding 1 to the numerator and k to the denominator of the maximum likelihood estimate.

Learning Curve Analysis

• Create a learning curve by varying the percentage of training data used (e.g., 10%, 30%, 50%, 70%, 100%).
• Keep the testing set unchanged.

file discription

• main.py: main python file so in this file task is running. There are total 3 task in this project. And in main.py the task can be checked. In order to check the result just enter the following code in the terminal
• preprocessing.py: this file contains a class having the variable and method to perform task1. To preprocess the data set
• training.py: this file contains a class having the variable and method to perform task2. This file contains Naive Bayes Classification training method and prediction method
• analysis.py: this file contains a class having the variable and method to perform task3. This file contains visualization and optimization methods.

Result

Tokenlixze
Tokenize all the review and get 1000 most frequently used words in the train data to use these words as a feature. And print 20 most frequently from 1000 words.

Train and predict new dataset
I trained the model using train data and using test data predict the rating (5- star, 1-star) and compare the result with real rating.

Visualize Learning Curve