This project focuses on sentiment analysis using machine learning classifiers SVM and MLP to investigate potential gender biases in the provided dataset.
The dataset comprises training, validation, and test sets available in three formats: raw, embeddings, and TF-IDF.
- pandas
- matplotlib
- sklearn
The datasets for training, validation, and test are loaded using pandas.
An 'Unnamed' column is dropped from each dataset, if present.
- Distribution of ratings in the training dataset.
- Gender-based distribution of ratings.
- ZeroR classifier is initialized and trained on the training data.
- Accuracy is evaluated on the validation data, further broken down by gender.
- MLP is trained using the TF-IDF and embeddings representation of the training data.
- Performance is evaluated on the validation dataset using accuracy, confusion matrix, and a classification report.
- Performance is also assessed separately on male and female reviews using the embeddings representation.
- SVM is trained using the TF-IDF and embeddings representation of the training data.
- Performance is evaluated on the validation dataset using accuracy, confusion matrix, and a classification report.
- Performance is also assessed separately on male and female reviews using the embeddings representation.
- SVM with an RBF kernel is trained on the embeddings representation of the training data.
- Predictions are made on the test dataset using this SVM.
- Submission: Predictions from the SVM are saved into a submission DataFrame.
Ensure all the required libraries are installed and the datasets (TRAIN.csv, VALIDATION.csv, TEST_NO_LABELS.csv, and their embeddings and TF-IDF versions) are available in the working directory.
Execute the Python cells to run the entire sentiment analysis in jupyter notebook.