Multiclass Sentiment Classifier

This model is a multiclass text classifier, which classifies a comment as a positive, negative, or neutral comment.

To train the model, we used sentiment analysis dataset. The dataset is open source and can be found in kaggle: https://www.kaggle.com/datasets/abhi8923shriv/sentiment-analysis-dataset.

We used TensorFlow as the ML framework.

Requirements

You need the following libraries to use the classifier:

• Tensorflow
• NumPy

How to install Tensorflow: https://www.tensorflow.org/install

How to install NumPy: https://numpy.org/install/

Train

To train the model, open the model_trainer.py and there you can enter your own dataset with your own parameters.

Please be careful with the VOCAB_SIZE and MAX_LEN since they are related with tokenizer file and dataset respectfully:

• VOCAB_SIZE: Number of tokens one allow in the token dictionary. The whole project must use the same VOCAB_SIZE.
• MAX_LEN: Lenght of the each sample. Each sample needs to have same size since they enter the Neural Network with input neurons of MAX_LEN.

We cannot offer the cleaned datasets because of the large size for GitHub, but you can create them with sentiment analysis dataset/cleaned_ds/dataset_cleaner.py.

Be sure to open the project in root folder of the project, or the directories might not work.

To increase the training time, increase the EPOCH parameter.

One can play with different optimizers and batch_sizes. More information on optimizers: https://www.tensorflow.org/api_docs/python/tf/keras/optimizers

Classify

To use the final model run the classifier.py. It does not have an user interface and for now work on the terminal.