For this project I got access to the tweets related to the 2019 Canadian Election, the dataset for tweets was rather small so the main job was to use a generic tweets sentiment dataset sentimenti.csv and then train on this generic-tweet dataset and see how well it performs on the election related tweets.
This is my first NLP project which is why I have not used tokenizer (used for tranformers) and more involved approaches like using trained word embeddings. For this project I have used work-freuqncy and Tf-idf, to encode tweets into numbers.
For modelling a grid search is performed on the common machine-learning models like,
- kNN
- Naive Bayes
- Logistic Regression
- Random Forest
- Decision Tree
For each algorithm a 5-fold cross validation is performed to tune the hyperparameters. More approaches like light-GBM and XG-Boost will be added later.