A bunch of different models on 25k Large Movie Review (IMDb) dataset. Results are reported in the attached paper.
Install dependencies: pip install -r requirements.txt
IF INITIAL RUN, in retrieve_and_pre function in each model file, make fromsave=False, as this will initialize the preprocessing step in the feature extraction pipeline, as well as save a local copy of the preprocessed data onto the machine. For the biLSTM model, set retrieve_and_pre parameter tfidfpca=False, as we want the function to return raw preprocessed sentences instead of scaled TFxIDF features.
To run model: python3 -m model.
Most models are validated using the standard 10-fold cross validation scheme, and average accuracy, F1-score, and standard precision/recall are reported.
For more information and analysis, check the .docx file.