yuanhuan177/project_amazon_reviews_nlp_recommender_system

Mining , pre-processing and embedding over 1 million Amazon Movie & T.V. reviews to build a multi class Naive Bayes model and later a CNN-LSTM model (that uses the Naive Bayes model as a baseline) to predict rating from text. Interpreting the original classifier using local surrogate models using the LIME library. Using LDA topic modeling to build a theme based recommender from the reviews and using a model based collaborative filtering system using SVD matrix factorization to build a second recommender system.

Project_Amazon_reviews_NLP_recommender_system

Mining , pre-processing and embedding over 1 million Amazon Movie & T.V. reviews to build a multi class Naive Bayes model and later a CNN-LSTM model (that uses the Naive Bayes model as a baseline) to predict rating from text. Interpreting the original classifier using local surrogate models using the LIME library. Using LDA topic modeling to build a theme based recommender from the reviews and using a model based collaborative filtering system using SVD matrix factorization to build a second recommender system.

LIME Explainer Screenshots from fitted Naive Bayes Model on the review data (Not rendering in the .ipynb notebook on this repo)

The LIME explainer notebooks based on my model's prediction on unseen reviews and their explanations is for some reason not rendering on the jupyter notebook on my repo so I decided to add the screenshots of those cells below.

Everything else is in the notebook.

Enjoy :)

5-Star Review(Clearly Positive)

1-Star Review(Clearly Negative)

3-Star Review(Somewhere in the middle)