Predict user scores(1-5) from their Reviews

• Used python evironment(anaconda)

pip install libraries

• scikit-learn

• Pandas

• NumPy

• os

• spacy

• nltk

• tensorflow, keras

• seaborn

• re

• Also need to download required nltk packages

Data is avaialable in https://www.kaggle.com/snap/amazon-fine-food-reviews, you need to put the data in a folder named "archive" in the same directory.

• Python files :

  • Amazon_Food_Review.ipynb : Main notebook.

• Python files :

  • textpreprocesser.py : Class with functions for preprocessing the text data.
  • textpreprocesser_tester.py : Unit tests for textpreprocesser.py
  • modelsk.py : Class to build machine learning pipelines by using sklearn.
  • modelsk_tester.py : Unit tests for modelsk.py
  • modelkeras.py : Class to build neural network models by using keras.
  • keras_plotters.py : Functions to plot learning curve for keras model.

• Loading & Exploring the data
• Preprocessing the data
  • Converting text to lower case
  • Removing html tags
  • Removing punctuations
  • Removing stop words
  • Lemmatization
• Creating Dictionary
  • Creating frequency dictionary
  • Get top n frequent words
  • Removing reviews longer than k words
• Building SKlearn Models
  • Naive Bayes
  • Support Vector Machine
  • Logistic Regression
• Building Keras Models
  • RNN
• Also in the end tried to analyze helpful reviews.(But i didn't have much time left)

Here as score 0 refers to 1, 1->2, ..., 4->5

• Since the data is imbalanced we can try resampling the data (downscaling the 5s or upscaling 1-2-3-4)
• Training multiple models then ensemble
• Hyperparameter tuning