This is an object-oriented-programming implementation of my solution to the problem.

The goal is to classify customer feedback messages into three categories: compliment, complaint and suggestion.

The three main scripts are: 1-eda: for exploratory data analysis, 2-preprocessing: for different transformations on the given data to prepare for eda and training; and 3-models: for training a model and testing and/or making predictions.

The desired script is run by: python run.py process path1 path2 path3.

"process" takes the following strings: "eda", "train", "test". The arguments "path"s refer to the input data frames that correspond to three sectors of businesses that the feedback messages are coming from.

One challenging aspect of this data was that the data set was kind of small, ~1800 samples. Here I have implemented the logistic regression model and the performance of the model is good, recall~88%.

To improve the predictions, I used a pre-trained deep learning model (BERT) which improved the recall by ~7%.

Results of the model are stored in the "results" folder.