German wings review challenge

The notebook shows two use cases. The data is found in a text file holding a set of reviews about the German wings airline (airline code ‘4U’).

• First use case:

The goal of this use case is to show that based on the customer review data how to predict the target variable 'Recommended'.

• Second use case:

The goal of the second use case is to find topics in the reviews.

The notebook covers the following topics:

• Data loading

  • loading the customer review data from txt to dataframe

• EDA

  • plotting distribution of the variables
  • plotting histograms
  • showing relationship between target variables to other variables
  • text analysis
    • plotting word frequencies
    • topic modeling with LDA

• Feature engineering

  • tfidf
  • count features
  • dimensionality reduction using truncated SVD
  • word embedding
    • glove

• Evaluation

  • logloss
  • roc curve
  • precision recall curve

• Machine learning models

  • logistic regression
  • navie bayes
  • gradient boost machine
    • xgboost
  • deep learning
    • lstm
    • gru
    • bidirectional lstm

• Error analysis