Failure-Detection

NLP algorithm for identification of customer reviews reporting product failures.

Data Set - Overview

• The dataset is made of reviews scraped from Amazon website.
• Those reviews deal with tablets.
• The attributes given in the datasets are the review id, the tablet model id, the customer review, the rating given to the product, the failure degree label.
• The labels are given in the "Failure class" column of the csv files.
  • "IF" means "Intolerable Failure", i.e. the product broke and is unusable.
  • "TF" means "Tolerable Failure", i.e. the product broke but can be used.
  • An empty field means that there is no known failure.

Evaluation Metrics

To evaluate the model performance, we rely on F1-score and Balanced Accuracy.