Reviewlytics

Predicts accuracy of Amazon reviews by sentiment analysis.

Strategy

• Select sample data
  • Picked software related categories for the sake of context
  • Pareto's Principle: the law of the vital few requires proper selection of vital group.
    • Selected as a sample generalization technique.
    • Dataset offers review votes; higher vote count with high helpfulness rate can be considered more "truthful".
  • Verified Purchases
    • An Amazon verified purchase tends to be more credible.
• Form vocabulary
  • Preprocces
    • remove punctuations
    • tokenize
    • remove stop words
    • stem
  • Select most frequent N-words
  • Fill Bag of Words (BOW)
• Model data
  • Logistic Regression
  • Multinomial/Bernoulli Naive Bayesian
  • Support Vector Classifier
• Run statistical analysis' on data
  • Use model to find and filter outliers (in respect to stdev of actual score to predicted score)
  • Use model to predict overall score
• Compare findings

Datasets

• Amazon Review Data