Skip to content

Latest commit

 

History

History
35 lines (28 loc) · 1.96 KB

README.md

File metadata and controls

35 lines (28 loc) · 1.96 KB

Utility Scoring of Product Reviews

  • Utility is the usefulness or reliability of a given review.

  • Utility score is the score given to a review based on its usefulness from 0 to 1.

  • Dataset

    • We downloaded reviews and related data in 22 different domains like Electronics, Clothing, Sports,Video games etc from amazon product data

  • Classification

    • Classify a given review into one of the 22 categories using

    • Neural Network Architecture

      • Embedding layer with the size of vocabulary.
      • 1-D convolutional layer with 32 filters of size 3 with relu activation.
      • Max pooling 1-D layer with pool size 2
      • Fully connected layer with relu activation function.
      • Fully connected layer with 22 nodes and softmax classifier.

    • Voted Classifier

      • Ensemble of 5 models (SVM, Decision Tree, Naive Bayes etc) to classify a given review using its TF-IDF vector.

    • Adjectives

      • Generated adjectives for each category using a seed set and recursively adding all its synonyms. The category with maximum count is the class of the given review.

    • Learning Algorithms

      • Support Vector Regression (SVR)
      • Simple linear regression (SLR)

    • Comparision

  • Conclusions

    • Classification
      • Accuracy in classification is highest in the case of Voted Classifier.
      • Voted classifier has the advantage of testing a review with 5 different classifiers.
      • Neural networks also works good but less accurate than voted classifier.
      • The naive method of counting number of matches in adjectives does not work well as there is no learning based in this method and also it is ambiguous