Yelp-Dataset-Classification-using-SVM

Classification of Yelp Reviews into Stars they can get

Approach followed by me:

1. Read the dataset and separate the reviews into positive and negative on the basis of stars
2. If Stars>3 then 1.0 else 0.0
3. Create Feature vector
4. Build the classifier
5. Do 5 fold cross validation
6. Try some trial and error using pos tagging and rejecting stop words for step 4 and 5

following libraries used: NLTK, SCIKIT learn

I also implemented the 5 class classification by using SVM Classifier, the approach was similar Below is the Output

Output: Dataset Loaded... Reviews Tokenized... Classification without any processing ###################################################################### TF Matrix Created... length of vector : 26051 Classifier Trained... Cross Fold Validation Done... accuracy score: 0.7674 precision score: 0.76834786108 recall score: 0.7674 classification_report: precision recall f1-score support

    0.0       0.67      0.68      0.68      1777
    1.0       0.82      0.81      0.82      3223

avg / total 0.77 0.77 0.77 5000

confusion_matrix: [[1211 566] [ 597 2626]] ###################################################################### Classification after removing stop words ###################################################################### TF Matrix Created... length of vector : 25924 Classifier Trained... Cross Fold Validation Done... accuracy score: 0.7632 precision score: 0.764003111764 recall score: 0.7632 classification_report: precision recall f1-score support

    0.0       0.66      0.67      0.67      1777
    1.0       0.82      0.81      0.82      3223

avg / total 0.76 0.76 0.76 5000

confusion_matrix: [[1198 579] [ 605 2618]] ###################################################################### Classification into 5 Classes ###################################################################### Classifier Trained... Cross Fold Validation Done... accuracy score: 0.4504 precision score: 0.445510219647 recall score: 0.4504 classification_report: precision recall f1-score support

      1       0.51      0.49      0.50       541
      2       0.29      0.24      0.26       497
      3       0.29      0.27      0.28       739
      4       0.38      0.40      0.39      1388
      5       0.58      0.61      0.59      1835

avg / total 0.45 0.45 0.45 5000

confusion_matrix: [[ 265 87 48 65 76] [ 100 121 102 99 75] [ 57 101 202 248 131] [ 45 71 201 549 522] [ 48 44 137 491 1115]]