Predict Ratings Using Hotel Images by CNN

Predict Ratings Using Hotel Images by CNN

I use Tensorlow to explore sentiment analysis on images taken by hotel guests. Does the image tell a clue about the rating of given by the guest? The answer is yes, with only 360 training images, th accuracy score is 100% on the training set with 5 classes(raiting 1 to 5). Although the accuracy score is low on the test set, with a more complex model and the aid of regularization, the performance on the test set will be greatly improved. I am working on getting more images and using other pre-trained weights from classic CNN(AlextNet, LeNet and VGG16) so to build a more powerful model to predict ratings by using hotel images alone. 

Tips: The rating for the image is 2 out of 5(5 as the highest).