floraxinru / nlp_hotelreviews Goto Github PK

(when revisiting this project in May 2019) Realized there is no correlation because the column Reviewer_Score contains scores for all the hotels from all reviewers, plus the ones that only left a score without reviewing.

To take care of that discrepancy, would probably need to filter out the scores posted by those who did not leave a review, and then find a way to match score with reviewer (while the purpose of the project is mainly to use the Naive Bayes Classifier for sentiment analysis).

At this point it might be better to go in the other direction of using the newly-developed ULMFiT and training a language model for hotel reviews. Also considering "diminishing returns", it might be better to explore ULMFiT with a larger, more nuanced data set.

A new approach for NLP was developed in 2018, using neural networks and inductive transfer learning for text classification. It would be very interesting to apply it here (there's already 1 related kernel on Kaggle), and train a language model and use it to classify hotel reviews (the result might also be applied to reviews for airbnb or other rentals)