I am going to classify the hotel reviews as positive, negative and neutral. For train the model I couldn't find any data set with labels positive, negative and neutral. I could find this Hotel Reviews data set but it only contains positive and negative labels. So I decided to create new training data set with labels positive, negative and neutral. To decide the label I thought to use the rating number. Is this a good idea or bad idea? Can some expert give some suggestions on problem because I am new to this field.
Sure you could do that, but why not try to predict the rating directly?
Unless you have a specific need for the 3 categories positive, neutral, negative, there's no reason not to use the rating provided since it also represents the sentiment of the review.
Notice that this would be a regression task since you predict a number (as opposed to a classification task where one predicts a categorical variable).