How to use:
YelpRatePrediction
├── yelp_academic_dataset_review.json
├── yelp_academic_dataset_business.json
├── yelp_academic_dataset_tip.json
├── yelp_academic_dataset_checkin.json
├── yelp_academic_dataset_user.json
├── code
| ├── data_to_vector.py
| ├── get_user_model.py
| ├── extract.py
| ├── linear_regression.py
| ├── splitDataset.py
| ├── generate.dataset.with.text.feature.ipynb
├── model
| ├── ridge.regression.ipynb
| ├── xgboost.ipynb
| ├── lightgbm.ipynb
├── generate.text.features
| ├──extract.restaurants.ipynb
| ├──aggregate.restaurant.reviews.ipynb
| ├──vectorize.reviews.ipynb
| ├──topic_modeling.py
| ├──process.topic.modeling.prob.ipynb
2.1: run splitDataset.py, this file split the whole review dataset into 3 parts
You should get data_modeling.json, data_training.json and data_testing.json
2.2 run extract.py, this file encode each item in the business dataset into a simpler representation
You should get restaurants_encoded.json
2.3 run user_model.py, this file gets you the user model
You should get user_model.json (note: if your computer does not have enough memory to run this, download the same file on the shared google folder and put it into the code foler)
2.4 NOTICE!! use the file data_to_vector.py as follows:
2.4.1: run it directly, you should get testing_X.json and testing_Y.json
2.4.2: modify the line 4,7,8 (no need to care about line 5), change the testing substring in them into training, then you shuold get training_X.json and training_Y.json
.
the 4 result files are training and testing data encoded into vectors.
2.4.3: run generate.dataset.with.text.feature.ipynb to get the feature table with text features, the rest are the same as you get from data_to_vector.py. For information about how to generate text features, see 2.6.
2.5 Fitting the model.
2.5.1 run linear_regression.py, the final result should be printed.
2.5.2 run ridge.regression.ipynb to fit a ridge regression model, in which we do feature selection using grid searching with Cross-Validation.
2.5.3 run xgboost.ipynb to fit a xgboost model, in which we do parameter tuning with Cross-Validation.
2.6 Generaet text features.
2.6.1 run extract.restaurants.ipynb to extract all the restaurants from the business dataset, and run aggregate.restaurant.reviews.ipynb to aggregate reviews for every restaurant.
2.6.2 run vectorize.reviews.ipynb to get the first text feature: text vector representations.
2.6.3 run topic_modeling.py and process.topic.modeling.prob.ipynb to get the second text feature: probability distributions over topics for each restaurant. Note training a LDA model can take a long time.