shyam31896 / yelp-recommendation-system

Yelp Recommendation System

This repository consists of the scripts required to train and predict the ratings that a user might provide for any new business on Yelp. The training script train.py provides the user with the option of training the Yelp data on three different model options (User based Collaborative Filtering, Alternating Least Squares method using the Scikit-surprise library, and XGBoost) and the performance of these models are evaluated using the prediction script predict.py The Usage examples are provided below for reference.

Apache Spark is used for the Big data processing. The code runs with the assumption that Apache Spark is pre-installed.

Usage:

Step 1: Download the required packages:

$ python3 -m pip install -r requirements.txt

Step 2: To download the training data and extract the models:

$ bash download_data.sh

Step 3: To train the models on previously constructed datasets:

$ spark-submit train.py -tr <Training file path.json> -ts <Test file path.json> -o <output trained model file path.model> -m <Training method [user_based_cf, surprise, xgboost]>

Step 4: To compute the predictions and calculate the RMSE of the predicted models:

$ spark-submit predict.py -tr <Training file path.json> -ts <Test file path.json> -md <trained model file path.model> -o <output prediction file path> -m <Training method [user_based_cf, surprise, xgboost]>