Rating prediction via review text scan for yelp database

This repository contains an implementation for a review (text) to stars mapping prediction in keras.

1) Installation

1. Install docker from https://docs.docker.com/.
2. Open command line window and navigate to the project folder with all files from this repository.
3. Install a container for cpu build via: docker build -t keras_cpu --build-arg python_version=3.6 -f Dockerfile_cpu . or for gpu build via docker build -t keras_gpu --build-arg python_version=3.6 --build-arg cuda_version=9.0 --build-arg cudnn_version=7 -f Dockerfile_gpu ..

2) Start Training and Evaluation

Start a docker container. For cpu version type: docker run -it -v <path_to_project_folder>:/home -v <path_to_yelp_database>:/data --env KERAS_BACKEND=tensorflow keras_cpu bash and for gpu version: nvidia-docker run -it -v <path_to_project_folder>:/home -v <path_to_yelp_database>:/data --env KERAS_BACKEND=tensorflow keras_gpu bash. Attention: For GPU Version, please install NVIDIA drivers (ideally latest) and nvidia-docker from https://github.com/NVIDIA/nvidia-docker.

To start the training of network, run train.py. The script takes several arguments to control the training process and the data extraction. To get an overview, type train.py -h for a help message.

After training there are three files generated as output:

1. The network model as *.hdf5-file or *.h5-file
2. A training log
3. A raw data file from pickle, which contains the word mappings from text tokenizer

3) Predict rating on new custom texts

Start a docker container (like you did to start a training). Start the prediction with the execution of predict.py. The scripts takes three parameters, type predict.py -h to get more information.

To enter new texts you can define strings direcly in the script with assigning them to variable input_texts.

Backup: Anaconda

If the docker container wont work you could take a look at the conda_environments folder. The folder contains *.yml files to generate conda environments. To generate an environment, install a valid anaconda version (see https://www.anaconda.com/) and type conda env create -f environment.yml to set the environment up (more information @ https://conda.io/projects/conda/en/latest/user-guide/tasks/manage-environments.html).