Recommendation Engine

Requirements

This API requires to be installed on a Linux distribution and to have Python3.6+ installed GCC, Python3-dev

Installation

1. Clone the repository with git clone https://github.com/Kanton99/RecommendationEngine.git
2. Create a Python virtual environment with python3 -m venv *environment name*
3. Activate the envirnment with source *repository path*/*environment name*/bin/activate
4. Install the required libraries with python3 -m pip install -r requirement.txt

Usage

To use the API start RecommendServer.py. Once it booted up to get a recommendation for a user, do a get call with this format

curl -X GET http://localhost:5000/recommend -H "Content-Type:application/json" -d '{"userId":*user id*}'

and the system will return the top ten recommendation for that user. By adding "itemId:*item id* the system will also diplay recommendations related to that item and add that item to the list tha items that the user has interacted with.

Movielens

By setting in config.json 'use_movielens' to true the system will use the movielens dataset.

Data files

The data files supplied to the system must be .json files, and there must be two kinds:

1. Interaction files, where are written user-item interactions, with at least user id and item id per interaction
2. item files, where is stored the item data data, with at least the item id

Example

interaction file name: interactions_#####.json interaction file format: {"user_id":"#######","item_id":"##########"}

video file name: video_#######.json video file format:{"item_id":"###########","tags":["t1","t2","t3"]}

user file name: user_#######.json user file format: {"user_id":"#########","tags":["t1","t2","t3"]}

Config File

The config file can be changed to fit whatever json files you use.

• "use_movielens" to tell the system to use the movielens dataset instead of your data files.
• "data_directory_location" is where are your file stored.
• "item_data_file_pattern" is the pattern to search in the file names to find item data.
• "user_data_file_pattern" is the pattern to search in the file names to find user data.
• "user_interaction_file_pattern" is the pattern to search in the file names to find interaction data.
• "item_ID_key" and "user_ID_key" are the key to find the user IDs and item IDs.
• "item_tags" key in the item jsons to find the item features.
• "user_tags" key in the user jsons to find the user features.