RS Experiment

This repository is home to the source code implemented for Interactive Data Visualization for a better understanding of Recommender Systems, Master Thesis by Dennis Leszkowicz, IT-University of Copenhagen, Denmark (January 2019).

The source code is two fold, first part is a Flask web application that implements the recommender system described in the paper. The second part is a collection of notebooks and scripts that shows how the system came along, making it possible to recreate the needed data structures from scratch.

Credits

Thanks to Ben Frederickson for the Implicit library

Thanks to Òscar Celma for creating the Last.fm 360K Users Dataset. The dataset can be downloaded here.

A cleaned Last.fm dataset in hdf5 used for this project has been borrowed from Ben Frederickson, more info here.

---

Getting Started

1. Install requirements.

This project has been developed in Python 3.7 in anaconda virtual environment.

conda create --name <envname> python=3.7

conda activate <envname>

pip install -r requirements.txt

Dendencies included in requirements.txt

flask h5py implicit Jinja2 markupsafe matplotlib numpy

pylast redis scipy seaborn tables tqdm wordcloud

Additionally make sure to install Jupyter notebook for the notebook .ipynb collection.

---

1. Download data

To download the needed data there are two options:

Option A (quick): Running the shell script sh download_data.sh will download and place all the aggregated data in flaskapp/data/

---

Option B (slow): Run scripts located in scripts/ folder to recreate the data structures used for the webapp. This process takes hours, and requires you to:

1. Download the lastfm dataset lastfm_360k.hdf5 here and place it in flaskapp/data/.

2. Download lastfm_tags_cleaned.csv dataset here and place it here flaskapp/data/.

In order to recreate the data structures used for this project,.py script in following order.

1. 01make_user_profile_dict.py

2. 02make_tag_aggregations.py

Development mode

# set to True if development mode, otherwise production mode.
dev_mode = False

In order to run debug mode in the Flask enviroment this variable is set dev_mode = True in app.py. This ensures that the debugger and hot code-reloading is on.

---

2. Run locally

1. start the redis server by runnning redis-server.

2. run python app.py from the flaskapp/ folder.

or alternatively set export FLASK_ENV=development / export FLASK_ENV=deployment and run flask run

NB! app.py should run just fine.

This will run the web application on http://127.0.0.1:6543/

3. Deploy with Docker Compose

docker-compose up -d --build

This will run the web application on http://127.0.0.1:6543/

to stop the service run:

docker-compose stop