grouple recsys

aka Sagiri

Sagiri is a hikikomori anime illustrator. She reads such a lot manga and doujinshi so she can recommend something suitable for anyone. Like as this app will

Application allow grouple.co item recommendations. Regularly retrains, calculates items similarity and so on.

Web scrapper for collecting user bookmarks data included

Models:

• implicit ALS for item recommendation
• tSNE + Agglomerative clustering / DBSCAN for initial item recommendations

sudo docker-compose -f docker-compose.yml up -d --build sudo docker-compose -f docker-compose-selflib.yml up -d --build

sudo docker-compose -f docker-compose.yml restart rumix

sudo docker-compose -f docker-compose.yml up -d -V --build --remove-orphans

Single container update

docker-compose stop selflib
docker-compose build selflib
docker-compose up --no-start selflib
docker-compose start selflib

Exploration update

There is a script ./database/explorations/explorations.py to get new explorational_recs.json

First, prepare items_features.csv (item_id,views), then run: python explorations.py SITE_ID

To change number of clusters, touch DBSCAN params (eps and min_samples). Increase it to decrease number of clusters. Finally put resulted file into config.explorations_path