Serendipitous Recommender Template

Alex Simes

This README is a work in progress.

This is the recommender used in my Master's thesis project. It provides both baseline (collaborative) recommendations and serendipitous recommendations. Serendipitous recommendations are pre-calculated upon training, using the algorithm proposed by Zhang et al. in their 2012 paper.

Below is a short animation illustrating how the serendipitous algorithm works.

Quick Setup

• Install PredictionIO, follow their guide
• Use pio command line to get this template. This command will copy the template to a directory 'Recommender'

pio template get alex9311/predictionio-serendipitous-recommender-template Recommender

• create a new PredictionIO application and build it.

cd Recommender
pio app new Recommender
pio build

• Add items to event store, use json to load in bulk. Json for items should have the following format

{"event":"$set","entityType":"item","entityId":"item1","properties":{"category":"a","title":"title for a1", "date_created":"2016-05-01"}}
{"event":"$set","entityType":"item","entityId":"item2","properties":{"category":"a","title":"title for a2", "date_created":"2016-04-01"}}
{"event":"$set","entityType":"item","entityId":"item3","properties":{"category":"b","title":"title for b3", "date_created":"2016-03-01"}}

• To load the file use the following command, your app id is shown after calling pio app new or pio app list after creating it.

pio import --appid <your_appid> --input ~/items.json

• Add views to event store, same basic idea as items. The view events json file should have the following format

{"event":"view","entityType":"user","entityId":"user1","targetEntityType":"item","targetEntityId":"item1"}
{"event":"view","entityType":"user","entityId":"user2","targetEntityType":"item","targetEntityId":"item2"}

pio import --appid <your_appid> --input ~/events.json

• train, deploy

pio train
pio deploy

• Use the engine

curl -H "Content-Type: application/json" -d '{ "user": "user1", "num": 4, "recommender":"baseline"}' http://localhost:8000/queries.json