Evaluation platform used in SCARS publications.
SCARS is a P2P recommender system, as describe in "Local and social recommendation in decentralized architectures" thesis.
The API is in the project Core. Extends fr.cnrs.liris.scars.api.Scorer to define your own algorithm.
The recommendation algorithms are located in the project Scorer.
This project defines the algorithm to evaluate and the datasets to use for evaluation.
Those projects contain the logic to transform a flat dataset file (or database) into a logical dataset in memory.
The following datasets contain data from www.epinions.com
- http://www.trustlet.org/epinions.html
- https://alchemy.cs.washington.edu/data/epinions/
- http://liris.cnrs.fr/red/
SCARS is compatible with other datasets, such as Flixster, Synpinion, etc.
Datasets much follow the structure below:
user_id item_id rating_float
user_id friend_user_id trust_float
Download the dataset locally and execute Evaluation/runScars with one of the following option:
Usage:
runScars --epinions dir
runScars --flixster dir
runScars --synpinions dir
runScars --database
OPTIONS:
--load dir
--reload
--save dir
--view
--process {stats|actors|friends|extended|items}
--metrics [atotal,acov,amae,rae,rrse,wae,rwse,srcc,asrcc,pcc,apcc,actors,scores,confidence,sconfidence] (',' separated multivalues)