Recommender evaluation

User 2570 | 7/13/2016, 2:49:29 PM

If I have a product recommender system based on image similarity, how can I evaluate and test the RSs system if want to compare with another RSs let say collaborative filtering recommender system (user/item).

Comments

User 954 | 7/18/2016, 6:49:18 PM

Hi, You can compare two models based on different evaluation metrics, like precision-recall, ROC curve, etc. In GLC you can use graphlab.recommender.util.compare_models for model comparisons:

https://turi.com/products/create/docs/generated/graphlab.recommender.util.compare_models.html