I have implemented a recommender algorithm that recommends a set of books based on the ontology. The next step is to perform evaluation. What I've done, is that considering the test set as relevant set of books simply calculated recall and precision based on top-recommendations.
But, now there are negative tests in the test set, for improvement of accuracy. So that in a test set, containing 10 elements, 2 can be negative, meaning we are sure that user will not like it at all. I wanted again to use recal/precision, but it seems that again negatives will be neglected.
How can I perform evaluation, having this in mind? I am new to amchine learning techniques that's why any feedback is appreciated.