search

RUCAIBox / RecBole

A unified, comprehensive and efficient recommendation library
https://recbole.io/
MIT License
3.38k stars 606 forks source link

Model Metric Question in Sequence Recommendation #1889

Open Afar8612 opened 11 months ago

Afar8612 commented 11 months ago

Hello, when I used the Sequential Recommendation Model to evaluate, I found that when I used the @Hit index and the @Recall index in the library to evaluate, what is the reason why the results are exactly the same when the calculation methods of the two are completely different?

Fotiligner commented 11 months ago

@Afar8612 Thanks for your attention to RecBole! For your issue, the hit rate means the ratio of items in the predicted rank list that falls in ground-true set, while the recall index refers to the ratio of items in the ground-true set that falls in the predicted rank list. Since the number of items in gournd-true set and predicted rank list are the same within the calculation of recall@k and hits@k, the final results would be the same.