Query on computing non-rated movies as ZERO in item-based collaborative filtering

[Swagato2415]

Respected Sir,
I was going through your code for building a recommender in the "item_based_collaborative_filtering.ipynb" file. My doubt lies in the sample dataset where you have considered "0" for movies not-rated by users since it will be affecting the cosine similarity computation on the denominator side. Or should we consider only those values for Cosine Similarity computation for which both the users have given ratings or reviews and neglect the Nan values or 0 values in the vector completely?

Kindly share your views on this matter. In such a case, we may have to build our own Cosine Similarity Metric since KNN is not NaN aware and raises an error.