It is important to recognize that some users, in reality, may interact with items (games) differently than other users in terms of gaming frequency and affinity by proxy. This discrepancy, in turn, could potentially translate to different implicit ratings based on the user-item interaction (hours of gameplay) data available for this particular analysis.
Based on our previous findings, we will now modify the model to account for user and item biases by including embeddings for average users and items (games) using the following code:
# caculate minimum and max ratingmin_rating <- steamdata %>% summarise(min_rating = min(rating_scaled)) %>% pull()max_rating <- steamdata %>% summarise(max_rating = max(rating_scaled)) ...
