Challenges
The memory-based CF model experiences a few challenges and are summarized as follows:
- Similarities are based on common items, and hence they become unreliable when there are many missing values and common values are less.
- To overcome this shortcoming, model-based CF methods have been proposed, which use rating data to learn the ML model for making predictions.
- The most common model-based CF methods include Bayesian beliefs nets (BNs) CF models, clustering CF models, and neural nets CF models.
Many approaches were used to deal with the sparsity problem using the dimensionality reduction techniques such as the Singular Value Decomposition (SVD) that eliminate less important users or items to reduce the dimensionalities of the user-item matrix.
The information retrieval method Latent Semantic Indexing (LSI) is based on SVD, or eigentaste based on Principle Component Analysis (PCA), where the similarity values between users are calculated based on the representation of the users in the reduced space. Nevertheless, eliminating users or items might lead to ignoring important data when making predictions or recommendations.