“…They differ in how to process the user‐item matrix. The model‐based algorithms (Khadem & Forghani, 2020; Lima et al, 2020) follow two steps: - The algorithm handles several matrices to generate an efficient model for representing the original rating matrix.
- The generated model is then used to estimate the predictions of active user ratings.
The most popular models for making recommendations include Bayesian classifiers (Gao et al, 2020), neural networks (Bandyopadhyay & Thakur, 2020), fuzzy systems (Barzanti et al, 2020), genetic algorithms (GAs) (Moses & Babu, 2020), latent features (Da'u et al, 2020), and matrix factorization (Liu & Ye, 2020). On the other hand, based on the steps outlined below, memory‐based algorithms (Mallik & Sahoo, 2020; Narayanan et al, 2020) use the entire rating matrix to achieve predictions: - The prediction ratings of the active user are estimated from the ratings of their neighbours.
- The similarity metrics are used to measure the distance between two users or two items by their ratings.
Memory‐based methods are divided into two main algorithms: - User‐based algorithms, where the method for obtaining neighbours works based on the users.
- Item‐based algorithms, where neighbours are obtained based on the items.
There are two main issues in RSs: cold start and sparsity.…”