CF algorithm is divided into two categories, based on the memory (Memory-based), also known based on the user's (User-based), the other for Model-based (Model-based), also called items based ( Item-based). The basic idea of User-based is that if user A favorite items a, user B like items a, b, c, user C like a and c, then consider user A and user B, and C are similar because they both like a, and the like a user also likes c, so the c recommended to the user A. The algorithm used to find the nearest neighbors of a user (nearest-neighbor) algorithm is a set of neighbors, the user set and the user has a similar preference, according to the algorithm to predict the user's preference neighbor. User-based algorithms exist two major problems: 1. Data sparsity. A large e-commerce recommendation system in general has a lot of items, which the user may buy less than 1% of the items, buy items overlap between different users are less, resulting algorithm can not find a user's neighbor, that is similar preferences User. 2. The algorithm scalability. Nearest neighbor algorithm to calculate the amount of users and with the increase in the number of items increases, the situation is not suitable for large volumes of data use. The basic idea is Iterm-based pre-computed similarity between items based on historical data of every user preference, and the user's favorite items with similar item recommended to the user. Or in the previous example, for example, we can know the items a and c are very similar, because I like a user also like c, but like a user A, so the c recommended to the user A. Because direct resemblance relatively fixed items, the calculated degree of similarity between the different items can be good in advance online, the result in the table, the look-up table when the recommendation, the user may calculate a score value, at the same time can solve the above two question.
Reproduced in: https: //www.cnblogs.com/lsai/p/3713595.html