User item recommendation method

Abstract

The invention discloses a user item recommendation method, which comprises the following steps of: 1) selecting an observed user / item pair (u, p) from a training data set and inputting the observed user / item pair (u, p) into a recommendation system model M to obtain a similarity score s (u, p) of a user u and an item p; 2) selecting an unobserved user / item pair (u, n) and inputting the unobserved user / item pair (u, n) into the model M to obtain a similarity score s (u, n) of the user u and the item n; 3) calculating a loss value by using a loss function; then optimizing the model M by adopting a back propagation algorithm according to the obtained loss value; bu is an auxiliary score corresponding to the user u; and 4) for a prediction item x, inputting the prediction item x into the trained model M, and if the similarity score s (u, x) is greater than bu, recommending the item x to the user u.

Description

technical field [0001] The invention belongs to the technical field of computer software, and particularly relates to a user item recommendation method. Background technique [0002] Facing the problem of information overload, recommender systems play an important role in efficiently providing users with useful information. As a widely used technique in recommender systems, collaborative filtering (CF)-based methods usually utilize user interaction behaviors to simulate users' potential preferences and recommend items to users according to their preferences. In general, given user-item interaction data, a typical CF method usually consists of two steps: (i) defining a scoring function to calculate the correlation score between users and candidate items, (ii) defining a loss function to optimize all observed The total relevance score for user-item interactions. From a loss definition point of view, CF methods are usually optimized by a loss function that assigns a higher sc...

AI Technical Summary

Problems solved by technology

While these approaches are successful, they ignore the fundamental mechanisms that lead to the vanishing problem

Intuitively speaking, hard samples are usually samples near the boundary, and it is difficult for the model to distinguish these samples well. Therefore, the accuracy of the corresponding recommendation system when recommending items for different users needs to be improved.

Images