Cold start recommendation algorithm based on implicit factor prediction

Abstract

The invention discloses a cold start recommendation method based on implicit factor prediction, and belongs to the field of machine learning and recommendation systems. According to score prediction, scores of the users on the items are calculated by predicting user preference factors, and the problem that new users cannot be recommended due to lack of historical scores of the users in a collaborative filtering algorithm is solved. The invention provides a cold start recommendation method based on implicit factor prediction by combining a matrix decomposition algorithm and a collaborative filtering algorithm aiming at the cold start problem of a user. According to the method provided by the invention, implicit factor feature vectors of a user and a project are obtained from a scoring data set through a matrix decomposition algorithm; if the user is a common user, a prediction score is calculated in combination with the implicit factor vector and the bias value of the user item. And if the user is a new user, obtaining an implicit factor feature vector (preference factor) of the user through a collaborative filtering algorithm, and finally calculating a prediction score. Compared with a traditional recommendation algorithm, the algorithm provided by the invention is higher than the traditional recommendation algorithm in both accuracy and expandability, and has certain practical significance.

Description

technical field [0001] The invention relates to the fields of machine learning and recommendation systems, in particular to a cold start recommendation algorithm based on latent factor prediction. Background technique [0002] With the intelligentization of Internet terminals and the popularization of the Internet, global data is growing explosively every day. Data is rich in information and has great value. However, when the speed of data generation far exceeds our ability to accept data alone When it comes to information, we not only need to spend time receiving information, but also need extra time to select the information we need from the huge data. In order to solve this problem, the recommendation system came into being. [0003] Since the collaborative filtering recommendation system was proposed in the 1990s, the personalized recommendation service provided by learning the user's historical behavior, hobbies and needs has played a good role in solving the problem of...

AI Technical Summary

Problems solved by technology

With the increase of the implicit feature dimension (the number of hidden factors) in matrix decomposition, the recommendation results of the model will also become personalized. When the number of hidden factors is very large, overfitting will also occur, which will lead to Recommended accuracy drop

Images