A Bayesian collaborative filtering recommendation method

Abstract

The invention discloses a Bayesian collaborative filtering recommendation method. The invention discloses a Bayesian collaborative filtering recommendation method. The method comprises the steps thatthe input of a model is a scoring matrix of a collaborative filtering recommendation system, and the scoring matrix is decomposed into two potential matrixes (please see the formula in the specification); wherein the matrix Uik for the M * K represents the probability that the user i belongs to the group k, and the Uik belongs to (0, 1); the matrix Uik for the M * K represents the probability thatthe user i belongs to the group k; Wherein for the N * K matrix Vjk, the evidence that the user group k likes the commodity j is represented, i.e., the prediction scoring matrix R '= UVT; Because thedata set R is relatively sparse, the observed entry can use the set = {(i, j) | Rij is observice}; Adopting a probability method for the problem; Representing a likelihood function for the observation data, and processing the potential matrix as a random variable; When each value of R is assumed to be from the product of U and V, The beneficial effects of the present invention are that: the userpreferences are diverse, and the taste is not similar to the taste reflected in a small data set. According to the present invention, with the present invention, the taste of the food can be improved,such that the food can be drunk by the user. A large number of data missing problems exist in the real data set, and if the value which is difficult to predict is predicted to be a median value or anaverage value, the recommendation significance is lost.

Description

technical field [0001] The invention relates to the Internet field, in particular to a Bayesian collaborative filtering recommendation method. Background technique [0002] With the emergence and popularization of the Internet, we can easily obtain a large amount of data, but a large amount of data makes it difficult for users to directly obtain effective information when searching for information, so that our use of information is reduced. Therefore, it is very important to use a recommendation system to effectively solve the problem of information overload. The recommendation system often recommends content according to the user's requirements, hobbies, etc. At present, recommendation systems have been widely used in many fields such as movies, music, shopping, social networking, and books. One of the most widely used and effective personalized recommendation techniques is the collaborative filtering recommendation algorithm. Collaborative filtering is mainly divided int...

AI Technical Summary

Problems solved by technology

[0008] Classical matrix factorization has two major disadvantages, one of which is that the decomposed matrix components are not constrained to be non-negative, which makes it difficult to understand the predicted meaning of each component, and the recommendation system has no interpretability

Another disadvantage is that the general matrix decomposition is non-probabilistic, and the solution method is to minimize the error between the original matrix and the approximate matrix, that is: This method will more easily lead to overfitting and ignore its uncertainty, and the recommendation effect is not good

[0009] There are some inherent shortcomings in the model-based recommendation system. For new items or when there are very few item ratings, it is difficult for a single model method to provide sufficient evidence, which will greatly affect the quality of recommendation.

Images