Matrix decomposition method and device based on convolution attention and electronic device

Abstract

The invention relates to a matrix decomposition method and device based on convolution attention and an electronic device. The matrix decomposition method based on convolution attention comprises thefollowing steps: expressing a user description document of an article as a word vector matrix; inputting the word vector matrix into a convolutional attention neural network to obtain a hidden factorof the article; obtaining hidden factors of the user by using probability matrix decomposition according to the scoring information of the user to the articles and the hidden factors of the articles;calculating an inner product of a vector for the hidden factors of the article and the hidden factors of the user, and establishing a prediction scoring matrix according to the inner product; and optimizing parameters of the convolutional attention neural network and the probability matrix decomposition by using an error back propagation algorithm. The matrix decomposition method based on convolution attention is good in interpretability, can effectively reduce the problem of cold start of articles, can improve the problem of data sparsity, and improves the accuracy of score prediction.

Description

technical field [0001] The invention relates to the technical field of data processing, in particular to a matrix decomposition method, device and electronic equipment based on convolutional attention. Background technique [0002] The recommendation method based on matrix decomposition is a commonly used basic method at present. By decomposing the rating matrix of a given user on an item, the product of two low-rank matrices is used to approximate the original rating matrix. The goal of the approximation is to make the predicted rating matrix The square error between it and the original scoring matrix is ​​the smallest. The two low-rank matrices are the feature matrices of the user and the item respectively, and these feature vectors in the matrix can be simply understood as the user's preference for different attributes. Compared with content-based filtering methods, matrix factorization has better domain adaptability and can handle unstructured data, such as music and vi...

AI Technical Summary

Problems solved by technology

[0003] Existing rating predictions regard users' rating information on items as pure ratings, without in-depth consideration of the deep meaning contained in users' ratings for different items and different users' ratings for the same item from the perspective of users and items. Simple Matrix factorization cannot add text features, and cannot deeply understand the user's comment text on items, so it cannot solve the interpretability problems of cold start and recommendation

[0004] Convolutional neural networks can be used to extract hidden features of text, but convolutional neural networks often ignore the feature extraction of single words, and different words in the text have different importance for scoring predictions, and most of the work does not consider different word pairs Variability in rating predictions

Images