The invention discloses a movie recommendation method fusing end-to-end training of visual features. The method comprises the steps of receiving an input user
score matrix, initializing a user featurematrix and a movie
feature matrix, and then building an initial model in combination with a visual
feature matrix; and performing the end-to-end training by utilizing the initial model to obtain a user
score prediction matrix, and recommending movies for users according to the user
score prediction matrix. According to the method, the learning of the visual features such as a poster, a key frameand the like, and a
recommendation model are fused to a unified framework, and the end-to-end training is carried out; the learned visual features not only have relatively high expression and classification capabilities; the obtained user score prediction matrix can fully reflect the preferences of the users to the visual features; and the movies recommended for the users also can better conform to the user preferences. The invention further provides a movie recommendation
system fusing the end-to-end training of the visual features, a
server and a computer readable storage medium, which havethe abovementioned beneficial effects.