Design of personalized recommendation method for entertainment news based on collaborative filtering algorithm

Abstract

In order to improve the personalized recommendation ability of entertainment news, this paper puts forward a personalized recommendation method of entertainment news based on collaborative filtering algorithm and deep semantic mining. Using word vector, neural topic model and other technologies to mine semantic information in news text to obtain news feature representation vector, and integrating all semantic features to represent user's preference vector, and then generating candidate sets that users may be interested in by matching, in the ranking stage of news recommendation, collaborative filtering algorithm is adopted to filter out interference options, and deep semantic mining technology is combined to realize dynamic mining and detection of entertainment news that users are interested in. Aiming at the news candidate set generated in the recall stage, the self-attention mechanism and other technologies are used to model the reading behavior sequences of users in different periods, and the learning of users' long-term and short-term preferences is completed by combining the attention mechanism, so that the click-through rate of candidate news can be predicted, and accurate recommendation can be made to users. The simulation results show that the personalized recommendation of entertainment news by this method has better pertinence and higher recommendation satisfaction, and improves the ability of emotional classification and feature enhancement of entertainment news.