Design and research of news recommendation system based on perceptron model in big data era

Abstract

Abstract With the continuous development of information technology, we are now in an era of information explosion, and we are also facing the problem of information overload, i.e., it is difficult to find the news that interests us from the huge news sources on the Internet. Based on this problem, this paper proposes a perceptron model-based news recommendation system, which uses the perceptron model to extract news keywords and realize the classification of the news; for user interest preferences, we analyze users’ news browsing preferences by collecting user behavior logs and use a multi-layer perceptron model to aggregate users’ interest features. The perceptron model is compared with traditional recommendation methods in terms of accuracy and recall, and one user is selected from 50 users who have been using the system for three months to analyze their news browsing preferences by single-day, short-term and long-term news data to verify the effectiveness of the perceptron model in aggregating users’ interest features. In terms of accuracy and recall, the accuracy of the perceptron model is higher than that of the traditional method for each news type, improving the accuracy by at least 5%-10%, e.g., 10% for real estate news and 9% for financial news. The recall rate of the perceptron model is improved by more than 6% for all news types except for social and entertainment news types. In terms of user interest preference mining, the whole change process of user browsing preference trend is in line with the human forgetting curve, which verifies that the perceptron model in the system is better used and meets the design requirements. Therefore, the applicability and reliability of the model proposed in this paper for news recommendation systems are verified in terms of accuracy, recall, and user interest preference mining.