Multiple collaborative filtering recommendation algorithms for electronic commerce information

Abstract

ABSTRACT The information filtering recommendation algorithm has poor security, low efficiency and user satisfaction, and high network energy consumption. A multiple collaborative filtering recommendation algorithm for e-commerce information based on user preferences is proposed. In the time series data of the original e-commerce information, using the sliding window for feature extraction, the trend value characteristics of the information variable will be obtained, and the suspected outfielders will be determined based on the trend feature value. Using the resulting eigenvalues for secondary detection, the abnormal data points are further judged, and the abnormal data therein is filtered using a filter. The filtering results are substituted into the information filtering recommendation based on the user’s preferences. According to the user’s preference knowledge, the user’s preference space matrix is constructed to complete the user similarity calculation, and the nearest neighbors with the same or similar preferences are obtained. The demand of the target user is predicted using the nearest neighbor and preference knowledge, and the multiple collaborative filtering of the electronic commerce information is completed. Experiments show that the user satisfaction of this algorithm is in the range of 91%–99%, with high recommendation efficiency and security, low network energy consumption, and strong practicality.