S-type improved similarity-based collaborative filtering recommendation method

Abstract

The invention relates to an S-type improved similarity-based collaborative filtering recommendation method. The method comprises the following steps of: carrying out data cleaning and filling on original recommendation scoring data; for the preprocessed scoring data, calculating similarities between users by utilizing an improved similarity model; constructing a nearest neighbor model for each user through a Top-N strategy and screening projects not recorded by the users through the nearest neighbor models; and calculating weighted scores, for the projects, of the users by utilizing scoring data of neighbors, and selecting projects according to the weighted scores. The method is capable of improving the recommendation correctness, decreasing the neighbor scale and improving the algorithm operation efficiency.

Description

technical field [0001] The invention relates to the technical field of item recommendation, in particular to a collaborative filtering recommendation method based on S-type improved similarity. Background technique [0002] With the rapid development of the Internet and the popularization of mobile smartphones, human beings have entered the information age. According to the statistical report on Internet status released by China Internet Network Information Center in January 2017, as of December 2016, the number of web pages in China was about 236 billion, an increase of 11.2% over the previous year. It is becoming more and more difficult to select items that people are satisfied with from the huge amount of information. Such problems exist in digital books, news, music, film and television works, and e-commerce platforms. The time cost for users to choose is getting higher and higher. Traditional solutions include classified navigation and search engines. When the amount o...

AI Technical Summary

Problems solved by technology

Traditional solutions include classified navigation and search engines. When the amount of information is relatively small, the above two solutions can alleviate the problem of information overload to a certain extent, but at the same time there are obvious shortcomings: all users get the same results , lack of personalized features, the recommendation system came into being at this time

In order to make the recommended results more in line with the actual needs of users, scholars have been trying to improve the classic collaborative filtering algorithm. However, with the increasing number of users and products, users cannot generate records for all products, but only a small part of them. The user-item matrix is ​​very sparse, even more than 99%. The extreme sparseness of the user-item matrix restricts the accuracy of the recommendation system

Images