Mixed collaborative recommendation algorithm based on WNBI and RSVD

Abstract

The invention relates to a mixed collaborative recommendation algorithm based on WNBI (weighted network-based inference with z-score normalization) and RSVD (regularized singular value decomposition). According to the method, a WNBI algorithm is firstly used for abstracting users and items into nodes in a network; information hidden in the network is used; deeper potential information between the items is mined; a neighbor set similar to the items is found; secondly, an RSVD model is used for decomposing a user-item grading matrix into a user feature matrix and an item feature matrix; the data density is improved through dimension reduction; finally, the item neighbor information of the items is used for regularizing the RSVD model so as to overcome the defects of a conventional method. The mixed collaborative recommendation based on WNBI and RSVD (RSVD_WNBI) can use the information, obtained by the WNBI algoritm, hidden in the user-item network for regularizing the RSVD model, so as to improve the recommendation accuracy and effectively solve the grading matrix sparsity problem.

Description

Technical field [0001] The invention relates to a personalized recommendation system technology, in particular to a hybrid collaborative recommendation algorithm based on WNBI and RSVD. Background technique [0002] With the rapid development of the Internet, the amount of data on the Internet has increased sharply. People are also facing the dilemma of "information trek" while obtaining the latest developments and massive amounts of information on the Internet, that is, how to filter out interesting information from a large amount of useless information. Useful information. Therefore, the recommendation system came into being. It collects various user information and data, analyzes the hidden user interests and behavior patterns, and provides users with customized personalized recommendation services based on the analysis results. [0003] Collaborative filtering (CF) is the earliest and most widely used algorithm in recommendation systems. The core of the algorithm is to analyze...

AI Technical Summary

Problems solved by technology

The former can be divided into two types based on user neighbors and project neighbors, mainly by finding a set of users (or projects) similar to the target user (or project) in the user group (or project group), and synthesizing these similar users (or projects) ) information to form the system’s prediction of the target user’s (or item’s) preferences. However, with the large-scale increase in the number of users and resources, this type of algorithm has problems such as data sparseness and cold start, and only uses the nearest neighbor similarity measure. It will inevitably lead to a significant decline in the quality of recommendations; the latter effectively preserves information content by reducing the dimensionality, and at the same time greatly reduces the complexity of calculations and memory requirements, but relies too much on the user-item rating matrix. When the data is sparse, Recommendation quality will also be greatly affected

Images