Information recommendation method and system based on flexible multi-dimensional clustering

Abstract

The invention discloses an information recommendation method and system based on flexible multi-dimensional clustering. The method comprises the steps of S1, collecting data and constructing a three-dimensional data tensor according to the data; s2, decomposing the three-dimensional data tensor to obtain a tensor block structure and multi-membership clustering information; s3, performing tensor filling according to the tensor block structure and the multi-membership clustering information, and judging convergence; s4, cyclically repeating the steps S2-S3 until it is judged that the filled tensor reaches a convergence threshold value, and outputting the filled tensor; and S5, combining the output filled tensor with the recommendation target to generate information recommendation. The system comprises a preprocessing module, a decomposition module, a filling module, an interaction module and a recommendation module. According to the invention, more accurate and fine-grained recommendation service can be provided for the user. The information recommendation method and system based on flexible multi-dimensional clustering can be widely applied to the field of data mining.

Description

technical field [0001] The invention relates to the field of data mining, in particular to an information recommendation method and system based on flexible multidimensional clustering. Background technique [0002] With the rapid development of information technology, terabytes of data are circulated in e-commerce or multimedia fields every day. However, the data in practical applications often has missing and damaged problems, and it is impossible to directly obtain useful information to recommend to users. [0003] The performance of recommendation methods and systems is closely related to the quality of data filling. The existing data mining embedded filling technology still has the following deficiencies: 1) The clustering method based on matrix decomposition cannot effectively analyze the real structural information of high-order multi-dimensional data; 2) Data mining methods based on "hard clustering", such as k-means, not only have a single target attribute, but are ...

AI Technical Summary

Problems solved by technology

[0002] With the rapid development of information technology, terabytes of data are circulated in e-commerce or multimedia fields every day, but the data in practical applications often has problems of missing and damaged data, and it is impossible to directly obtain useful information to recommend to users

[0003] The performance of the recommendation method and system is closely related to the quality of data filling. The existing data mining embedded filling technology still has the following deficiencies: 1) The clustering method based on matrix decomposition cannot effectively analyze the real structural information of high-order multi-dimensional data; 2) Data mining methods based on "hard clustering", such as k-means, not only have a single target attribution, but are not suitable for multi-membership clustering tasks, and the clustering results are sensitive to missing values ​​and singular values. Weaken the quality of data filling; 3) The data filling technology based on exponential weighting fails to consider the clustering characteristics of "weighted sum is 1", which limits the interpretability and accuracy of multi-membership clustering results; 4) Limited by the ability of data calculation and storage, the existing technology cannot deal with the filling problem of large-scale sparse data well

Images