Multi-dimensional user feature vector screening method

Abstract

The invention relates to a multi-dimensional user feature vector screening method, which is characterized by comprising the following steps of 1) extracting a certain feature vector quantum set in a multi-dimensional user feature vector; 2) calculating the dependency degree of each feature vector in the feature vector quantum set on the user class label; 3) adding the feature vector with the maximum dependency degree in the feature vector quantum set into a candidate feature vector quantum set, and determining the information increment of the candidate feature vector quantum set; 4) judging whether the information increment of the candidate feature vector quantum set meets the requirement or not by adopting an integrated learning method and according to a preset judgment condition, and if the information increment does not meet the requirement, entering the step 2) to re-extract the feature vector quantum set in the multi-dimensional user feature vector; and if the information increment meets the requirement, taking the candidate feature vector quantum set as an optimal feature vector quantum set to complete the screening of the multi-dimensional user feature vectors. The method can be widely applied to the field of large-scale multi-dimensional user data pre-processing.

Description

technical field [0001] The invention relates to a method for screening multi-dimensional user feature vectors, and belongs to the field of pre-processing of large-scale multi-dimensional user data. Background technique [0002] In recent years, with the continuous development of science and technology, human daily life has become more abundant, the dimension of user data has become higher and higher, and the scale of users has become larger and larger. Therefore, in the face of large-scale multi-dimensional user data, it is increasingly urgent to propose an efficient feature vector screening method. At present, the common eigenvector screening methods include filtering, wrapping and embedding. The filtering method first screens the user's feature vectors, and then trains the learner. The result of the user's feature vector screening has nothing to do with the subsequent learner. The wrapping method uses the performance of the learner to be used as the evaluation criterion ...

AI Technical Summary

Problems solved by technology

[0004] However, the existing multi-dimensional user feature vector screening methods have the following problems: 1) The calculation is mainly concentrated on all multi-dimensional users. When the number of users is huge, the calculation cost is high, and the calculation cannot even be completed

1) It is necessary to calculate the feature screening results on the selected part or the whole subset, and the feature vector redundancy is prone to appear in the integration process of the sub-feature screening results

Images