Method for recommending consumption package to user based on recommendation algorithm of deep learning

Abstract

The invention relates to a method for recommending a consumption package to a user based on a recommendation algorithm of deep learning, and the method specifically comprises the following steps: S1, model training: collecting desensitized user portrait data as sample data, the user portrait data comprising basic information of the user, consumption information of the user and package types subscribed by the user, building a deep learning network of a recommendation algorithm network model NFM and loading training parameters to form a deep learning prediction model; inputting data of the user into the established model, and selecting a package with the highest occurrence frequency as a recommendation result; S2, new sample prediction: calculating input user data through a deep learning prediction model, and selecting a package with the highest occurrence frequency as a final prediction result; and comparing the prediction result with the package actually used by the user, and calculating the accuracy of the prediction result. A deep learning algorithm is used to mine deeper data relevance, so that the recommendation accuracy of the model is greatly improved.

Description

technical field [0001] The invention relates to the technical field of communication data processing, in particular to a method for recommending consumption packages to users based on a recommendation algorithm based on deep learning. Background technique [0002] With the increasing competition among operators, in order to retain users and reduce the churn rate of users, it has become a problem for all operators to recommend new packages that meet user consumption habits. With the popularization of 5G, many new packages have also emerged. Some users have replaced the new packages, and most users are still using the old packages. Some of the packages used can no longer meet the user's consumption situation. Select the users who have not upgraded the new package from the total number of users, and recommend the new package that best suits their consumption habits to the user based on the user's personal situation and the user's consumption information. [0003] The traffic p...

AI Technical Summary

Problems solved by technology

This method has limitations: (1) Each user needs to calculate the similarity with all the data, resulting in too much calculation; (2) When the data scale is large and the feature columns are sparse, the accuracy of the recommendation algorithm will decrease

[0006] 2. Matrix decomposition algorithm, matrix decomposition combines the characteristics of latent semantics and machine learning, and can dig deeper connections between users and items, so the prediction accuracy is relatively high, and the prediction accuracy is higher than that of neighborhood-based collaborative filtering and The disadvantage of the content-based recommendation algorithm is also that the amount of calculation is large, and the model training is time-consuming.

[0009] The recommendation effect of traditional recommendation algorithms has limitations, and the accuracy of single feature analysis is not high, while the recommendation algorithm based on deep learning can dig out deeper data relevance

Images