User position prediction framework based on clustered graph federal learning

Abstract

The invention provides a user position prediction framework based on clustered graph federal learning. The user position prediction framework comprises the following steps that S1, using a sequence prediction model to carry out training locally by users; S2, uploading the model parameters and the implicit state of the original sequence data passing through an encoder to a server by users; S3, learning a similar graph structure by using an implicit state; S4, obtaining an embedded representation of the users through a graph convolutional neural network; S5, dividing the users into a plurality of clusters through a clustering method, and executing a federated average algorithm by the users in each cluster; and S6, downloading the embedded representation and the averaged model parameters to the corresponding users, splicing the implicit state and the embedded representation by each user, then outputting a prediction result, and updating the server model parameters. The method has the advantages that federal learning protects data privacy; the graph convolutional network solves the problem of insufficient training cost caused by label scarcity; and the graph clustering algorithm enables more similar users to execute a federated average algorithm so as to solve the problem of heterogeneity between the users.

Description

technical field [0001] The invention belongs to the technical field of smart device user location prediction, and in particular relates to a user location prediction framework based on clustering-based graph federated learning. Background technique [0002] User location prediction aims to predict the mobile trajectory or position of the user in the real scene, which allows the intelligent system to assist the user to improve the quality of life, and has been involved in many fields such as smart services, smart cities, and healthcare. In recent years, with the popularity of smart wearable devices and the advancement of location-based smart service technology, the problem of user location prediction has become a research hotspot in both academia and industry. In previous research, user location prediction on a specific data set has achieved relatively good results. However, user location prediction has its unique characteristics and challenges: (1) Privacy protection. Loca...

AI Technical Summary

Problems solved by technology

However, user location prediction has its unique characteristics and challenges: (1) Privacy protection

Location data carries a large amount of user privacy information, which is often very sensitive. Existing methods usually use centralized batch processing to train data, which not only leaks user privacy, but also requires a very large storage space to save For all user historical trajectory data, how to improve the effect of model prediction while protecting user data privacy is one of the problems that need to be solved

(2) Scarcity of labels

Labeled activity data is always limited, and it is expensive for individual users to obtain enough movement trajectory data for model training

If users train individually, the limited amount of data is an extremely serious problem

(3) User heterogeneity

Since the user's characteristics or interests are diverse, different users have different location movement patterns, which means that the user's data is heterogeneous, that is, non-independent and identically distributed, so the generalization model obtained through federated learning Unable to achieve the best performance on a specific client

Images