Method for recommending next position of campus user based on space-time attention network

Abstract

The invention discloses a method for predicting the next position of a campus user based on a space-time attention network, and the method comprises the steps: employing an SMM algorithm with track data preprocessing according to the distribution characteristics of a time sequence, removing abnormal data through a segmentation merging and adaptive adjustment method, and solving a large number of pinpong effects of track data; mining a sequence mode by adopting a frequent item set in a reverse order, and iteratively removing a redundant item set through the mined sequence mode set to obtain a to-be-mined sequence mode set; dividing the whole historical track of the user into a plurality of time windows, learning the user number, the position number, the sign-in time and the space-time effect of each piece of historical data, and converting into vector representation to construct a multi-modal embedded layer; constructing a self-attention aggregation layer, aggregating important related positions in a user track, and capturing expression of sign-in of each time of long-term dependence updating; and constructing an attention matching layer. According to the method, the database is greatly compressed, the track mode mining performance is improved, and the personalized prediction recall rate is effectively improved.

Description

technical field [0001] The invention belongs to the technical field of next geographic point of interest recommendation, and in particular relates to a method for predicting the next location of campus users based on a spatio-temporal attention network. Background technique [0002] With the rapid development of information technology and the widespread application of location-based service systems, facilities such as smartphones, wearable devices, and automobiles have recorded a large amount of trajectory data based on time and location. One of the most important applications is next POI recommendation, which aims to predict the next POI based on the user's historical check-in activity sequence. Due to the wide application of location technology, the research on location prediction has a long history. previous job [1-2] Sequential recommendation methods based on Markov chains are proposed, but they only consider the influence from the last check-in activity. Inspired by ...

AI Technical Summary

Problems solved by technology

Previous jobs [1-2] proposed Markov chain-based sequential recommendation methods, but they only considered the influence from the last check-in activity

Inspired by the success of word2vec in sequence problems, some researchers proposed to learn the embedding vector of POI and make recommendations based on temporal POI embedding, failing to consider the influence of spatiotemporal background on POI recommendation at the same time.

[0003] For the trajectory data set of campus WiFi detection, due to the unstable AP access point signal and the crossing of different AP signals, the pingpong effect will cause the behavior trajectory data to be too complex and increase the additional overhead of the algorithm; the traditional trajectory data is stored in the form of time stamps, which lacks The description of users, especially students and commuters, has strong periodic characteristics, resulting in low algorithm efficiency

Images