User remote access position prediction method based on cross-city interest point matching

Abstract

The invention discloses a user remote access position prediction method based on cross-city interest point matching, which comprises the following steps of: constructing an interest point-category bipartite graph, and obtaining an embedded representation of an interest point in a low-dimensional continuous vector space based on a word hopping model; constructing a directed weighted interest point network, and sampling interest point sets from different cities based on random walk; realizing cross-city interest point matching based on a maximum mean value difference theory; modeling sign-in behaviors of different city users by utilizing Bayesian personalized sorting; migration of user sign-in preferences among different cities is realized through multi-task joint modeling; training the model; and calculating sign-in probabilities of the user for different interest points in the target city according to the model parameters, and selecting the interest points corresponding to top-K probability values to form an ordered list as a final prediction result. According to the method, the access position of the user in the non-settlement city can be effectively predicted.

Description

Technical field [0001] The present invention belongs to the field of user access location prediction technology, especially involving a location -oriented information service provider -based user -based user -based user access location prediction method. Background technique [0002] With the advancement of transportation technology, people's travels have greatly shortened at the time of the way, and the phenomenon of users "unlocking" new cities is becoming more and more common. Users who predict mobile information services in mobile information services in different places help understand the user behavior mode at a higher level, thereby improving the quality and user experience based on location services. However, location -based information services usually have localized characteristics. Existing solutions are more concerned about the prediction of users' access location in the local (settled city). The prediction schemes for users in different places (non -settings) are sca...

AI Technical Summary

Problems solved by technology

However, location-based information services usually have localization characteristics. Existing solutions focus more on the prediction of users' access locations in the local area (settled cities), and the prediction solutions for users' access locations in remote places (non-settled cities) are relatively scarce.

The intuitive strategy to solve the problem of user location prediction in different places is to transfer the user's check-in mode in the settled city to the target city. Because different cities have different cultural and regional characteristics, users often show different interest preferences in different cities, so the simple Knowledge transfer is difficult to solve the problem of user interest deviation

[0003] Existing schemes realize cross-city migration of user access preferences by modeling users’ check-in behavior in settled cities and non-settled cities. However, the precondition of these schemes is that users not only have sufficient check-in records in settled cities, but also The assumption that there are positional trajectories to model does not match reality

Since different cities have different cultural and regional characteristics, simple knowledge transfer is difficult to solve the problem of user interest deviation

Considering that there are differences between cities, users often show different interest preferences in different cities, so it may be counterproductive to directly migrate the check-in preferences of users in the settled city to the target city, resulting in "negative migration"

To sum up, there are two major challenges in user remote access location prediction

[0004] First, users often do not have check-in behavior in non-settled cities, resulting in extremely sparse location trajectories, and it is difficult to use existing methods to directly model users' check-in behavior in different places; Its own inherent mobile behavior mode is also restricted by the local cultural and geographical environment. Existing technical solutions cannot simultaneously take into account the migration of user interest preferences (city-shared characteristics) and the transformation of interest preferences (city-specific characteristics)

Images