Representation learning-based user life mode identification method

Abstract

The invention discloses a representation learning-based user life mode identification method. The method comprises the steps of 1, feature selection; firstly, a sign-in interest point category transfer sequence is extracted from original trajectory data; secondly, the sign-in interest point category transfer sequence enters a preprocessing layer, and life mode primary representation of each user is output; 2, a representation learning step; a word2vec CBOW representation learning method is used for learning vector representation of a user, and meanwhile semantic and time information of user movement is reserved. 3, a module identification step; and clustering the life mode through a classical clustering method. By analyzing the user movement data, the semantic-based user space-time behavior mode is identified from the space-time data with rich semantics, and the root cause for governing the user movement space-time behavior, namely the life mode of the user, is understood. The understanding of the life mode of the user is helpful for understanding the social and economic conditions of people and understanding the distribution of social capital in cities. The model can be used for deducing occupational and residential distribution, occupational conditions and the like.

Description

technical field [0001] The invention belongs to the field of mobile data analysis. By analyzing user location data, the user's spatio-temporal behavior pattern based on semantics is identified from the spatio-temporal data with rich semantics, and the root cause that dominates the user's mobile spatio-temporal behavior is understood, that is, the user's life pattern. Specifically, it relates to a method for recognizing user life patterns based on representation learning. Background technique [0002] The spatio-temporal behavior unit in the user semantic space is time slice + behavior, for example, working in the office at 9:00 in the morning and eating at a fast food restaurant at 12:00 noon are all units. If a group of people have the same spatiotemporal semantic behavior at different times of the day, then this group of people has the same life pattern, and they have similarities in the purpose / semantic level of movement. This spatiotemporal pattern is not affected by the...

AI Technical Summary

Problems solved by technology

[0004] First, there is no ready-made method to obtain user behavior characteristics from variable-length check-in records,

[0005] Second, it is very difficult to select the category accuracy of check-in points of interest. Low-precision categories cannot represent user habits well, while high-precision categories have a large number, and there are different degrees of similarity between categories, such as "Chinese food" and " The correlation between "fast food" is much greater than that between "Chinese food" and "factory".

[0006] Third, it is very difficult to define the similarity of life pattern behavior among users, because there are two dimensions of semantics and time that need to be considered at the same time

The existing distance function cannot effectively measure the effects of two factors at the same time

Images