Personalized place semantic recognition method based on multi-context embedding

Abstract

The invention relates to a personalized place semantic identification method based on multi-context embedding. In the traditional personalized place semantic identification method, the situation information is insufficiently considered and the inherent limitation of the feature engineering problem is solved, and a method for collaboratively identifying the personalized place semantic and the mobile phone App relevance is proposed. For personalized place semantic recognition, the depth neural network is used to get the representation of multi-contextual information (time, user, user activity, etc.). For mobile phone App correlation recognition, we use word embedding to get the representation of different mobile phone Apps. By sharing the representation of mobile App, the parameters of the model are cooperated with each other. The invention combines context perception and presentation learning to carry out personalized place semantic identification, and has broad application prospects inthe fields of pervasive computing and location-based service (LBS).

Description

technical field [0001] The invention relates to the field of semantic recognition of personalized places, in particular to a method for semantic recognition of personalized places based on multi-context embedding. Background technique [0002] With the rapid development of the mobile Internet, the LBS carried in smart devices is gradually emerging. For a geographic location, it can be precisely located by the basic longitude and latitude. However, in order to provide a more intelligent LBS, the longitude and latitude information of the geographical location is obviously not enough, so it is necessary to give it more human-understandable labels, such as home, school, dining place, etc., to expand its semantic information. These tags are called venue semantics. The introduction of location semantics can make LBS more intelligent. For example, location-based search and discovery can be realized in location-based social network applications, and location-based semantic push of...

AI Technical Summary

Problems solved by technology

However, the existing work lacks sufficient consideration of contextual information, usually only considering a small number of contexts such as time and users, while ignoring important contexts such as user activities

In addition, existing work usually extracts features separately for different contextual information, ignoring the correlation between different contexts in the same place

Finally, the existing work uses the traditional feature engineering method to extract features. The features defined by feature engineering are usually obtained under the guidance of people's experience. These features are often incomplete and require long-term design and verification.

Images