Activity-based social network activity feature extraction method

Abstract

The invention discloses an activity-based social network activity feature extraction method, which comprises the following steps: 1, constructing a social relationship graph of a user, vectorizing thesocial relationship graph by adopting a graph embedding algorithm, and extracting social network features of the user; 2, clustering the longitude and latitude of the activity, dividing the activityinto different areas, and extracting geographic position features; 3, dividing the activity time into time periods according to user habits, and extracting activity time characteristics; 4, generatingsemantic vectors of activities and user preferences for the semantic factors of the users and the activities by adopting a potential semantic analysis algorithm, and extracting social semantic features; and 5, processing and splicing the four feature vectors to obtain feature representation vectors of the user and the social activity. According to the method, the user and social characteristics are extracted to serve as the input of a subsequent training neural network on the basis of carrying out operation processing on the attributes as few as possible, so that the dependence of an activitysocial network recommendation problem on experience knowledge of researchers is reduced.

Description

technical field [0001] The present invention relates to the technical field of activity-based social network, in particular to a method for extracting social activity features in the activity-based social network. Background technique [0002] Activity-based social network is a new type of social network that combines users' online virtual social relationship with offline actual face-to-face communication. Users form interest groups online. Members in the group can communicate online and publish activity notices. At the same time, users can actually participate in offline social activities initiated by the group at the scheduled time and place, and communicate face-to-face with members in the group. The active social network realizes the effective combination of the real world and the virtual world of the network. [0003] At present, the content of activity-based social network research mainly includes the following aspects: community detection, recommendation problem, use...

AI Technical Summary

Problems solved by technology

[0006] The feature selection of existing social activity recommendation algorithms largely depends on the experience of researchers. The recommendation effect of the first scheme depends on the calculation of influencing factors, and the selection and calculation methods of influencing factors vary depending on the experience of researchers. There are differences, and each has its own reasons; in the second scheme diagram, there is no unified normative method for setting weights among entities. Each researcher looks at the problem from a different perspective, and the setting of weights also varies greatly. Even with the same graph model idea, the recommendation results vary due to the different weights between entities in the graph

It can be seen that the modeling of features will determine the recommendation effect to a large extent. Manually selecting features is a very laborious method that requires professional knowledge. Whether the selected features can effectively improve the algorithm effect largely depends on experience and luck. , the optimal recommendation effect cannot be guaranteed

Images