A social group classification method and system based on multi -characteristic fusion

Abstract

The invention discloses a social group classification method and system based on multi-feature fusion. The method is as follows: 1) For a dataset of a target social group, preprocess the trajectory data of each user in the dataset, remove noise and interpolate the missing position information; 2) Use a frequent sequence mining algorithm to extract the trajectory data from each user's trajectory data. The trajectory pattern of the corresponding user is excavated in the method, and then the trajectory pattern of the user is regarded as a time series, and the time series is encoded by LSTM to obtain the trajectory code of the user; 3) According to the social relationship, a graph network is generated, and the social The relationship is projected into the low-dimensional space, and the embedding representation of each user is learned; 4) The trajectory encoding of each user and the embedding representation of the corresponding user are combined into the softmax layer to determine the category of each user and realize the classification of the target social group. The present invention greatly improves the accuracy of group classification.

Description

Technical field [0001] The invention is a data mining-classification-group classification technology, involving a social group classification method and system based on multi-characteristic fusion. Background technique [0002] The purpose of the classification method is to construct a classification function or classification model based on the characteristics of the dataset (also often referred to as a classifier), which can map the unknown class sample to the given category. Generally, the classifiers obtained by machine learning can be expressed as a classification rules, decision -making trees, or mathematical formats; the classifiers obtained by deep learning are mainly used to find the most probability of label value with neural network structures such as CNN and RNN. [0003] At present, the existing group classification methods based on social networks mostly use the existing classification models to achieve social network user characteristics such as social relationship...

AI Technical Summary

Problems solved by technology

However, in the above methods, features such as user trajectories and network behaviors that can also reflect social attributes are often ignored

[0004] Although the above classification methods have achieved certain results in group classification problems, such methods require the use of data sets containing specific forms of feature values, which makes it difficult for irregular features such as user trajectories and network behaviors to participate in the classification process, or features There is a certain lack of information

In the group classification problem, such irregular features have important reference value for the classification results, and ignoring the irregular features may lead to a decrease in the accuracy of the classification results

Images