A social robot detection method based on variational autoencoder and k-nearest neighbor combination

Abstract

A social robot detection method based on variational auto-encoding and K-nearest neighbor combination belongs to the technical field of abnormality detection. The present invention obtains public data of social robots through a network, extracts features through preprocessing, uses data for training, and then uses variational After encoding and decoding, the normal sample features are more similar to the initial features after decoding, while the abnormal samples are very different from the initial features. This method considers that in the large social network environment, the number of abnormal user groups is smaller than that of normal user groups, so in the process of data collection, the collection of abnormal users is relatively troublesome. The method proposed by the invention solves the shortcomings of high-cost labeling and unbalanced positive and negative samples in the existing social robot detection methods, and realizes efficient detection of social network machine users by reducing abnormal samples to participate in model training.

Description

technical field [0001] The invention belongs to the technical field of abnormality detection, and in particular relates to a social robot detection based on differential self-coding. Background technique [0002] With the great popularity and development of the Internet, a large amount of real online user behavior data has been provided for the study of human behavior. As of December 2020, the number of netizens in my country reached 989 million, and the daily active users of Twitter reached 192 million. As of September 2020, the number of monthly active users of Weibo was 511 million, and the average number of daily active users was 224 million. Such a huge number The number of users generates terabytes of data every day, which records the rich online behavior of thousands of users. Social media has become an integral part of people's lives to obtain and share information. In general, social media sites like Twitter and Weibo have brought us unprecedented opportunities to ...

AI Technical Summary

Problems solved by technology

In order to reduce the participation of abnormal samples in the training of the model, the present invention proposes a social robot detection method based on a combination of variational autoencoders (VAE) and anomaly detection, by using data for training, and then using variational autoencoders for encoding And decoding, the normal sample features are more similar to the initial features after decoding, but the abnormal samples are quite different from the initial features, the original features are fused with the decoded features, and then the abnormal detection method is used for abnormal detection. The method proposed in the present invention solves the problem of The shortcomings of high-cost labeling and unbalanced positive and negative samples in the existing methods of social robot detection are realized, and the efficient detection of social network robot users is realized by reducing abnormal samples to participate in the training of the model

Images