Recognizing anomalous user behaviors in online social network using recurrent neural networks

نوع: Type: thesis

مقطع: Segment: masters

عنوان: Title: Recognizing anomalous user behaviors in online social network using recurrent neural networks

ارائه دهنده: Provider: Fereshteh Ebrahimzadeh

اساتید راهنما: Supervisors: Muharram Mansoorizadeh

اساتید مشاور: Advisory Professors: Mohammad-Reza Feizi-Derakhshi

اساتید ممتحن یا داور: Examining professors or referees: Mehdi Sakhaei-nia-Hassan Khotanlou

زمان و تاریخ ارائه: Time and date of presentation: wednesday-20 oct-at 16 :00 pm

مکان ارائه: Place of presentation: room 04

چکیده: Abstract: With the development of Internet technology in recent decades, various aspects of lifestyle have been influenced by the digital world. Therefore, social networks have become common among different classes. Users to communicate and perform their social activities to these social networks They join that in these networks, according to the sequence of activities, there is a possibility of security failure for each user, and if the abnormal behavior of users is not detected and prevented in time, it will cause attacks. In the meantime, selecting appropriate features in analyzing the abnormal behavior of users and also finding a new method using recursive neural networks and considering the separate modeling of the balanced behavior range of each user who interacts with another user to reduce errors in detecting behavior. Abnormal user can be challenging. This dissertation examines the problem of analyzing abnormal behavior of users using feedback networks. The proposed method is a feedback neural network that first uses one of the basic preprocessing statistical techniques. It is then classified with a recursive neural network. Finally, it is clustered by one of the clustering methods. Anomalies in social networks were improved. The proposed method was also used independently to detect anomalies. The proposed method in Python environment was performed on the 2008 Vast Challenge and Enron-dataset Email datasets. The results showed that the proposed method is more accurate.

فایل: ّFile: Download فایل