Υλοποίηση συστήματος ανίχνευσης δραστηριότητας λογαριασμών ρομπότ κοινωνικών δικτύων
Nowadays, the Internet has become the main, daily source of information formany people worldwide in the form of applications that simulate our society, theso-called social networks. The basic feature of social networks is the creation of aprofile, which is used by corresponding us...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Language: | el_GR |
| Published: |
2021
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/11610/21834 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Nowadays, the Internet has become the main, daily source of information formany people worldwide in the form of applications that simulate our society, theso-called social networks. The basic feature of social networks is the creation of aprofile, which is used by corresponding users as the most realistic representationof themselves in real life, aspiring to socialize with other users, the kind that theywould like to communicate with, in the real world, as well.This digital environment now hosts profiles that don’t belong to genuine users,the so-called Social Bots. The word “bot” is the short form of the word “robot”.Social Bots operate in an automated way and disturb the proper function of theaforementioned networks. Thus far, the scientific community has developed multi-ple techniques and suggestions for identifying and removing such malicious profiles.However, since Social Bots are a human creation, their behavior changes over time,imitating human behavior, to avoid being identified by the mechanisms of the net-works. This dissertation proposes a machine learning framework, which will initiallybe trained with a dataset, so that it can solve a classification problem: bot or notbot? In other words, the model uses the method of Supervised Learning. It is im-portant to note that the steps followed for the creation of features are thoroughlypresented and explained, along with the rationale which led to the selection of thebeforementioned features.Afterwards, the model uses the STL Framework, in order to obtain the ability tolearn from unknown data and to be able to identify profiles with deviating behavior.The creation of those features, as well as the structuring of the machine learningalgorithm was achieved through the Python programming language.Finally, while the number of chosen features is smaller in comparison with otherresearches in the same topic, the results were encouraging, indicating that the selec-tion of the features was efficient, meaning that those features can accurately describethe classification problem, so as to identify the differences between a genuine userand a bot. Concerning the implementation of the STL Framework, the results werepromising and the Precision of the model was increased. Nevertheless, the Recalldecreased, because the model can successfully identify profiles with deviated behav-ior, but there is a greater possibility that it will falsely classify them as genuineusers. |
|---|