Hybrid features for detection of malicious user in YouTube
Social media is any site that provides a network of people with a place to make connections. An example of the media is YouTube that connects people through video sharing. Unfortunately, due to the explosive number of users and various content sharing, there exist malicious users who aim to self-pro...
Saved in:
| 主要作者: | |
|---|---|
| 格式: | Thesis |
| 語言: | eng |
| 出版: |
2017
|
| 主題: | |
| 在線閱讀: | https://etd.uum.edu.my/6562/1/816170_01.pdf |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 總結: | Social media is any site that provides a network of people with a place to make connections. An example of the media is YouTube that connects people through video sharing. Unfortunately, due to the explosive number of users and various content sharing, there exist malicious users who aim to self-promote their videos or broadcast viruses and malware. Even though detection of malicious users have been done using various features such as
the content, user social activity, social network analyses, or hybrid features, the detection rate is still considered low (i.e., 46%). This study proposes a new set of features that includes features of the user, user behaviour and also features created based on Edge Rank concept. The work was realized by analysing a set of YouTube users and their shared video. It was followed by the process of classifying users using 22 classifiers based on the proposed feature set. An evaluation was performed by comparing the classification results of the proposed hybrid features against the non-hybrid ones. The undertaken experiments showed that most of the classifiers obtained better result when using the hybrid features as compared to using the non-hybrid set. The average classification accuracy is at 95.6% for the hybrid feature set. The result indicates that the proposed work would benefit YouTube users as malicious users who are sharing non-relevant content can be detected. The results
also lead to the optimization of system resources and the creation of trust among users. |
|---|