The popularity of social networks provides people with many conveniences, but their rapid growth has also attracted many attackers. In recent years, the malicious behavior of social network spammers has seriously threatened the information security of ordinary users. To reduce this threat, many researchers have mined the behavior characteristics of spammers and have obtained good results by applying machine learning algorithms to identify spammers in social networks. However, most of these studies overlook class imbalance situations that exist in real world data. In this paper, we propose a heterogeneous stacking-based ensemble learning framework to ameliorate the impact of class imbalance on spam detection in social networks. The proposed framework consists of two main components, a base module and a combining module. In the base module, we adopt six different base classifiers and utilize this classifier diversity to construct new ensemble input members. In the combination module, we introduce cost sensitive learning into deep neural network training. By setting different costs for misclassification and dynamically adjusting the weights of the prediction results of the base classifiers, we can integrate the input members and aggregate the classification results. The experimental results show that our framework effectively improves the spam detection rate on imbalanced datasets.

With the rapid development of social networks, spam bots and other anomaly accounts’ malicious behavior has become a critical information security problem threatening the social network platform. In order to reduce this threat, the existing research mainly uses feature-based detection or propagation-based detection, and it applies machine learning or graph mining algorithms to identify anomaly accounts in social networks. However, with the development of technology, spam bots are becoming more advanced, and identifying bots is still an open challenge. This paper proposes a new semi-supervised graph embedding model based on a graph attention network for spam bot detection in social networks. This approach constructs a detection model by aggregating features and neighbor relationships, and learns a complex method to integrate the different neighborhood relationships between nodes to operate the directed social graph. The new model can identify spam bots by capturing user features and two different relationships among users in social networks. We compare our method with other methods on real-world social network datasets, and the experimental results show that our proposed model achieves a significant and consistent improvement.

In recent years, privacy leakage events in large-scale social networks have become increasingly frequent. Traditional methods relying on operators have been unable to effectively curb this problem. Researchers must turn their attention to the privacy protection of users themselves. Privacy metrics are undoubtedly the most effective method. However, social networks have a substantial number of users and a complex network structure and feature set. Previous studies either considered a single aspect or measured multiple aspects separately and then artificially integrated them. The measurement procedures are complex and cannot effectively be integrated. To solve the above problems, we first propose using a deep neural network to measure the privacy status of social network users. Through a graph convolution network, we can easily and efficiently combine the user features and graph structure, determine the hidden relationships between these features, and obtain more accurate privacy scores. Given the restriction of the deep learning framework, which requires a large number of labelled samples, we incorporate a few-shot learning method, which greatly reduces the dependence on labelled data and human intervention. Our method is applicable to online social networks, such as Sina Weibo, Twitter, and Facebook, that can extract profile information, graph structure information of users’ friends, and behavioural characteristics. The experiments show that our model can quickly and accurately obtain privacy scores in a whole network and eliminate traditional tedious numerical calculations and human intervention.

Because social networks exemplify the phenomenon of homogeneity in complex networks, researchers generally believe that a user's privacy disclosure is closely related to that of the users around them, but we find that the related users studied in previous methods were not correct. That is, the analyzed user groups may have had nothing to do with the privacy disclosure of the target users. Since private information is time-sensitive, information held by users who are no longer in the same environment as the target user may no longer be true and have lost its value. For example, considering students and members of the working class, transfers to another school for further studies or job changes entail dramatic changes to most of their information. This lack of timeliness has an overarching impact on the effectiveness of social network analysis and privacy protection, but this problem has not been addressed by researchers. Therefore, we study and characterize this problem, add the user's behavior trace to solve this problem and measure the user's privacy status more accurately.