*6.2. Group View*

The group view is aimed at displaying the whole topology of the network and helping us make a preliminary classification of data. Multidimensional scaling (MDS) reducts based on the distance between points, so if the points are more similar, the closer they are after dimension reduction. With the using of it, we can reveal the distribution of each ego by their features and find out whether the algorithm results are effective (T1). The result is shown in Figure 4a, and in order to distinguish anomaly scores of each ego, we color code their score into five colors, representing their anomaly degree (T2). Scatter plot can help experts to have a detailed understanding of the internal situation of the social network. However, due to dimensionality reduction, on the one hand, it is impossible for experts to have an understanding of egos through x and y coordinates, so we need to provide the specific characteristics of each ego to help them make decisions. As shown in Figure 4b, we design a list based on the ranking of LOF scores to show the detail of each ego. On the other hand, the MDS

dimension reduction brings similar points together. With the increase of data, visual clutter becomes more serious and we cannot understand the overall situation of the network, so we provide statistical data based on the detection results. After experiments, we find that the score of users below 1.5 are more than 90%, so we take 1.5 as the boundary and classify the users into 6 groups. We select the maximum, minimum and average values of alters and contact times for each segment. These allow us to have an intuitive understanding of each fraction. Above all, we design a statistical view to help experts analyze. (T2). The design is shown in Figure 4c.
