All of us are aware of graph algorithms and we have seen in the past that some of them move towards NP-hard. In case of social networks too, we encounter humongous, enormous graphs. It is extremely challenging to analyse such huge graphs for patterns and behaviors. As a solution, the authors of the paper introduce "Social Action", a tool that allows for users to view visualizations of the graph based on different filters, detect outliers if any, find interesting patterns and trends and similar and dissimilar groups of people and their activities.
The main idea behind Social action is the Hierarchical clustering approach. The nodes are thus ranked by users on different criteria and are visually rendered. It keeps the user in mind and so presents information in a comprehensible, ordered list form with color coding and dynamic query methods to provide flexibility.
The feature that I liked is that the authors have taken care of the fact that the graph can grow breadth-wise and hence they give the users the ability to aggregate nodes based on link structure. I also found the method of their ranking the nodes interesting where importance is not given to individual nodes but rather on the structure information of the graph itself and that the rankings are assigned different colors. I also appreciate the idea of sub group detection and I 'm curious to know what happens in case of overlapping sub group structures. Also, I am not sure of how accurate the sub groupings are as the users themselves are allowed to aggregate the nodes based on some criteria. I think the users must be able to give custom criteria or merge two or more of the graph parameters that the authors discuss.
0 comments:
Post a Comment