Information Cascades and Node Popularity
Information cascades occur when people make decisions sequentially, where later people watching the actions of earlier people infer something about what the earlier people know. It is not mindless imitation, but rather the result of drawing rational inferences from limited information. So how do people make decisions on city-wide activities or answer the question of ‘‘do I want to take part in the activity’’ and ‘‘am I interested in the activity?’’ Researchers at Beihang University in Beijing went about researching these questions and propose a new information spread model.
The research was conducted on a large online social network of DouBan, which is a social hobby and user-generated content network. The focus was on the characteristics of the cascade subgraph made during the decision spread process. These characteristics include the cascade’s scale, scope, diameter, and density. The mechanisms that lead to information diffusion are noted as equal probability, node similarity, and node popularity. Douban provides information on “city-wide activity” which allows users to decide if they would like to participate in these activities, and that the researchers can use to analyze decision-making processes. Once a user decides to participate in an activity, this decision is sent to all their fans’ pages, so this activity spread direction is opposite to the fan’s direction. Using Douban’s API the researchers tracked activities that users took part in. The new model they created to track information diffusion works as follows: the model randomly selects a node in the network as the node which has made a decision, namely the beginning node. In each iteration, researchers decided whether information diffusion has occurred by considering every node which has made the decision in proper order and for each node connected with this node by the in-edge according to one of the three mechanisms defined above. The model repeats this iteration until the diffusion process stops.
It was found that equal probability, fan’s popularity, and PageRank have more obvious impacts on information diffusion, especially the fan’s popularity. Furthermore, the effects of equal probability and the fan’s popularity are stronger than that of the PageRank, so information diffusion has a great relationship with the nodes’ popularity.
The research found that a node’s popularity plays an important role in the spread of information. We can think back to our discussion of power in social networks and the importance of who is connected to who. It was found that by setting the key node based on the information spread, the researchers could predict the scope of the information spread. Overall, the research showed how easily people are swayed according to the activities they see others engaging in. We can predict how people will act according to who they follow, and see that herding occurs often and is heavily influenced by a node’s power in the network structure.
Sources:
Chao Tong, Wenbo He, Jianwei Niu, Zhongyu Xie, A novel information cascade model in online social networks, Physica A: Statistical Mechanics and its Applications, Volume 444, 2016, Pages 297-310, ISSN 0378-4371, https://doi.org/10.1016/j.physa.2015.10.026