Online Information Cascade in Response to Crisis
This study is focusing on information cascade on Twitter during crisis situations. The researchers demonstrated a model of the diffusion of actionable information in the context of an armed robbery occurred near to the Rensselaer Polytechnic Institute (RPI) campus. Based on the results, the researchers explained their findings about the following three aspects of information cascade in response to a crisis.
- What kinds of information are requested or shared
- What kinds of information cascades or patterns are observed
- What these patterns tell us about information flow and the users on Twitter
Instead of focusing on the details of their findings, I’m more interested in the model they adapted to analyze the above three questions. The model has three major components. Firstly, a network of nodes with configurable attributes, which is regarded as the input of the model and stands for individuals in the collected data set. Secondly, the weighted edges in the network that represent the relationship between individuals (nodes) in terms of trust, that is, how likely the information will be considered as believable when transmitted from one node to another. Thirdly, internal information sources that spread the information through the network through “seed” nodes that “follow” the information source.
Given these three major components, the interactions within the network is easy to unfold. Firstly, information sources give out messages, and seed nodes spread it out by retweeting. Then, other nodes receive information from different sources and evaluate the credibility of each received information. Based on their evaluation, each node will behave differently. They can be susceptive, indifference, believed, etc., but the ultimate evaluation of their action is – whether they shared the message or not.
This model integrated some aspects of the concept of information cascade we learnt in class. The most obvious commonality is the four basic ingredients of information cascade.
- Some decision to be made
- People make the decision sequentially over time
- Each person has some private information
- You can see what the earlier people did, but not what they know
Which corresponds to the following four ingredients in the crisis context described in the study.
- Whether to share (tweet/retweet) the message or not
- Individuals make their own decisions as warnings and new information were kept sending out (e.g. Prior first warning, first warning, second warning, etc.). However, in this case people are not making decisions exactly one by one, some of them might make decisions simultaneously, which is slightly different from what we assumed in class. But the idea that some people make decisions after the others are largely the same
- Some individuals might witness the crisis or hear other people talking about it; some users follow the information source on Twitter while others not.
- Individuals can see whether other people shared on Twitter and what they shared, but they don’t know to what extend that person know about the crisis (the private information that person possess)
These four ingredients built the common grounds for us to understand some of the online information cascade patterns in emergent situations. However, this model also possesses novel concepts that are not covered in class but are interesting to think of in terms of how the patterns provide new insights on the users involved in the information cascade.
Firstly, the model takes the relationship between individuals into consideration. Since crisis is a very special kind of situation where credibility is of the highest priority, it’s important to take the concept of trust into consideration. The researchers did this by using weighted edges as the representation of the degree of trust among people and information sources, which is a very smart way to incorporate graph theory as a tool to analyze individual behaviors during information cascade.
Secondly, their model puts more emphasis on the role of information sources. This is also reasonable, because unlike the experiment we learnt in class, this study focus on real life situations instead of laboratory context. In the former network, different nodes do have different powers (e.g. Formal sources are more reliable, and therefore more powerful than personal, informal sources), but in the latter one, peoples’ powers are approximately the same (e.g. Everyone draws one marble from the urn).
Finally, their emphasis on power differences also gave rise to one more new concept – “degree centrality”, which suggest that the degree to which a user is retweeting and being retweeted indicated the role of that user in the information cascade. More retweeting indicates that the user propagates information, while frequently being retweeted suggest that this user might be a reliable source of information.
Source: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.299.8817&rep=rep1&type=pdf