Information Cascade in Social Media and the Mechanisms of Prediction Model
In the discussion of information cascade, we learned that it is a phenomenon that has the potential to occur when people are trying to make decisions sequentially, with signals from people before them, for example by watching the actions of earlier people, reading the comments from previous buyers, and searching the review for previous movies. From these actions people can infer something about what the earlier people know, or even general common knowledge in a certain network. Considering the current situation that most people are used to frame their ideas, utter their thoughts, and comment on each other or simply make observations by collecting, reposting, or liking someone else’s posts anonymously, it is worth delving deeper into the information cascade and the patterns of how people follow the crowd on social media platforms.
In the research paper published by scholars from the Department of Computer Science & Engineering, National Institute of Technology Delhi, information cascade and propagation on Twitter has been analyzed carefully. The paper Predicting Information Cascade on Twitter Using Random Walk talks about how information spreads like sharing news could shape the opinions of the users and how businesses can better launch their products by analyzing the information diffusion on Twitter. Governments can also take advantage of such trends to gain support during election periods. In order to avoid intentional propagation or better distinguish the source of information, the scholars introduced a method that has not been covered in the class which is random walking. For most of the current predicting models on social media platforms, there are four prominent features that need to be taken into account.
First is the content-based features, which are features that are based on the literal content of the action. The basic idea is that if the content consists of keywords or phrases related to trending topics, COVID-19 in 2020 for example, it will eventually become popular. The ranking of the texts and topics are modeled by different NLP algorithms. This feature is especially easy for Twitter analysis and prediction due to the text-based nature of Twitter compared to image-based Instagram. The second feature is what we discussed in the class about the network effect and how a connected network of users can be used to model the flow and cascade of information. The third feature is related to the structure of the network, including the density of a cluster in a network. This leads people to the idea of clubs and sub-groups on Facebook where the user network is more dense than a random forum and therefore a complete cascade is more likely to be formulated. Finally, there is a temporal feature of the prediction that the model would need to take into account of the recency of the topic. The researchers suggested that with all these features combined with a random walk model, an integrated version of the predicting model can provide more accuracy for information cascade prediction, people will more likely to find out the reason behind a popular topic or how popular the topic will achieve beforehand. On the positive side, it is helpful for businesses and institutions to gain the most popularity and have the information spread out on the social network the most. On the other hand, it may also be exploited by other parties for purposes that may lead to unpleasant results.
Source: https://doi.org/10.1016/j.procs.2020.06.024