As the election for the next president draws near, there is more attention in the media. The importance of media is very high as it is a means of communication for everyone around the world. It is supposed to report the truth for recent events, which is why it is important that the news is true. If the news isn’t true, then it is considered fake news and can affect real life events such as the election. This can lead to a devastating result such as the election of an unworthy president. Another recent event in which fake news has impacted real life events is the spreading of false information regarding COVID-19. There are countless fake news about the current virus such as “Dark skin people are protected from the infection” or “Hand sanitizer can’t kill viruses”. If people believed in the former and decided to go to the public without safety measures, it would result in a higher chance of getting the virus and may result in death. This is why it is important that fake news is ceased and deleted as soon as possible to save people around the world.
The challenges of identifying fake news can be broken down into three parts:
- Unverifiable Data
- Huge Volume of Data
- Evolving Behaviours
The use of algorithm detection often leads to false-positives and incorrect results near the edge cases(scenarios where the shared news is hard to determine as it is on the borderline). Instead, a better approach is to calculate a reliability score for each content. Automatically allow or disallow the content if the reliability score is high or low respectively. When the reliability score is around the middle, use graph theory to identify the likelihood of fake news.
Such a graph is created using three nodes with the relationship shown in figure 1:
- Account(age, history, connections)
- Post(Timestamp, flags/reactions, IP address)
- Article(Host URL, author)
These three nodes can help to find various connections that can be an indication of whether the content is fake or not. For example, accounts that are relatively new and share a content should be flagged as unreliable. Usually accounts with purpose to spread fake news don’t only share one content but multiple content. This would result in a graph like figure 2.
Another easily identifiable link that suggests fake news is one that is called “link farm” in which multiple accounts are used to increase the popularity of an account. Figure 3 shows this exact scenario where three accounts from the same IP address all share the same three articles which results in a diamond shape.
The above example is also a scenario which can be described as unusual behavior. It can be hard to detect using algorithm detection without graph theory.
In conclusion, it is important to stop the spreading of fake news as it can severely impact the outcome of various real life events. Thanks to Cambridge Intelligence, they have used the above to create such a program. This program is called KeyLines which attempts to filter fake news using the graph theory and a reliable data set from BS Detector and Buzzfeed Fact check.
Reference:
- Disney, A. (2017, February 28). Using graph technologies to weed out fake news. Retrieved October 09, 2020, from https://cambridge-intelligence.com/detecting-fake-news/
- Doughton, S. (2020, July 05). COVID-19 meets Election 2020: The perfect storm for misinformation. Retrieved October 09, 2020, from https://www.seattletimes.com/seattle-news/health/covid-19-meets-election-2020-the-perfect-storm-for-misinformation/