When rumours spread through social networks, their impact on human lives becomes more and more pronounced; hence, it is vital to research the fundamental rules of rumour transmission in online social networks and propose effective rumour management techniques. Since the outbreak of Covid-19 last year, the volume of fake news, rumours, and doctored stories and videos circulating on social media has risen. Recent news reports that the Maharashtra Police Department has been fighting the dissemination of false information on social media and the Internet for a long time. As a result, during the crisis, a total of 852 offences were recorded.
The Importance of Rumour Detection
Rumours can contain a wide range of information, from simple gossip to sophisticated propaganda and commercial material. According to the researchers, the phenomenon of viral marketing is based on rumour-like mechanisms, in which corporations use their consumers’ social networks on the Internet to promote their products via the so-called “word-of-email” and “word-of-web”. Furthermore, a group of scholars say that rumour-mongering is the foundation for gossip algorithms, a class of communication protocols used for large-scale information distribution on the Internet and in peer-to-peer file-sharing services. A variety of machine learning (ML) algorithms has been deployed to detect rumours on social media. There are three types of ML approaches: supervised, unsupervised, and hybrid. Let’s have a look at these approaches in general.
Sign up for your weekly dose of what's up in emerging technology.
Supervised Learning Techniques
A group of researchers from the United States framed several indicators in the social media ecosystem that enable users to judge information authenticity after the Yahoo research team introduced engineering elements for credibility evaluation on Twitter. Support Vector Machines (SVM), decision trees, decision rules, and Bayes networks were among the supervised learning approaches used. Similarly, scholars in China have addressed an early study on rumour analysis and identification on Sina Weibo, a famous Chinese social media site. SVM using the Radial Basis Function (RBF) kernel was trained in the first experiment utilising a subset of the originally provided characteristics (content-based, account-based, propagation-based). Furthermore, K-Nearest neighbour (KNN) was used on user-based features, and Naive Bayes classifier (NB) was used on content-based features in an annotated Twitter data set. The list goes on and on.
Unsupervised Learning Techniques
In the case of unsupervised learning, the situation is similar. Australian researchers proposed a clustering-based approach for detecting political rumours on Twitter. Furthermore, the researchers from Singapore regarded the detection of fake rumours on Sina Weibo as an anomaly detection process. They came to the conclusion that for rumours, an unsupervised technique is superior to K-means. In a similar vein, Indian researchers proposed four major methods for spotting rumours on Twitter, as a result of which, a greater proportion of tweets are classified as False Positives/Negatives due to sentiment scores above the threshold.
Researchers from Japan have also presented a framework for detecting rumours on social media sites. Online bulletin board kakaku.com, Japan’s most widely used marketing site, was used to collect word-of-mouth messages. To detect rumour information, a Swiss researcher created graph topology-based distance. It was used to track changes in the topology of concept graphs over time. Some Chinese studies blend the supervised and unsupervised learning methodologies. In order to do this, they extract elements from the content of retweets and comments and look for rumours on Sina Weibo. Scholars from the United States investigated rumour spreading trends in the social media ecosystem in a similar way.
Prospects for Rumour Detection in the future
With the rise of social media, all users now have quick access to information. Because there is no strict social media policy in place to regulate and monitor online activity, most individuals are disseminating bogus news on social media and have been victims of it. For detecting rumours via social media, many machine learning models have been developed. There is usually a lot of interest in detecting rumours on social media among researchers; however, a viable ML model is necessary. Scholars and ML startups will need to put in a lot of study time.