Article Preview
TopIntroduction
In the past, people received their information through various communication channels such as TVs, radios, newspapers, and magazines. However, since the extremely high prevalence of smartphones, human’s life style has been changed significantly. There are more than one billion smartphones sold worldwide each year. People rely on smartphones to perform their daily activities like texting, checking messages, watching news, playing games, paying fees, etc. Mobile messages become the major information and news source for many people. Unfortunately, many messages received on the devices may not be correct. Incorrect information and news from the traditional channels are a big headache for societies. The problem is even worse for mobile messages since the messages could come from anywhere, individuals or organizations, instead of few fixed sources. Some of the messages may not be correct and are delivered intentionally or unintentionally. For example, the claim of election fraud with mail-in ballots has many people lose their faith on election justice. This claim is arguable, but at least should not be broadcasted without further proof, and it may be sent intentionally to affect the election results. This kind of mobile misinformation could have huge impacts on societies and needs to be identified and stopped. The consequences of health or medical misinformation are even worse than the ones of generic misinformation because the former may be fatal. For example, the misinformation of eating garlic able to prevent COVID-19 may lead people to lower their vigilance against coronavirus. Figure 1 shows the possible impacts of misinformation during a pandemic.
Figure 1. Possible impacts of misinformation during a pandemic
Data mining is an effective tool to fight misinformation (Horowitz, 2021). This research tries to propose a method for effectively identifying mobile health misinformation, especially pandemic misinformation, sent to users’ devices by using various mobile information retrieval and data mining technologies including lexical analysis, stopword elimination, stemming, and decision trees. The mobile text messages are classified into five classes: true, fake, misinformative, disinformative, and neutral. Users could make better judgement regarding the messages after consulting the classes found by our method. For example, if a text message is labeled as fake, the user may ignore it just by browsing its title without reading the details. Experiment results show the accuracy of the proposed method is above the threshold value 50%, but is not optimal. It is because the problem, health misinformation identification, is intrinsically difficult and the text messages do not provide much information to dig. Further refinements are needed before it is put to actual work.
The rest of this paper is organized as follows. Section 2 presents related works on misinformation identification. The system structure and its components are given in Section 3. Section 4 proposes our major work, a decision tree, for detecting mobile health text misinformation. The experiment results and discussions are given in Section 5, followed by a conclusion including future research directions and references.