Rumor Identification Using Relevance-Based Rumor Recognition


Rumor is unverified information in circulation that may be true or false and has significant influence on different aspects of society. It can be spread via various channels, especially online social networks and messaging applications. It is important to understand the underlying causes of rumors in order to control their spread and avoid their negative impacts. Rumor can be categorized according to its credibility and importance. This paper aims to investigate the correlation between these two factors and to propose an effective model for rumor verification.

A rumor is an unsubstantiated, often false story that spreads from person to person in a community in a manner similar to the spreading of communicable diseases. It can affect the behavior of crowds for good or bad, and in some cases can even lead to riots. The rumor can be in the form of news, gossips, or even jokes. It can be about events in a locality, government, or even in the world.

Some scholars have studied the rumor-spreading process by analyzing its textual features, such as fuzziness and importance. However, these features do not fully describe the characteristics of a rumor because they only depict the feature of words rather than their persuasiveness to people. A more reliable attribute that can capture the credibility of a rumor is its relevance to people’s lives. In other words, if a rumor is about something that is relevant to people’s daily life, they are more likely to share it with their friends.

To investigate the relationship between relevance and rumors, we analyze the rumor documents of two datasets: Twitter and Telegram. We divide the rumor documents into TRs and FRs, and calculate the spread power of each document. The results show that a rumor with higher SPR has more spread power than a rumor without SPR. This suggests that the SPR is a useful indicator for rumor identification.

The effectiveness of the SPR model in identifying rumors is verified by conducting several experiments. The first experiment involves classifying the rumor documents using the SPR feature and other set of content-based features that have been used in previous studies. The second experiment is a cross-validation study that compares the SPR model with other popular models.

The final result shows that the SPR is more effective in recognizing rumors than other models. Moreover, the SPR is an effective rumor detection feature because it can help save the verification system’s time by acting as a preliminary step. This can allow the verification system to focus on examining the validity of documents with high SPR. It also helps to reduce the number of false documents that need to be verified. It is therefore important to include the SPR criterion in rumor detection systems. Moreover, the SPR can be used as an alternative to the traditional stance feature in a rumor detection system. This can improve the overall accuracy of a rumor detection system.