Rumor detection and analysis is a complex and multi-dimensional problem in which the content, user’s communicational strategies, or diffusion of rumors can be taken into account. Several researches considering a dimension have been conducted. However, the semantics is an important gap that can be seen in existing studies.
To overcome this issue, I proposed evaluating texts leveraging a philosophical tool, named critical thinking, which is usually used to discover fallacies in reasoning. I hypothesized some fallacies which are applied in rumor-making as several semantic features. Then, to detect semantics in texts, I offered a methodology by which I would take 4 major steps that as the first step, for each semantic feature, there need to be a suitable dataset, as the problem is a classification. Then, I proposed using active learning method to overcome the labeling bottleneck and to achieve high accuracy using as few labeled instances as possible, thereby minimizing the cost of obtaining labeled data. In the second step, I suggested an architecture utilizing deep learning techniques to capture the sequential nature of my data and discover dependencies in different layers. After evaluating the importance of features and investigating how much correlation between rumors and features exists as the third step, in the last step, there can be a prediction, utilizing other researches’ data sets of rumor and non-rumor, to determine a post’s reliability.