Leveraging uncertainty modeling for suspicious tweets detection

Analyzing information from microblogs like Twitter is an important issue for the modern society that confronts new challenges. Its Security is the most crucial problem that requires one to anticipate different events, to track specific information or given persons. The detection of suspicious tweets has a main task in this issue, however tweets are generally not well written, contaminated by errors and may use metaphors, slang terms, and colloquialisms. Probabilistic methods inherited from information retrieval are systematically applied to social data analysis. In this paper, we combine probability theory and fuzzy logic in a harmonious approach, where the first one is more related to frequencies, whereas the second concerns qualitative modeling of suspicion and normality. The experimental results show that the algorithm leveraging probability and fuzziness outperform the one based only on probability.