Entropy-Based Model for Estimating Veracity of Topics from Tweets

Micro-blogging sites like Twitter have gained tremendous growth and importance because these platforms allow users to share their experiences and opinions on various issues as they occur. Since tweets can cover a wide-range of domains many applications analyze them for knowledge extraction and prediction. As its popularity and size increase the veracity of the social media data itself becomes a concern. Applications processing social media data usually make the assumption that all information on social media are truthful and reliable. The integrity of data, data authenticity, trusted origin, trustworthiness are some of the aspects of trust-worthy data. This paper proposes an entropy-based model to estimate the veracity of topics in social media from truthful vantage point. Two existing big data veracity models namely, OTC model (Objectivity, Truthfulness, and Credibility) and DGS model (Diffusion, Geographic and Spam indices) are compared with the proposed model. The proposed model is a bag-of-words model based on keyword distribution, while OTC depends on word sentiment and DGS depends on tweet distribution and the content. For analysis, data from three domains (flu, food poisoning and politics) were used. Our experiments suggest that the approach followed for model definition impacts the resulting measures in ranking of topics, while all measures can place the topics in a veracity spectrum.