Text annotation using textual semantic similarity and term-frequency (Twitter)

Researchers on social-media understandably assert that the contributions social media has made on various sectors is massive. Business development managers today have directed a huge amount of effort in strategizing efficient collaboration with both customers and other organizations using social-media. Despite the visible impact social media has made, a lot of digitally shared information is yet to be revealed. Gradually twitter has become the main hub for many Information system researchers, because tweets can freely be accessible in real-time by any one. Motivated by earlier studies where IS researchers addressed big-data analysis and management by employing content analysis techniques, this paper proposes a novel approach to perform unsupervised classification of the tweets into different labels. It introduces a unique algorithm that uses semantic similarity between texts, Term-frequency and a determinant threshold to perform content analysis. The goal of this approach is to extract relevant features from a tweet thus reducing dimension and preparing training datasets that would be used to build classifiers