Improved Topic Modeling in Twitter Through Community Pooling

Social networks play a fundamental role in propagation of information and news. Characterizing the content messages becomes vital for different tasks, like breaking news detection, personalized message recommendation, fake users flow characterization others. However, Twitter posts are short often less coherent than other text documents, which makes it challenging to apply mining algorithms these datasets efficiently. Tweet-pooling (aggregating tweets into longer documents) has been shown improve automatic topic decomposition, but performance achieved this task varies depending on pooling method.In paper, we propose new scheme modelling Twitter, groups whose authors belong same community (group who mainly interact with each not groups) user interaction graph. We present complete evaluation methodology, state art schemes previous models terms cluster quality, document retrieval tasks supervised machine learning classification score. Results show that our Community polling method outperformed methods majority metrics two heterogeneous datasets, while also reducing running time. This is useful when dealing big amounts noisy user-generated social media texts. Overall, findings contribute an improved methodology identifying latent topics dataset, without need modifying basic machinery decomposition model.