Twitter account classification using account metadata: organizationvs. individual

Organizations present their existence on social media to gain followers and reach out the crowds. Social media-related tasks applications, such as graph construction, sentiment analysis, bot detection, are required identify entities' account types. Some applications focus personal accounts, whereas others only need nonpersonal accounts. This paper addresses classification problem using minimum amount of data, which is metadata account's profile. The proposed approach classifies accounts either organization or individual, in a language-independent manner, without collecting accounts' tweet content. model uses long short term memory (LSTM) network for processing textual properties fully-connected neural numerical features. We apply our solution collection Twitter it one most widely used networks. Our classifier, based solely metadata, achieves an average 97.4% accuracy under 7-fold cross-validation. experiments show that qualified resource accurately estimating