Real-time event detection and classification in social text steam using embedding

Taming data will always be a significant challenge in online social networks. These networks are rapidly becoming the emerging source for users to explore the primary sources to seek information in the form of events. Rich informational data can be extracted from various social platforms like twitter text streams for direct insights into enduring topics and classifying them based on their similarities. To address the research issues of event detection and classification, we model events as evolving clusters over a period of time. The inability of conventional clustering algorithms to process the data streams mandates the use of a fast yet robust method. Therefore this work employs quick comparisons of data coming from social streams relying on a twin network known as the Siamese network, which can detect the novel event based on clustering by comparing their content dependent feature. We also trained dataset derived from the social text stream from twitter and other sources, where embedding encode every word representation mapped to a vector. This representation of word into real valued vectors provides a specific processing task for event classification. Finally, we compared the proposed technique with the existing methods, and the results obtained through several experiments are a clear indicator of the efficacy of the proposed scheme.