Detecting Cyber Security Threats from Short Noisy Texts

• Used Deep Learning and Natural Language Processing techniques to build a model that can identify cyber security events from Tweets.
• The neural model utilizes meta-embeddings learned from domain-specific word embeddings and task-specific features to capture maximum contextual information as tweets are short and noisy.
• Also employed a two channel architecture that combined both LSTM and CNN. Used meta-embeddings for feature learning via LSTM and CNN, and their feature maps are concatenated with contextual embeddings in the Fusion Layer.