WEAC: Word embeddings for anomaly classification from event logs

摘要: Dramatic progress has been made in the usage of semantic word embeddings for solving analogy tasks recent years. Word or vector representation words key to many advances natural language processing. This paper presents a novel application Word-Embeddings Anomaly Classification (WEAC), where we detect whether an event log entry is anomalous one not. Additionally, WEAC helps us classify anomaly by identifying feature(s) log. For example, unusual network activity such as store transaction server logging into dropbox.com would be automatically flagged because wrong feature associations entries corresponding works with two training models: Skip-Gram (SG) and Continuous Bag Words (CBOW). Negative sampling used boost training. The initial results on wikipedia text8 dataset, well investigation enterprise HTTP logs are promising. model achieved average detection rate 65–100% classification accuracy 85–100%. was superior state-of-the-art techniques.