Latent Semantic Analysis (LSA) for syslog correlation
Gabriel Slomovitz · 2017
Latent Semantic Analysis is a novel method to extract the principal components of a text corpus which has been initially used for categorization and information search. However, due to the significant results obtained, similar to human processing, LSA has become much more than a simple method to analyze text. In this work, we propose to use LSA in order to infer similarity degree of syslog messages by discovering hidden relations between them. Using real syslog message samples, we show that LSA is able to highlight the most correlated messages by topic. This method can be used to avoid complex event correlation systems which usually need signatures or rule set definitions and high expertise for its configuration.