An Automatic Approach to Validating Log Levels in Java']Java

A log statement is used to record important runtime behavior of software systems for diverse reasons, which is inevitable to develop most of the software systems. However, developers do not tend to deeply consider an appropriate log level in their source code. In order to address the issues, this paper proposes an automatic approach to validating log levels in Java in consideration of the syntactic as well as semantic features. We first build up the Word2Vec model and generate semantic and syntactic log feature vectors, then train the machine learning classifiers to automatically validate the log levels. For the evaluation, we collected six open source projects of the message-oriented middleware domain, and obtained the 88% precision and the 87% recall respectively.