The document proposes using N-gram IDF, an extension of Inverse Document Frequency, to classify bug reports into real bugs or non-bugs. It finds that an N-gram IDF model outperforms a topic modeling technique in classifying reports from three open source projects, with statistically significant improvements in an evaluation using cross-validation and separate training and testing datasets. The N-gram IDF is able to extract variable-length terms that serve as useful features for the classification model.