....¹

For the sake of simplicity, we ignore here the timing of messages. Because system logs sometimes span a long time period, it would generally be necessary to address this issue, for example, by processing only message instances that occurred within a certain time frame.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... log.²

The method of using tf-idf weights [8] to rank words in documents in information retrieval engines, is comparable to our scoring method though it uses a different formula.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

....³

In case of ties, the coordinate with the larger index is ranked higher.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... feature-sets.⁴

Since the Frequent Message matrix is not a correlation matrix, its eigenvectors include imaginary parts and therefore cannot be plotted.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.