On Automating XSEDE User Ticket Classification

The XSEDE ticket system, which is a help desk ticketing system, receives email and web-based problem reports (i.e., tickets) from users and these tickets can be manually grouped into predefined categories either by the ticket submitter or by operations staff. This manual process can be automated by using text classification algorithms such as Multinomial Naive Bayes (MNB) or Softmax Regression Neural Network (SNN). Ticket subjects, rather than whole tickets, were used to make an input word list along with a manual word group list to enhance accuracy. The text mining algorithms used the input word list to select input words in the tickets. Compared with the Matlab svm() function, MNB and SNN showed overall better accuracy (up to ~85.8% using two simultaneous category selection). Also, the service provider resource (i.e., system name) information could be extracted from the tickets with ~90% accuracy.