In a study of randomly selected 100 solved tickets resulting from email interactions with the Global Help Desk the correspondence between emails and initial ticket descriptions was found to not be significantly better than random assignment of tickets to emails. A semantically rich ontology was developed in OWL 2 to cover all the computer-related concepts in the tickets and emails. Some aspects of this ontology, such as email delivery success rate, had to be adjusted to take local conditions into account. The semantic concepts in this ontology present in each email, in each initial ticket description, and in each ticket solution were determined. The semantic distance and direction from the (possibly multiple) emails from the requestor to both the initial ticket description and the ticket solution was determined using a semantic metric between sets of semantic concepts that took into account both specificity and difference and the correspondence between these two vectors was measured for each ticket. The measured correspondences were not significantly better than random assignment between requesting emails and tickets, but were significantly worse than many methods of matching tickets with requesting emails, including picking the highest number of common word stems between the initial email from the requestor and the intial ticket description. Attempts to introduce either of these or related better-performing methods in practice were forcibly rebuffed.
SURPRESSED, now intellectual property of HP
SURPRESSED, now intellectual property of HP