You are here

Data Science in Action: An Algorithmic Approach to Process High Volume Service Data and Make Improvements to IT Services by Natural Language Processing and Machine Learning

Mrs. Wen (Bonnie) Hoschke

Western Sydney University

In today’s high education sector, many organisations have implemented IT service management platform to support the service delivery. Currently at Western Sydney University, we have ITSM platform which include a ticketing system as an interface with customer to request for service. With each ticket raised by customers, the service data is able tobe collected in the system. Every month there are thousands of tickets processed and handled by various resolving groups via the system. The descriptions for the service requests and solutions are captured in textual form. Some tickets are similar to each other. Some similar nonstandard requests could be converted to standard request that can quickly reach the correct resolving group and treated by formulated solutions to improve service efficiency. Similar incident tickets likely are sharing similar root cause. Group them together can help us to track down and develop the solution to eliminate the re-occurrences or mitigate the impact. There are hundreds and thousands of tickets logged in the system across different months and years. Tickets data, for example descriptions and solutions, are entered by many different users and at various stages.

As the descriptions for request/incident and solution are all free input text eld, it present challenging to group the similar tickets by basic analytics methods.

The high volume of ticket data makes it difficult to group the tickets by manually reading. Here we like to describe the works on using text mining techniques to identify and group the similar tickets together and the experiments on visualize methods for easy interpretation of the results. The methods include using natural language processing for data cleaning, parsing, stemming, tagging, etc. Then using a machine learning algorithm to cluster the data by their text similarity. From the machine learning, each tickets are labelled with a cluster identifier. The results also contain keywords to describe each cluster. With the outcomes from this method, it become lot easier to group similar tickets together by cluster identifier and to produce a recommandation list for task standardisation, incident root cause identification etc.