Automatization of incident resolution

• Autor(a) principal: Costa, Jorge Tafarel Morais
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/20472
• topic: Incident management; Machine learning; Natural language processing; Text mining; Information retrieval; Gestão de incidentes; Aprendizagem automática; Processamento da língua natural; Processamento de texto; Recuperação de informação; Gestão de sistemas de informação; Processamento da linguagem

Incident management is a key IT Service Management sub process in every organization as a way to deal with the current volume of tickets created every year. Currently, the resolution process is still extremely human labor intensive. A large number of incidents are not from a new, never seen before problem, they have already been solved in the past and their respective resolution have been previously stored in an Incident Ticket System. Automation of repeatable tasks in IT is an important element of service management and can have a considerable impact in an organization. Using a large real-world database of incident tickets, this dissertation explores a method to automatically propose a suitable resolution for a new ticket using previous tickets’ resolution texts. At its core, the method uses machine learning, natural language parsing, information retrieval and mining. The proposed method explores machine learning models like SVM, Logistic Regression, some neural networks architecture and more, to predict an incident resolution category for a new ticket and a module to automatically retrieve resolution action phrases from tickets using part-of-speech pattern matching. In the experiments performed, 31% to 41% of the tickets from a test set was considered as solved by the proposed method, which considering the yearly volume of tickets represents a significant amount of manpower and resources that could be saved.