Automatic incident processing tasks
The large companies Service desks process a huge flow of requests and incidents. To radically improve their work, it is necessary to automate some tasks:
- Detect mass failures and service denials in real time. Inform the dispatcher and the head of the support service about them in order to take the necessary measures in a timely manner.
- Reduce the time for incident routing. Automatically determine when registering an incident whether it relates to a typical, mass or an already known problem.
- Offer a ready-made solution if the incident relates to a known problem.
- Receive up-to-date information for analyzing the quality of the supported systems and services. Generate reports regularly to manage product quality and properly allocate the efforts of developers and support staff.
To effectively automate incident handling, you need to solve two interrelated technical problems:
- Clustering. Identify a list of incident groups, including previously unknown groups that may appear as a result of failures and mass service problems.
- Classification. Check whether each incident belongs to a particular group.
NLU technologies for incident handling
Such tasks are usually solved using machine learning (ML). However, applying ML methods to processing technical support requests is not so easy. Descriptions of incidents that are similar in meaning may use different words (but not synonims), while incidents that are different in meaning may use the same words.
In order to qualitatively group incidents, it is necessary to extract the essence from their text description. This can be achieved using NLU (Natural Language Understanding) technologies. Processing incident descriptions using NLU includes several stages:
- analysis of the grammatical structure of sentences
- determination of the lexical basis of each word
- search for concepts, the formally defined terms
- building a “semantic portrait” of each phrase, extraction of the formal statements.
How does EKG Language Processing work?
Our algorithm evaluates the commonality of incidents, even if they are described using different words. It "understands" negation and modalities - such expressions as "can", "must", "requires".
We present DataVera EKG Language Processing - a tool for natural language texts processing. It can be used for clustering and classifying incidents. The tool integrates with any Service Desk systems and is deployed in the Customer's infrastructure. You do not need to set incident groups and typical problems in advance - the service will determine them itself. You can manage the service, helping it to determine incidents that are similar in meaning, or showing the difference between similar incidents, different from a business point of view.
Processing results are displayed in the platform's user interface. Incidents that were successfully grouped are linked to typical problems. The names of typical problems are formed automatically using summarization. In the properties of each incident, we can see elements of a structured description of its meaning, which are used when comparing with other incidents. These descriptions contain links to the concepts mentioned in the incident description. The platform interface allows analysts to explore the results of incident classification and tune the algorithm. The technical support specialists are still working in their familiar Service Desk system, where the classification results are exported.
If it is important for you to improve the work of the technical support service and improve problem management, contact us. We will demonstrate our product and discuss how to solve your tasks using it.
