Servicenow language model - Powering Conversational Experiences

[shivramanna]

Every company that wants to use Natural Language Understanding (NLU) for Virtual Agent has different conversational needs. This stems from variations in how users converse about a topic in a specific context and organization. ServiceNow NLU allows customers to create their own NLU models providing them with the flexibility and control around these models and the experiences it powers.  

However, there are a lot of commonalities across companies in how requests are raised, kind of issues employees report and how they talk about it. ServiceNow caters to a significantly broad base of customers who use our platform and we come across these patterns very often. Over the past couple of years, we've worked with customers and found patterns in how their users use language.  We have combined these commonalities, learnings and data to improve ServiceNow NLU through a pre-trained ServiceNow language model* that can benefit all customers.  

The newer version of ServiceNow NLU, made available with the latest Rome release utilizes this Servicenow language model behind the scenes. This benefits all customers while still providing them with the flexibility and control to create their own models and customize their conversational experiences. The benefits of the ServiceNow language model include: 

• A smarter model that can understand what the user is asking for – With ServiceNow language model, NLU can now understand your users’ language better. NLU will be able to better understand the intent of your user and drive them through right conversation/automation flows thereby deflecting a greater percentage of tickets that your ServiceDesk might have historically been tasked with.  

• Reduce development time – With NLU being powered by ServiceNow language model, the need for a large number of utterances or large vocabulary is reduced for training customer specific models. The overall time to implement good quality NLU is reduced as a result 

• Identify a broader range of IT catalog items – ServiceNow NLU can now identify software or hardware terms in your end users’ utterances automatically without requiring any additional annotation of utterances making it easier to build conversation flows 

This only marks a beginning, and the ServiceNow language model will improve over time as we source and train with more data from a wider range of customers. Also, the applications of this language model will extend to natural language interactions powering other experiences like Search and Question and Answering. 

In future articles, we will talk in more detail about specific features like Model Optimize that utilizes this language model  and how you can benefit to create better conversational experiences. 

*What is ServiceNow language model? 

ServiceNow language model is a pre-trained model that powers ServiceNow NLU and has been trained to understand enterprise terminology better. This language model is not only based on the industry leading BERT language model but improves upon it further by training on enterprise data. This has been made possible by utilizing millions of conversations, incidents, and knowledge articles across a range of anonymized ServiceNow customers.