Domain Prediction

How does Moveworks classify domains?

Previously, Moveworks determined the domain of an issue using a combination of four data points. In January 2023, as our models got increasingly accurate, we decided to simplify domain classification by predicting the domain based on the user utterance alone.

What happens when the domain is wrongly classified?

Sometimes, because of the nuances between certain domains and the intricacies of business processes, the domain predicted can be different from what some stakeholders expect. When this happens, there are two solutions we can take.

Firstly, we are looking into adding the domain of the KB article presented as an additional column in Answers Insights. This offers a more deterministic domain classification compared to machine learning classification. Even though this does not override the domain predicted by the model, providing this data allows stakeholders to do their own analysis after downloading the data.

Secondly, if an organization prefers to have the domain predicted based on the utterance overridden by the KB domain or handoff domain, we can enable this setting on our end.

Entity Prediction

How does Moveworks detect entities?

In every user utterance, Moveworks extract all the entities that refer to business applications, geographical locations, people, companies, hardware, formalized processes etc. These entities allow Moveworks to better resolve employees’ questions or issues.

Moveworks detects multiple entities in each user utterance and then selects one entity as the primary entity, which is referred to as the “topic” on our dashboards.

For example, in the utterance “What is the wage limit for social security tax,” “wage” and “tax” are both detected as entities. Occasionally, the primary entity selected by the machine learning model may not be aligned with what stakeholders might expect, depending on how the utterance is phrased, the business processes, and the model performance in that instance.

How granular can the entities get?

Entity detection serves the purpose of allowing the bot to serve the best resources in order to resolve employees’ issues. We are constantly adding important entities to our corpus.

While the entities data on Answers Insights could also be helpful for reporting on users’ issues, adding overly granular entities can interfere with the bot’s ability to serve up the best resources. For example, we have “payroll” as an entity to provide the signal the machine model needs to serve up the best resources, and adding “payroll change” as an additional is not a suitable change.

Can I request certain entities to be renamed?

Yes, you can request certain entities to be renamed as long as the change applies to all instances of this entity. For example, you could elect to rename “Disability insurance” as “Disability”.

At Moveworks, we are currently building the next generation of data products and machine learning tooling to make it easier for you to understand how the bot works, diagnose issues, report issues, and perform fixes. Thank you for your partnership.