The Zendesk QA Sentiment filter identifies the sentiment and tone of customers and agents, and filters to those that are noticeably either positive or negative, so that you can identify the best conversations to review.
This article contains the following sections:
Using the Sentiment filter
The Sentiment filter shows you all of the conversations in which Zendesk QA has found either 'Positive sentiment' or 'Negative sentiment'.
The messages in the conversation that have positive or negative sentiment will be marked either with a green smiley face or a red frown face respectively.
You can also use the timeline to identify where these messages lie in the thread and quickly jump to them.
To turn on the Sentiment filter
- Follow the instructions to create a new filter.
- Add the rule Sentiment is Positive/Negative.
Understanding how the Sentiment value is calculated
We use state-of-the-art Natural Language Processing models to assign sentiment to customer messages.
Our data shows that sentiment strongly correlates with survey feedback values. In other words, negative conversations tend to get lower feedback survey values. This makes the sentiment filter an efficient tool for learning which kind of conversations results in unsatisfied customers.
Sometimes, negative sentiment does not necessarily mean that the customer is frustrated, but it can mean that the message describes a complex problem that is conveyed with some negative vocabulary of a standard language.
Of course, no machine learning model is perfect and the quality of the output is dependent on the type of text. Content such as production code, long sequences of numbers or letters, and other non-word text make it harder for the machine to properly process the input, and can lead to an inaccurate judgment.