Early Adopter Programs Overview

New feature: content organization, collaboration and filtering

This article describes a new feature that may not be available to all users. If you don't see this feature and want to learn more, please contact your Tethr admin.
Section only

What is Sentiment?

Tethr’s Sentiment model is designed to provide insight against overall agents' and customers' sentiment. By identifying factors such as top contributing reasons, behaviors, teams and agents that are influencing sentiment, businesses can provide tailored coaching against undesirable and preventable interaction events that maximize participant emotions, thereby improving your overall sentiments. 

Early Adopters help Tethr:

  • Validate and improve model accuracy 
  • Understand how the data is being applied to your organization
  • Shape the product experience

What you can do with Sentiment:

  • Easily identify your positive and negative conversations
  • Identify individual emotions that drive the overall sentiment
  • Monitor and track your sentiment over time to measure improvement 
  • Track sentiment within different levels of your organization (i.e. line of business, team, service, product, offer)
  • Analyze individual emotions and overall sentiment by different participants (i.e. agent vs customer)

How it works 

The Tethr sentiment model includes an overall sentiment rating ranging from negative to positive and 30+ individual emotions that are identified within a conversation. Different participants in a conversation are rated separately to provide a more detailed representation of the emotional context.

Sentiment modeling and analysis involves taking unstructured text and identifying one or more emotions represented by the text. To accomplish this, a large collection of example phrases, labeled with the emotions they represent is used as training data since any given emotion may be expressed in a  variety of ways. For example, to accurately model the emotion of “anger”, a large number of phrases must be provided, some of which represent anger and some of which do not. Mathematical relationships between words and word sequences in the phrases and the emotional label of anger, represented as a probability of 1 or 0, are then solved.

Once created, fully trained models representing different emotions can then be applied to phrases and conversation text. These models provide values for each phrase representing the likelihood to calculate overall sentiment representing the overall positive or negative sentiment for a conversation.



Articles in this section