Skip to content

Question: How do interaction analytics solutions determine emotion or sentiment in conversations?


Emotion detection and sentiment analysis in interaction analytics (IA) solutions examine events in live or recorded conversations and extract insights from the nuances in verbal and written communications. This is performed by analyzing linguistic, acoustic, or interaction events, or a combination of the three, occurring during the conversation. Linguistic events address the presence of keywords or phrases, intensifiers (e.g., very, totally), negations (e.g., not, never), and/or other text-based indicators (emoticons, all capital letters, speed of chat response, etc.), that are associated with positive or negative attributes. Acoustic events, which are typically associated with emotion detection, can include language patterning, volume, pace, agitation, intensity, tone, pitch, over-talking, silence, etc. Interaction events may be a customer’s request to escalate the call to a manager or to cancel a product or service. Emotion detection uses language patterning and acoustic metrics to identify and measure the types of customer and agent emotion in each interaction. Sentiment analysis utilizes natural language processing (NLP) to detect, extract, and classify whether the conversation is positive, negative, or neutral, and includes the ability to differentiate sentiment by agent or customer. Speaker-separated analysis is an important feature for these capabilities, allowing results to be filtered based on what is expressed by the customer versus the agent.

Emotion detection and sentiment analysis can be applied on a real-time basis to identify angry, frustrated, or unhappy customers or escalating situations. Based on threshold-based alerts, supervisors can be notified so they can assist the agent and/or join the interaction. Real-time guidance can also be triggered to provide agents with tips for acknowledging customer frustration, taking ownership of the problem, and getting the interaction back on track. On a historical basis, emotion or sentiment scores can be used as search criteria, along with other metadata (e.g., agents, teams, types of calls) to identify conversations and/or customer surveys with high emotional content, and positive or negative sentiment. Positive sentiment/emotion can be used to identify exemplary interactions for agent recognition. Using emotion detection and/or sentiment analysis in combination with root cause analysis identifies policies or actions taken by the organization that provoke a negative response. It can also be used to give organizations an indication of customer satisfaction and identify positive or negative customer experience (CX) trends.