Elementary, my dear Watson -- IBM's cognitive system learns to spot differences in tone
Human conversations are littered with subtle indicators, like tone of voice or slang, which can make the meaning of the dialog very different from what the words actually say. Online the situation is made more complex still by the increasing use of emojis.
All of this can leave automated chatbots and customer service systems floundering when interacting with unhappy or frustrated individuals.
IBM Watson is seeking to tackle the issue by making a new Tone Analyzer feature available from today. Trained specifically using customer support conversations on Twitter, the new feature is optimized to help companies monitor agent/customer communications, detect anomalies, and highlight opportunities to improve. It can also help make chatbots tone-aware and give unique responses to frustrated, sad, or satisfied customers. Additionally, it can detect tone from common emojis, emoticons, and slang.
It can track how tones progress throughout conversations, and note where agents need to be more sympathetic or polite to customers' situations, or if they should show more excitement when they resolve the issue.
Tone Analyzer is already being used by learning and development company KNOLSCAPE to test employees with simulated customer service scenarios and evaluate their responses.
"The quality of communication between the company and the users is a top priority for businesses, especially an international one like KNOLSKAPE where the majority of communication happens with customer service representatives in English," says Chaithanya Yambari, director at KNOLSKAPE Labs. "Tone Analyzer for Customer Engagement gives valuable metrics which help in assessing the quality of this communication to train customer service agents to be consistently polite."
Tone Analyzer is available now on the Watson Developer Cloud and you can learn more about how it was developed on the Bluemix blog.
Image Credit: IBM