Sentiment analysis in customer service: Understanding human emotions

Modern customer service cannot work without the support of technical systems. One of these systems is sentiment analysis, which can be used to identify the emotionality of a post or a caller's mood during a conversation. How does it work? And how can businesses use it to optimize their customer service?
Sentiment analysis had its first major practical application in the context of Web 2.0, when internet users were transformed from pure consumers to content generators who communicate and discuss the products and services of businesses. The first approaches to analysis were relatively simple. To begin with, lists of words indicating positive or negative statements were drawn up – for example, adjectives. Then, user comments were compared with these lists and categorized accordingly. Naturally, dividing them up into only two categories – positive and negative – was very simplistic. Over time, this rule-based approach has been refined and expanded.
Today, artificial intelligence (AI) is also used with sentiment analysis. What is required for a system like this to actually work intelligently is, as always when using AI, good training based on the largest possible amount of data and with the support of a team that oversees the training.
But that doesn’t mean that classic, rule-based approaches are out-of-date. Both methods have their strengths, and experts are in agreement that the best solution is to combine both approaches. For example, it may be that AI can presort pieces of information well, but the final categorization of the information works better with a rule-based approach. This could be due to it being much easier to implement than carrying out extensive additional training of the AI. Moreover, when a new sentiment analysis solution is set up, simple rule-based approaches can provide the data needed for AI training and offer an overview of what the systems should find and recognize.
From text mining to video analysis
The next step in sentiment analysis is to evaluate spoken language, for instance in a service center. For this, there are the same content markers as in written language, but there are also various acoustic properties of the speech signal, including volume level, pitch changes or the degree of overlap if speakers interrupt one another. Dr. Damian Borth, Director of the Deep Learning Competence Center at the German Research Center for Artificial Intelligence, is convinced that it won’t be long before AI can also recognize facial expressions. This means video chats could also undergo sentiment analysis. Systems can already reliably recognize the age and sex of a person in a photograph.
The obvious area of application for sentiment analysis is still the social web, where it can be used to categorize comments according to language and content. But the process can also be used – subject to compliance with all data protection regulations – to analyze consultations in the interest of measuring and improving quality. Additionally, sentiment analysis provides indicators which can be used to optimize human-machine interaction. For instance, these might be warning signs when communication technology has reached its limits and a human customer-support specialist needs to take over. Because even though the systems are becoming ever more powerful and the algorithms ever more sophisticated, human empathy is still irreplaceable.
Author: Editorial team Future. Customer.
Image: NIKOLAS HOFFMANN – AdobeStock