The sentiment analysis process: technology, not telepathy

Forget the telepathy you’re used to from science fiction.  Unfortunately, reading the minds of customers isn’t as easy as donning wired headgear and watching their thoughts as a movie-style montage.

That’s not to say that technology can’t help you tap into the customer mindset, however. Sentiment analysis isn’t the futuristic telepathy you’ve seen in film, but it can help you understand how your customers are really feeling. Here, we explain the sentiment analysis process and how to use it for smarter service.

What is sentiment analysis?

Sentiment analysis uses machine learning methods to extract, identify and categorise the sentiment of content. It’s sometimes referred to as opinion mining, or emotion AI. Put simply, sentiment analysis is the process of reading the emotional tone behind a piece of text and identifying the attitude and feelings of the writer.

Using natural language processing and biometrics, the technology automatically studies information. From there, it categorises that information with a positive or negative sentiment score.

Why use it?

At present, debate surrounds the accuracy and necessity of sentiment analysis software. Let’s face it: if an organisation has a handful of customer emails to respond to, the most effective way to measure sentiment is to manually read through them and reply accordingly.

But what if you have thousands of emails? Machines will never be able to gauge emotion as well as a human being — not even humans will agree all the time. However, as customer demand increases, organisations are looking for new ways to improve their transactions and manage enquiries. Enter the sentiment analysis process.

What can the software do?

For businesses, sentiment analysis can automatically identify a customer’s attitude from any piece of incoming text. This can be taken from an email, social media message, live chat session, SMS or document —allowing the organisation to analyse text from any of their multiple communication channels.

But the best kind of sentiment analysis process doesn’t stop with analysis. Using business process automation software, sentiment analysis results should end in automated action.

Let’s say a company received an email which rated as ‘highly negative’ according to sentiment analysis. This score can trigger automated escalation notifications to management. Depending on the nature of the business, words to trigger negative sentiment analysis might be ‘cancel’, ‘refund’ or ‘urgently’. The software can then automatically create an escalated support ticket with the customer’s enquiry, flagging it as important. This places the ticket at the front of the queue, enabling customer service agents to deal with it as quickly and efficiently as possible.

Different strokes

Intelligent business automation software should not limit its categories to simply a ‘positive’ or ‘negative’ sentiment. No two organisations operate in the same way, so it’s vital that sentiment analysis tools offer flexibility on keywords.

A negative term for one organisation may not be so urgent for another. For example, an online gaming company may want to highlight terms such as ‘crashes often’ or ‘lagging’. A subscription based food service, on the other hand, may opt for phrases such as ‘allergies’ or ‘there’s a slug in my kale’. (Well, maybe not those exact phrases, but you get the drift.)

Conditional action

Organisations can then generate conditional execution of actions into the sentiment analysis process. They can also create different classes of sentiment, like sales enquiries, support or customer services.

For example, if a customer sends an email to an ecommerce retailer to complain that they have not received a refund they expected, the email and details can be directed straight to the accounts team. Then, business automation software can deliver an immediate response to the customer to inform them that the company has received the email, and that the accounts team will be in touch.

The bigger picture

Sentiment analysis doesn’t only apply to incoming messages from customers. It also extends into mentions of organisations or certain product names on social media. For example, organisations can use sentiment analysis to monitor Twitter for global messages. From there, they can create automatic responses for positive tweets, or alerts to management when negative sentiment tweets are identified.

Naturally, the accuracy of sentiment analysis depends on algorithms. This can be complex when dealing with unpredictable human speech and opinion. So, for any sentiment analysis process to be effective, it’s vital to employ a hybrid model. Use automated sentiment analysis, but use it alongside humans to ensure all customer interactions run smoothly. However, there are other ways to improve the accuracy of sentiment analysis software automatically.

Machine learning

Over time, intelligent sentiment analysis programs will use machine learning techniques to develop more advanced analysing tools. By solving customer problems and identifying enquiries on a regular basis, the software will begin to understand how to manage certain queries faster and more accurately. Our own ThinkAutomation, as an example, is a quick learner. The more data it consumes, the more intelligent it becomes.

The sentiment analysis process in practice

As customer demand continues to increase, organisations are looking for new technology to assist customer service operations. Naturally, implementing sentiment analysis technology isn’t a one-step solution to managing the thousands of customer interactions that some businesses receive on a daily basis. However, when implemented carefully, the benefits of sentiment analysis are substantial.

Sentiment analysis cannot provide the mind reading abilities we’ve seen portrayed in science fiction movies and novels. But for the transactional side of customer service, it certainly helps things along.

Useful links

Applying sentiment analysis software in customer service: how, where, and why

ThinkAutomation and sentiment analysis

WhosOn and sentiment analysis