Sentiment analysis involves analysing text, normally online, to assess customer opinion. It uses Natural Language Processing (NLP) and machine learning to analyse extracts of text, categorising them as positive, negative or neutral.
Also known as opinion mining, sentiment analysis helps you keep tabs on what customers think. It’s useful for monitoring your brand reputation and qualifying customer reviews. You can also use it to assess and categorise customer service requests, organising by priority, urgency or type of help required.
You can use sentiment analysis to analyse data from several different sources, such as social media, review sites, your own website or CRM system and survey responses. In fact, here at SurveyMonkey we use sentiment analysis to evaluate the responses to your open-ended questions.
There are three main types of sentiment analysis. Which type you use depends on the level of detail you want to go into.
This takes the broad categories of positive, negative and neutral and breaks them down further into: very positive, positive, neutral, negative and very negative.
This type of analysis evaluates text to determine whether it reflects various emotions, such as happiness, sadness and frustration. To do this, it uses lexicons, or lists of words that have been categorised according to the emotion they express—for instance, “thrilled” to indicate happiness, “disappointed” to indicate sadness.
Aspect-based sentiment analysis involves determining not just what a customer feels about a product or service, but what exactly they are happy or unhappy about. For instance, in reviews for a gym, a customer might be satisfied with the equipment, but unhappy with the cleanliness or class booking system, and it’s useful to have that information.
Sentiment analysis is just one type of NLP, where artificial intelligence algorithms are used to extract meaning from human language. Search engines and digital assistant tools like Alexa and Siri use the same technology.
Sentiment analysis normally involves these steps:
Rule-based: using manually created rules and lexicons (word lists), this approach compares the number of negative and positive words to determine the overall sentiment.
Automatic: using machine learning, classifiers are trained on sample texts and tag sentences as positive, negative or neutral.
Hybrid: using a combination of the best bits of rule-based and automatic approaches.
3. You’re up next. Delve deeper into the results, for instance filtering responses by location, or filtering responses by emotion to identify trends and draw conclusions.
As you can see, sentiment analysis is a powerful tool for understanding what people are thinking. But it’s not without its challenges.
These challenges stem from the complexity of language and human expression. We express ourselves in so many different ways, using different language and slang. And it’s constantly changing. This makes rule-based analysis, used for emotion detection, incredibly difficult, as words need to be constantly added or recategorised. Rhetorical devices such as sarcasm and irony, as well as implicit meaning, can also stump sophisticated programs, meaning text is sometimes misclassified.
And that’s before we even start thinking about taking sentiment analysis across borders, with different languages, dialects, cultural specificities and unique forms of expression. It’s clear that multiple markets and languages can pose a problem for sentiment analysis.
Let’s imagine you work in the marketing team for a sportswear brand. You could extract product review data from your own website and from third-party vendor sites then run it through hybrid sentiment analysis. This will allow you to:
You can use sentiment analysis as a PR or brand reputation tool by analysing online mentions of your organisation across a range of media. With effective brand monitoring, you can stay up to date with the conversations your customers, investors and competitors are having about you. This allows you to detect and respond to unhappy customers before things escalate, and also to capitalise on what you’re doing well, and even improve your brand reputation.
For instance, let’s consider the incredibly influential Elon Musk. In recent times, his comments on Twitter have seen Samsung Publishing share prices soar, and Bitcoin plummet. Being aware of what people are saying about your company as it happens means you can react or respond quickly to mitigate any risks.
Gain greater insight into the state of staff satisfaction or engagement by using sentiment analysis in your surveys. This involves mining opinions expressed in open-ended questions and can help you understand:
Analysing open-ended questions can also help you put other answers in context. For instance, of those staff that say they are actively looking for other opportunities, how many are anxious or unhappy about returning to the office? Maybe this is a blocker for them.
Brandwatch is a great tool for keeping tabs on what people are saying about your brand or organisation. You can also check what they’re saying about your competition and stay up to date on the latest trends.
Whatever type of survey you’re running—be it customer satisfaction, market research, brand awareness, or something else entirely—SurveyMonkey’s Sentiment Analysis tool allows you to categorise your open-ended questions. It sorts them into positive, negative and neutral. You can also assign relevant tags to responses and then filter based on tags. Plus you can filter overall results and individual questions by sentiment.
HubSpot’s customer feedback tool organises reviews based on sentiment, displaying them on a dashboard that gives a great overview of customer satisfaction.
Critical Mention monitors and analyses broadcast and digital media, including TV, radio and podcasts for brand mentions.
MonkeyLearn provides a free tool that you can use for real-time monitoring of digital content about your brand, product or service.
If your customers are spread across the globe, then Rosette is probably your best bet. Its sentiment analysis tool works across over 30 different languages, without the text needing to be translated first.