What Is Sentiment Analysis?

Sentiment analysis is an AI technique that automatically determines whether a text has a positive, negative or neutral tone. You use it on customer reviews, social media posts, support tickets or survey answers: the model reads the text and labels the underlying attitude, so nobody has to read everything by hand.

The technique falls under Natural Language Processing (NLP), the field concerned with automatically understanding human language.

How does sentiment analysis work?

There are roughly two approaches:

• Lexicon-based: the model uses word lists in which words have a preset sentiment value (for example "fantastic" is positive, "slow" is negative) and adds up or combines those values per sentence or document.
• Machine learning and language models: the model is trained on large amounts of text with known sentiment labels and learns to recognise patterns itself, including nuances such as context and word combinations that a lexicon misses.

Modern sentiment analysis often goes beyond just positive, negative or neutral. With aspect-based sentiment analysis, for example, the model recognises that a review is positive about delivery time but negative about customer service, within the same text.

Why is sentiment analysis useful for SMEs?

Many SMEs receive text from customers every day: reviews, emails, chat messages, social media comments. Reading and judging all of it by hand takes time and doesn't scale to larger volumes.

Sentiment analysis does three things for you:

1. Monitoring customer satisfaction without reading every review or survey by hand.
2. Spotting escalations sooner: a strongly negative message can go to an employee automatically before a complaint grows.
3. Recognising trends over a longer period, for example whether satisfaction with a specific product is dropping.

If you get hundreds of reviews or support tickets a month, automated sentiment analysis can take over a large part of the reading and bring up only the cases that stand out.

A practical example

A web shop collects hundreds of product reviews every month. Instead of reading them one by one, they are analysed automatically for sentiment. Reviews with strongly negative sentiment about delivery time are grouped and shared with the logistics department. Positive reviews about a specific product can feed back into marketing. That gives you insight without anyone having to read all the text.

When does it work, and when not?

Sentiment analysis works well, but it isn't equally suitable everywhere.

Sentiment models make mistakes, especially with irony, sarcasm and industry-specific jargon. A model that wasn't trained on Dutch or sector-specific text often performs worse than you'd expect. So check the output on a sample before you run sentiment analysis fully automatically.

Which terms go with it?

• Natural Language Processing (NLP): the broader field that sentiment analysis belongs to.
• Named entity recognition (NER): recognising names, organisations or locations in text, often combined with sentiment analysis to see who or what the sentiment is about.
• Text classification: sorting text into categories. Sentiment (positive, negative, neutral) is one specific type.

Using sentiment analysis in your business

Want to know how sentiment analysis fits your organisation, for example through an AI agent that analyses customer feedback automatically, or in AI customer service? Talk it through in a conversation about AI consultancy, or run an AI scan to see where text data in your business is going unused.