Understanding the technical side of employee sentiment analysis

While lots of HR leaders don’t have time to explore the intricacies of how AI such as NLP works, we also know you might be one who wants to learn more about it. In this brief article, you’ll find out about:

• The technology behind employee sentiment analysis
• Why qualitative data is so important
• How Natural Language Processing interprets the data
• Knowing when you’re seeing a trend

This article works on the basis you already know what employee sentiment analysis is, and why you might want to implement it in your business. For a detailed introduction to employee sentiment analysis, you should take a read of how to measure employee sentiment and why you should do it.

Now, where were we…

What is Natural Language Processing (NLP) and why do you use it?

Employee sentiment analysis, very simply, interprets the data people write and say and works out how they’re feeling. To do that, it uses a machine learning algorithm called Natural Language Processing, NLP (This is not the same as Neuro-Linguistic Programming, which is sometimes used by coaches to aid personal development and self-awareness).

NLP technology has a ton of applications and has developed significantly over the last few years, including contributing to the development of AI programs such as Microsoft Co-Pilot. If you’ve ever used a search engine or voice-to-text translation, for example, you’ve experienced NLP systems in action.

The thing is, the human language is full of idiosyncrasies. English is especially guilty. Between homophones, double negatives, colloquialisms and more, it’s a lot to deal with. But that’s just the tip of the iceberg.

NLP really shines at identifying how our views and experiences unconsciously inform our language. And that’s how it helps with measuring employee sentiment.

Qualitative versus quantitative data

In traditional employee engagement surveys or performance reviews, many of the answers are numerical. The results can be converted easily into an average or scale of some sort, and leave you with a figure you can quote as a statistic. But that isn’t where the richness comes from. The really good analysis is found in the qualitative responses. The written answers which tell you how people really feel.

For example, you may have asked employees to rate their work environment out of ten. Someone might rate their workplace a seven, and that’s all the context you get. You’ve no idea why they knocked off those three points, or even what they like about their office to cause it to be a seven.

You also don’t know whether seven is good or bad.

In theory, an “average” rating would be five out of ten. But a lot of people consider that to be a bad score. When was the last time you got excited about a film or video game where five meant “good”. Very often, sevens and eights are the average. Anything less is a bad review.

Interpreting the data in employee sentiment analysis

When you use qualitative questions, you can analyze the results with NLP. People write responses on their own words, so, even if someone’s being diplomatic, sentiment analysis picks up on their true feelings.

An example of NLP analysis

Let’s say you’re evaluating the need for hybrid setups, so you’re asking people about their commutes.

4 people have rated the statement “I am productive and focused when I’m in the office” as agree. 2 people have rated the statement “strongly agree” and 4 “disagree” or “strongly disagree”.

Here’s how your results might look:

Traditional statistic: 60% of people agree or strongly agree they’re happy working in the office.

NLP analysis: Many employees would like to have the option to work from home.

So where does the difference appear? The first data point doesn’t take into account the comments people are making (or the pressure they might feel to give a positive rating).

The NLP analysis, however, picks up on the number of people referencing long commutes, poor traffic, and having to rush to leave when picking up the kids. People aren’t likely to write “My commute’s terrible, I hate getting up early and driving to work still tasting the toothpaste, and the kids being in bed before I get home really sucks” – that might be deemed unprofessional.

But sentiment analysis puts all the innocuous statements together and identify a pattern. It flags that some people would benefit from getting to work from home at least some of the time. So you can adjust your proposals to take the views of parents and carers into consideration.

Why NLP is so great for sentiment analysis

Using NLP algorithms, you can also dig deeper into an issue.

One linguistic choice of a phrase or word could be considered a coincidence or someone having an ‘off day’. Let’s be honest, if you were doing the analysis yourself, you’d probably ignore it. But several, especially on a similar topic or in the same part of the business? That makes a pattern.

People might not say they hate your new open-plan office. But, if you’re getting lots of incidental noise complaints, or concerns about lack of privacy, there are only so many causes or office changes which will lead to that response.

Sentiment analysis helps you get to grips with what people think, not what they think you want to hear. And that’s why it’s such a great tool for sentiment analysis.

How we use NLP in Engage365

Engage365 has an in-built NLP engine. It uses machine learning to understand the sentiment of feedback and conversations that form part of the employee check-in. We use one of the largest natural language datasets in the world to ensure maximum accuracy. An independent study in 2019 showed that across a dataset of 15,000 text paragraphs the datasets we use had an overall accuracy of 77% when compared with a human manually assessing the text.

As well as detecting the sentiment of the text within a check-in, Engage365 also looks at the sentiment of the question being asked. This provides a weighted sentiment which is more reflective of the actual feedback provided.

What the sentiment data tells you

There is no good or bad level of sentiment, more that it should be treated as a relative rather than absolute measure. This is because the culture of different organisations leads to different preferences in communication style and often, we see large differences in ‘baseline sentiment’ between organisations that you may not expect from other engagement scores.