Dispelling the myth of sentiment analysis 

As a metric, sentiment is something that marketers, consulting firms and analysts have touted for years as the key to understanding audience opinion and unlocking success in the digital world.

But what is sentiment? And is sentiment analysis actually worth the effort?

What is sentiment analysis?

Sentiment analysis (or ‘opinion mining’) is the process of extracting the subtext of piece of text – like a review, tweet or an article – using computational linguistics. It’s a way to mine qualitative data to identify and categorise opinions about particular topics, products or companies.

It is not an exact science. Sentiment as a metric alone isn’t a precise measure of opinion. Yet a lot of sentiment analysis is automated now using specialist software, powered by algorithms that are supposedly capable of spotting risk, predicting movements in the stock market, tracking public opinion about brands and more.

Sound too good to be true? It is.

In reality, sentiment analysis is old news and automated tools are little more than glorified natural language processors and text analysis systems. Most of them aren’t capable of understanding the context of conversations or the subtleties of human communication like sarcasm, irony, humour and hyperbole.

Despite this, sentiment analysis is relied upon by both brands and consumers.

But most of the hype around sentiment analysis is just noise created in the marketplace, and it misrepresents how and when sentiment can profitably be used as a metric.

Is sentiment still useful? Yes.

Sentiment analysis can still be a useful exercise. But like any metric, sentiment is only useful when applied to the right data, which means targeting the right conversations.

In general, sentiment analysis is useful when the conversations you’re monitoring are:

  1. Big, in the sense that they have a large number of participants who express many opinions over a long period of time.
  2. Polarised, ideally with two articulated and diametrically opposed opinions (eg conversations in which participants fall into ‘pro’ and ‘anti’ groups).
  3. Prominent, with extensive coverage on online media such as news sites and blogs. This enables machine learning algorithms to capture a more accurate reading of general sentiment, simply because there’s more data (ie text and content) to analyse.

Online conversation about the practice of hydraulic fracturing (or ‘fracking’) is a good example of a conversation that matches these criteria. There are thousands of articles, forums, social media posts and discussions dedicated to the topic, and participants in these dialogues generally form into very clear ‘pro’ or ‘anti’ groups that write in noticeably adversarial tones.

For conversations like these, tracking sentiment over time is a highly useful way to see how opinions shift, react and evolve. Changes to how people talk about a particular topic can be a leading indicator of risk, so it’s worth keeping an eye on.

Sentiment analysis: use it wisely

In a nutshell, sentiment is a good metric for broad, thematic analysis and can help you (roughly) gauge which way the wind is blowing. But be wary of how you use it, target the right conversations and remember to balance automation with good judgement.

Sentiment can help measure the tone of how people write about a topic, but that’s very different from analysing the topics that someone is writing about. This is why we believe in using risk – rather than sentiment – as the primary lens to understand the content and context of online conversations. Risk can tell you so much more – it can identify early warning threats to your supply chain, quantify your reputation and surface potential compliance issues. And you can’t do that with a simple smiley or frowny face.