Sentiment analysis is one of the most important media metrics needed to measure and protect corporate reputation; it has risen to the top of the must-tackle metrics list for communications professionals.

Yes, sentiment has always been critical to understanding how influencers, stakeholders and consumers think about your brand. But the actual number of positive and negative mentions is a tough nut to crack in a social media landscape where data changes at hyper-speed. This makes it extremely difficult to find insights about the reasons for the favorable and unfavorable comments about your brand.

This results in a massive volume of mentions, which means that human coding for sentiment is too time-consuming and costly. And the accuracy of automated sentiment for an evolving, complex language has always been a contentious topic.

But that too is changing.

Analytics platforms such as Zignal Labs are investing in innovative sentiment solutions, introducing new technologies to increase the accuracy of automated sentiment.

Determining the Accuracy of Automated Sentiment

As mentioned, until recently there were two approaches to evaluate sentiment – hand-coding to detect sarcasm and other tricky language issues or automated sentiment, which typically has low accuracy rates. Yes, many analytics platforms claim 60% to 90% sentiment accuracy, but either can’t or won’t demonstrate the proof to support it.

A more current advance is a combination of artificial intelligence and machine learning – which can help you to process large quantities of data, and comprehend complicated language such as sarcasm, irony, slang, double-positives, industry jargon, new phrases and trends, and misspellings.

For example, think about words such as brilliant, which could be genuine or sarcastic, or funky which can be negative or positive, depending on the context. Words like thin can have more than one meaning – a thin smartphone is positive, but thin walls in an apartment is not. And think about how this mixed-sentiment sentence might be analyzed: “The hotel service isn’t great but we are still happy to be on vacation.”

Machine learning provides computers with the ability to learn from previous results, without being explicitly programmed. Of note, machine learning can also be trained to understand sentiment as language evolves. With that in mind, Zignal Labs developed a new solution that is improving the way that you can track and understand the sentiment about your brand.

Differentiating Consumer Opinion

Zignal Labs has introduced a patent-pending technology for all media channels, based on 1) a machine-learning technique called “deep learning” that analyzes text; and 2) a system for quality assurance. Notably, this solution can demonstrate results that are 75% accurate – in other words, this level of accuracy can be proven; there is no “secret sauce” that is hidden behind a claim.

In addition, this new model provides functionality that allows you to apply filters to widgets, so you can determine the sentiment of different audiences on a given topic. This further increases your ability to evaluate sentiment and gather findings that give you insights to inform your future strategy and build and protect your brand reputation.

This deep-learning technique began with machine training, which came from a group of experts (human analysts) who reviewed a set of stories and coded for sentiment on a 3-point scale (positive, neutral and negative). Those values were then used to train the deep-learning model, which is continuously taught to adapt as conversations and languages evolve.

Unlike most automated sentiment, this system does not look for individual sentiment-bearing words such as good, bad, satisfied or angry; rather it looks at the statistical similarity of the story’s overall language in comparison to the training. You will find that deep learning:

  • Enables sentiment analysis to pick up on slang, idioms and industry jargon
  • Is constantly trained by experts’ input
  • Can be continuously measured and demonstrated

The Zignal Labs sentiment model is available in English, French, Portuguese, Spanish, Italian, German, Arabic and Russia.

Summary

Sentiment in the media is harder to figure out now than ever, with data coming in at lightning speed, coupled with our evolving and complex language that can be ambiguous and have multiple layers of meanings.

The Zignal Labs new patent-pending sentiment analysis solution, with demonstrable 75% accuracy, provides a context to online opinions about your brand, which helps you adjust your communications plan to further protect your company’s reputation.

Learn more about how to protect your company’s reputation in our eBook: Building and Protecting the Health of Today’s Purpose-Driven Brands.