With the rise of artificial intelligence and ensuing hype, many companies in the media intelligence industry and beyond began touting their use of AI. But the story often stops there without further explanation.
Communicators don’t have to be data scientists, but it is worth asking your media intelligence provider how they employ AI. In the world of textual media analytics, there are best practices as in any other industry. If your provider is not following them, it could have serious consequences for the accuracy of your communications data.
Media Analytics Best Practices
1. Use Ongoing Supervised Machine Learning
Cultural conversation changes quickly. The meaning and connotation of words are situational and evolve over time. This is why several studies have found artificial intelligence employed in the media analytics space must be supervised. One study from communications experts at the University of Wisconsin-Madison and the University of Georgia found, “the combination of computational processing power with human intelligence ensures high levels of reliability and validity for the analysis of latent content.” Another from researchers at the University of California at Berkeley and Northwestern University found that unsupervised machine learning, “does not perform well in picking up themes that may be buried within discussions of different topics” and therefore missed several mentions of the topic they were tracking of economic inequality. The concept of inequality, whether in the economy or in the workplace, might very well be something a communicator would want to track – and certainly other nebulous concepts like it.
Computers can improve at processing language, but they need to be told what’s right and wrong. A computer cannot tell when the use of sarcasm in an article contradicts the normal sentiment of a word it has already learned to label positive, so it will continue incorrectly analyzing your content until it is corrected.
That’s why media intelligence providers cannot take a “set it and forget it” approach to AI. A constant feedback loop is required to educate the computer in the nuances of language. If the data set remains static, it will make your analysis inaccurate and irrelevant.
2. Target Analysis Specifically to Your Company and Your Perspective
The most accurate communications analysis comes from ongoing supervised machine learning targeted specifically to your business. Every organization has different goals, challenges, and perspectives on the world. Two companies can read the same news article or social post and analyze it completely differently based on their point of view. A solar energy company and electric utility company would categorize and tone the same article about energy regulation very differently. If you use the same data set across clients, you run into the same problem again in that the computer will continue analyzing content as it originally learned, not accounting for the context of what a specific organization cares about.
At PublicRelay, we perform client specific media analysis leveraging ongoing supervised machine learning to ensure that our clients are getting the most accurate data possible. This accurate, contextual analysis tailored to their business goals enables them to not only understand what they’ve done, but yields insights that tell them what to do next.