Public relations and data analytics used to be worlds apart. In the old days, PR measurement consisted of soliciting opinions through surveys and by measuring behaviors based on opinions, like buying or voting. Still, for most variables, data was expensive, limited, lagging or all three.
As a result, measuring public relations or media campaigns relied heavily on leaps of faith. Measurement focused on outputs instead of outcomes, or efforts simply went unmeasured.
Now, text analytics, widely available demographic data and the ability to harvest data from media in near-real time can provide more detailed insights faster than ever before. Big data-processing tools, theoretical and technical advances in text analytics, and better image-processing tools are rapidly changing PR data analysis and how we measure content.
So what are four of the biggest PR measurement trends?
1. Going deeper than positive-negative
In media analysis, we can measure the world as it is in yes/no, positive/neutral/negative sentiments. An audience will support or not support a policy, buy or not buy a product or stock, and vote or not vote for a candidate. But to influence, we need to understand deeper factors.
For example, here are the results from Cision’s social media listening solution for a search about realtors in a geographic area. The terms that are much more likely to be found in positive comments towards realtors are on the left in green, terms common to both are in the middle, and terms more likely to be in negative comments are on the right in red.
There are some counter-intuitive findings, as well as some predictable ones, such as “bad” and “fake” in the positive comments and “closing” and “appraisal” in the negative comments. A closer examination of the articles shows that realtors helped clients avoid some “bad” mistakes and that good “fake” flowers helped in stagings. We also find that first-time buyers wanted more help from their realtors in understanding appraisals and closings.
2. Tremendous progress in text analytics
IBM’s Watson project, as well as Google’s Knowledge Graph, do something that sounds simple but gets remarkable results. They’re creating databases of facts and developing programs that use these facts to analyze texts.
They’re use these databases to answer questions (which is how Google immediately produces the answers to many kinds of queries or how Watson beats humans in trivia contests like Jeopardy).
In text analytics, tools can recognize entities and apply related information. For example, a system might “know” that Steve Jobs was CEO of Apple and Tim Cook is currently the CEO. It also recognizes the relationship “predecessor.” The system would apply that knowledge to an article that identifies “Tim Cook’s predecessor” and tag the article as referencing Jobs, even though the article never explicitly mentions jobs.
However, we’re still quite far from this goal. For example, a human can very rapidly tell the difference between ambiguous meanings, based on subtle clues in the context. For example, if I mention a taste for spicy food at the beginning of a review, and later mention that the restaurant’s Danger Zone Pizza wasn’t hot enough, most humans would understand that I meant “spicy.”
If I said that the pizza came right out of the oven “too hot to eat,” readers understand that “hot” refers to temperature and I’m not complaining, since the pizza could cool down. Or if a human reads “a lot of Hawai’ians like spam,” they will use their existing knowledge of the world, which contains facts like “nobody likes unsolicited junk email” and possibly “Hawai’ians have a high per capita consumption of SPAM®” to understand that the sentence refers to the food rather than to junk email.
When we ran that same sentence through several semantic tools that include entity recognition, most of them identified “spam” as meaning unsolicited email.
While it seems very likely that we’ll reach near-human computer capacity to analyze text, we still need humans.
3. Finely segmented data everywhere
One of the most famous cartoons about the Internet, published in 1993, says, “On the Internet, nobody knows you’re a dog.”
Twenty-plus years later, the Internet has the data to tell, with more and more accuracy, not only whether a user is a dog (albeit a smart one), but the user’s demographics and favorite products.
This gives invaluable data to PR professionals, since they can target communication by demographics, attitudes and interests, and also analyze message effect by audience.
4. Increasing visual media content
Anthropologists more or less agree that writing originated in pictures. Looking at how much we communicate online via photos, visual memes and emoji, we just might be coming full circle.
Whoever first said a picture is worth a thousand words probably wasn’t thinking about metadata, but the information in even a fairly simple image could well take a thousand words or more of metadata to describe.
Social media based on images and video is booming. Some social media images are capable of shaping opinions long after they were posted. According to a 2014 study by Wisemetrics, tweets have a half-life of 24 minutes and Facebook posts have 90 minutes. In other words, after 24 minutes, half of the people who are going to see a largely text-based tweet have seen it. After three hours, a tweet reaches three-quarters of all the viewers who will see it. A Facebook post takes five hours. A 2013 Piqora study found that pins of images have a half life of 3.5 months.
Public and media relations used to be almost entirely based on “soft skills.” Clear and persuasive writing and speaking, professional relationship development, understanding the media, all of these are still necessary. But more than ever, these fields require strong quantitative and analytics skills.
Ann Feeney is the information science specialist at Cision. A version of this article originally appeared on the Cision blog.