Our in-house data scientist, Fletcher Stump Smith, spends his days building machine learning models that identify news articles relevant to AirPR customers and predict the business impact of those articles. In layman’s terms, Fletcher’s work allows us to solve a pretty itchy problem: the problem of proper PR measurement.
Recently, Fletcher gave a presentation about what data science looks like at AirPR so our team could get a greater window into how his expertise makes our PR analytics software possible.
Why share it with you? Great question. Gaining a stronger understanding of the data science behind PRTech can show you how PR data is harnessed to help you achieve your public relations and communications goals. And when you have a grasp on the science behind the tech, you can better explain issues tied to PR measurement to your C-suite leaders.
Let’s break it down.
Articles in their raw form aren’t measurable. That said, we rely on technology to identify, categorize, and analyze entities found in a given document (or piece of content). Let’s look at a slice of an article from our blog to illustrate what a computer sees:
After entities within an article are recognized, our technology identifies constituent parts of each sentence and determines how the words relate to one another. Those entities are then categorized and labeled.
If a piece of content on the web includes the sentence “AirPR is a PRTech company that provides analytics, insights, and measurement solutions to the evolving PR industry,” our tech will capture relationships between words (indicated by arrows) and identify parts of speech (signified by colored abbreviations).
From this analysis, we can see that AirPR is a proper noun and a nominal subject, both of which can be indicative of salience. In other words, we can identify which entities are most prominent or relevant to a piece of content. You could liken it to how “Sir Mix-a-Lot” in the sentence “Sir Mix-a-Lot likes sunny days” is more important than “sunny days” in terms of relevancy. Other features that can be identified within an article include word count and paragraph location.
Cool beans. What does that mean for PR?
Say you’re an in-house PR manager for a big technology company like Apple that’s mentioned nearly every day in the news. If you manually read through Google Alerts every time your company is mentioned, you’d be swimming in email alerts for decades. A lot of these mentions would be incidental and not necessarily worth tracking from a PR perspective.
What you really want to know is when and where notable articles about product launches and other company news publish. Being able to distinguish between irrelevant mentions and relevant news is invaluable — and it starts with PRTech.
Thanks, Fletcher, for the window into your work!
Live in San Francisco? Be our guest at tonight’s Salesforce panel: “PR Upgraded.”