Extracting the Right Information from Your Data

If AI at its core is about discovering patterns in large data sets, when building a PR prediction machine – for content generation, thought leadership, or crisis communication - we must figure out which patterns are most valuable to identify.

Starting sinisterly, can we identify the best influencers/reporters to target for producing a hatchet job article/video on a competitor? Why do this? Do people really do this? Yes! All the time. Perhaps we want to cause one of our competitors’ most important customers to delay a key purchase, giving our company a chance to catch up on product development. We want to submit a competitive bid for a big RFP and our CEO asked the PR team to create a delay.

Note: If you are not the kind of person willing to do a hatchet job on a competitor, I understand. But remember your competitors will do it to you. The point is, be prepared for it even if you wouldn’t do it yourself. Using AI, identify “sensationalistic” or unscrupulous influencers before your company “gets got” by your competitors. As the saying goes, “…Be as gentle as a dove, but as wise as a serpent.”

The question is, how can you and your data scientist set up your AI model to figure out a “hatchet job” campaign? Yes, most media databases already promise to perform such tasks. They show who writes what, where, and when, and they’re getting better at predictive analytics. But if you can train your company’s AI to identify patterns unique only to your business/products, it seems wise to keep that to yourself. To build your fully owned sustainable competitive AI advantage, not your data broker’s media product.

So how do you take your company’s historical data and turn it into a trend-spotting predictive content machine?  Start by understanding the most popular data mining techniques:

Clustering Data – Simply put, this means grouping your data into clusters that share similar characteristics. For example, which influencers publish weekly, monthly, or quarterly? Easy peasy. That tells us who to contact right away for our hatchet job piece. Another cluster could be around long-format text vs short format videos. Because we want to own this data, we can also add unique data points if we have them, like individual personality traits like “disagreeable” vs. “agreeable.”  The better we structure and cluster our data, the more accurate our predictions can potentially be.

Association Rule Mining – This technique is about finding common co-occurrences in data. For example, how can we find influencers whose opinions skew negatively overall and who write about new technologies? That information will work well for our hatchet job needs. Different from Data Clustering, in Association Rule Mining, our AI model will identify co-occurrences for us. We won’t have to place content identifiers in our data before asking the AI model to predict which influencers are best for our purposes.

Ultimately these two popular data mining techniques, and many others, are used to find patterns in your data. The next step in the process of data extraction and analysis is predicting future occurrences. This is the fun part, and this is where predictive analytics comes in, which is covered in another post.