Machine Learning Models for PR

If you understand AI as essentially “pattern recognition” of the highest order, you must ask, what can I do with that capability in communications? Well, if you’re doing PR at a startup, you could use AI’s pattern recognition to predict the success of a press release or a marketing campaign based on past pickup of your news or contributed articles. But before doing that it’s important to understand a few concepts about how AI “learns” or is “trained” to recognize patterns in your old PR content and then make recommendations based on those patterns.

A basic understanding of four Machine Learning models (methods for analyzing your old content) can be useful, especially if you’re working with data scientists to tease out a repeatable way to analyze your past PR content and then plan your future content. The core concept is Machine Learning, which is the use of algorithms (a set of software instructions used to analyze your data) to find a way to transform input data (past PR content) into predicted outcomes (future PR content). There are four basic Machine Learning models:

1. Logistic Regression (video): A “classification” model often used for binary outcomes (e.g., predicting if a press release will get picked up by a list of your most important media outlets). Logistic Regression has a long history in the field of statistics, and it remains a staple in data science for predicting probabilities. “What is the probability my news release will get picked up by X, Y, Z media outlets as it is written now? What are the probabilities of pickup if I change the text/images to version 2 of a news release?”

2. Decision Trees (video): This is a Machine Learning model that splits data based on key variables, creating a tree structure to make predictions. For example, it may predict media pickup based on news topics and release timing in the corporate earnings news cycle. The predictability splits are chosen to maximize predictive accuracy.

3. Random Forests (video): This is an “ensemble method” combining multiple decision

trees to improve accuracy. For example, Your press release has three industry-changing news elements (3), and you’ll be announcing the news the day before your major competitor will be holding its Wall Street earnings call. Each tree is trained on different subsets of the data, and the final prediction is based on the majority vote of the trees.

4. Neural Networks (video): Modeled after biological neurons, these networks take inputs

and transform them through layers of neurons to make complex predictions, such as facial recognition. While highly accurate, neural networks can be difficult to interpret.

There are many more types of Machine Learning techniques (explainer video) used today, but the four above are most popularly understood and give a baseline for working with your company’s data scientist. For more about other techniques, look into:

·       Boosting

·       Support Vector Machines (SVM)

·       Neural nets – more complex than neural networks

·       LASSO

·       Ridge

·       Weighted Regression

·       Kernel Regression

Takeaway: Members of the communications team won’t necessarily get into which machine learning models are used to effectively predict which news will hit where, when, and how effectively. However, understanding how your data scientist is creating the “prediction machine” that you are relying on can help immensely when explaining your data-driven opinion on which way to present news, when, how, and where to issue the news.