At Boomtrain, we don’t believe in a one-size-fits-all approach to predictive modeling. Your content and audience are unique, and it’s a best practice to build a custom model for each client. This leads to better predictive accuracy, which results in better engagement.


News Media Case Study

For a news media organization, we implemented a predictive algorithm tailored for a 24-hour news cycle. In this case, we predict the most popular article based on what we observe about each new piece of content within the first 3 hours after it was published. As part of our research, we proposed a dozen different factors that might contribute to predicting article popularity:

  • Published timestamp
  • Seasonality (time of day, week, month, year)
  • Early engagement
    • 3 hour cumulative page views
    • % of page views received in the first hour after publication
    • % of page views received in the second hour after publication

Next, we validate which of the proposed features provide predictive power. To answer that question, we turned to a methodology that is designed find all the relevant features from the candidate set.

At a high level, the method builds a predictive model with the entire candidate feature set plus new features that are built by randomly shuffling each of the initial features. The idea is that we can compare (via statistical test) the effectiveness of the real features to the random permutation features. Any feature that is not significantly more predictive vs. the control can be safely discarded because it is not helping drive accuracy.



In this case, we were able to remove some irrelevant features, such as month-of-year, which let us build a more accurate model.

Remember, these results were tailored for a specific industry vertical. The most predictive feature of other types of content (e.g. long-form topical content) can and will be very different. For example, extracting features from the article tile and topics from the full-text may be better for other industries. The point is that there is no one-size fits all approach that. Think of Boomtrain as your data tailor who can help build bespoke models for your business.


Want to incorporate predictive models into your marketing strategy? Learn more about Boomtrain Marketing Engine here.

Share This