ML Recommendation Model

TLDR

Built a content recommendation model for Dolby.com that combined user navigation behavior with article category similarity to surface relevant content, achieving a 66% success rate during testing. Each article page now includes three article recommendations at the bottom.

Context

Dolby.com had very little personalization built into its content experience, which meant users were often left to navigate articles without guidance toward related material. I also wanted to introduce more data science practices to the Dolby.com team, because the team had data they were collecting but were not always using it fully.

I set out to build a model that could analyze how users moved from one article to another and combine that behavior with category similarity to generate smarter suggestions, turning underutilized data into a tangible product improvement.

Process

I started by creating a brief outlining the KPI goals and the methodology so the team had full visibility into the approach before any code was written. I then ran tests on the data to confirm there were enough meaningful patterns to support a model, which there were. Using a Multinomial Logistic Regression model, I analyzed patterns in how visitors moved between articles and combined those navigation signals with article category similarity to generate ranked recommendations. After training and tuning the model, I implemented the recommendations at the bottom of each article to begin measuring real-world engagement.

Outcome

The model achieved a success rate of approximately 66% during testing, correctly predicting a relevant next article roughly two out of three times. I pitched both the creation and the results of this project to my team, and the recommendations were subsequently deployed. Each article page on Dolby.com now includes three article recommendations at the bottom, improving content discovery and increasing time on site for users who engage with the suggestions.