Mobile data sourced straight from the network gives a behind-the-screen look into consumers’ brand affinities and behaviors, showing app installs, uninstalls, mobile web visits and persistent location data. This uncovers true interests and behaviors, revealing who consumers are as individuals and offering the most comprehensive view of a brand’s ideal audience.

But in the world of real-time bidding and programmatic buying, there’s a need to better match user behaviors with bid requests to reach the right user in real-time. While app installs are a great starting point to measure behaviors, the crowded marketplace makes it increasingly difficult to quickly associate relevant descriptions with individual interests and behaviors.

In an industry where every millisecond counts, our data science team developed an algorithm to model app descriptions based on topics. We analyzed the 23,000 most used apps from Google Play and found 65,000 words associated with these apps. That’s almost 40 percent of the total number of words in the English language.

We then narrowed words down to nouns and simplified them to a root word. This stemming process looked a little something like this: Merry, Merrier, Merriest is shortened to Merri. Brave, Braver, Bravest is stemmed to Brav. This process, called topic modeling, assigns a distribution of topics based on the app descriptions, which uncovered a mixture of categories per app, giving a deeper understanding of what an app “is about.”

For example, common words associated with the topic of food are bread, bake, pasta, and salad. But we also noticed the word “photo” appear. While this word on its own does not seem relevant to what this particular app “is about,” its frequent appearance with the words mentioned above, as well as words like “recip” (the stemmed version of recipe) and “chocol” (likewise for chocolate) start to paint this user’s interests as food; thus, identifying a foodie.

Topic modeling analysis gleans insights into individual interests beyond publicly-available app descriptions, serving as key algorithm inputs to better match individual user behaviors with bid requests. It also gives brands an opportunity to expand audiences based on interests, making behavioral targeting more accurate and relevant than before.

Each device is as unique as the fingerprint of the person who uses it. We get behind the lock screen to understand the various life stage of consumers and their intent to engage. This discovers your best audience, identifying the most meaningful way to connect with them.