How Pandora Knows What You Want To Hear Next

Have you ever noticed that, after 6 p.m. on weekdays, you tend to listen to harmony-laden, lo-fi, guitar-based songs with medium-to-fast-paced rhythms and a strong backbeat -- but you'll skip ones that are too distorted? As opposed to weekend mornings, when you follow up a local news podcast with slower piano tracks sung by a solo female vocalist, with strings and horns, angular melodies, multiple sections (but no solos) and a touch of melancholy throughout? Chances are, you've never thought about your listening choices in such a detailed way. But Pandora's musicologists and scientists have, and that's how -- with the help of artificial intelligence, machine learning and the analysis of the listening habits of its more than 65 million monthly users -- it knows which song you'll want to hear next. "We treat every individual very specially, and focus on contextual recommendations to understand what you like, what you listen to," says Oscar Celma, Pandora's vice president of data science, of how the company maps the DNA of every piece of audio in Pandora's millions-wide song library and compares that with explicit and implicit user preference feedback to yield bespoke programming.