Record stores may be hard to find, but there are plenty of alternatives for online music downloads. Some of them -- Amazon is a prime example -- have placed great emphasis on fashioning their software to make personal recommendations extraordinarily personal. Much of that was home brewed.
Upstart players like PonoMusic have little choice but to join the recommendation engine competition; still, they can do so on their own terms, building systems based on resources and strategy and buying the software where that makes sense.
"Amazon has done a great job of filtering recommendations up to users," said Randy Leasure, vice president of business development at PonoMusic. "That is what we are looking to do, too."
The route taken to recommendation is a bit different, however, as PonoMusic took a "buy" path to its recommender. Leasure, along with Kevin Fielding, head of engineering, and others, opted for Datameer's big data analytics platform as a means to quickly bring recommendation capabilities to a website that offers high-fidelity music downloads, including ones that run on PonoMusic's own specialized music player.
Fielding said he has done programming work that ranges from low-level firmware development to cascading style sheets. Although he felt he and his crew could have built their own recommender, they chose to use the Datameer combined data preparation and analytics platform to speed their project.
Mission to recommend
In this case, the objective of the recommendation system is to make appropriate music suggestions for audiophiles looking for sound fidelity superior to that found in the ubiquitous MP3 format.
That is an important part of an overall mission of a company started by Rock and Roll Hall of Famer Neil Young to enable listeners to hear albums as original artists may have heard them during studio recording sessions. Another part of the effort is a $399 music player dedicated to playing music with 24-bit depth of sound.
An early step is to use the Datameer platform to improve business analysts' views into customer preferences. Produced by Datameer Inc., the software runs natively on Hadoop, includes recommendation algorithms and supports an interface into Excel. That interface allows line of business users to connect user website activities to data from Salesforce back-end databases.
"This saved us as much as six weeks of building our own service, maybe even more if you consider that the [Datameer-provided] algorithms for recommendation were mature. As a result, there was less testing required," Fielding said.
He added that Datameer support engineers' experience with variety recommendation engine implementations gave PonoMusic a leg up on the process. That is, their experience with different scenarios for handling web user sessions helped cut time to production.
Fielding said recommendation results so far "have been spot on." He estimated that the system's recommendations do as well as manually curated collections that PonoMusic workers have produced.
For now, the site's recommendations are refreshed daily. Among next steps is a move to improve the site's approach to sales, with personalized messaging to users. "We are looking to do deeper marketing via email," said Leasure, a music industry veteran who hopes the recommendation software can take on the role of the knowledgeable sales clerk in a record store from the distant past.
"For us, the ultimate goal is to bring the store and the user together and to make the discovery experience as easy as possible," he said. "The goal of the software is to enhance the site in order to get people excited about coming back, creating deeper, better customer engagement.
Datameer software's use of Hadoop, he indicated, positions PonoMusic for growth, because the company eventually plans to offer ''millions of SKUs" -- or stock keeping units -- that being the parlance for describing downloadable digital songs.
(Editor's note: PonoMusic suspended purchases from its music store on July 20, likely for several weeks. The company said it's transitioning to a new digital music delivery platform because its previous contract was canceled after a sale of the platform provider's technology and assets.)
Learn about the role of recommendation engines in 'the second machine age'