Best Buy Adopts RDF and the Open Web
May 3, 2010
I had a chance to talk with Jay Myers of Best Buy the other day. Jay works on the retailer’s web site(s), and he’s one of the people on the front lines of the semantic web. Jay has two interesting problems: he wants people to find product information much easier than they do today, and he wants the Best Buy stores to have their inventory found more easily, to make a tighter connection between buyers and sellers.
Let’s look at the second problem first. Best Buy has a certain number of products that are returned to the stores, and the stores’ goal is to sell them on a special “open box” shelf at reduced prices. Each store has its own inventory that is trying to find a buyer at a specific price. Jay could have done what everyone else does – make his own API and publish his own data feed online using keywords and his own version of XML, but he wanted a more scalable solution. He’s been working with Martin Hepp, whose GoodRelations product ontology helps create linked offers that search engines can see easily, without any fancy tricks.
Goodrelations is, at the moment, a container for describing offers using RDF, the language of linked data. It lets you describe a product, its location, the hours when the location is open, and the price of the item. It doesn’t have descriptive fields for each kind of item, but we all expect that a group effort will fill that gap eventually (more on that in a minute). This summary from the BestBuy.com/GoodRelations gives the figures:
BestBuy.com has just started to serve a complete RDF/XML dump of their products and price information to the Web of Linked Data, using the GoodRelations vocabulary for e-commerce. The data dump is updated on a daily basis and contains detailed descriptions for roughly 450,000 individual items. With about 60 triples per item, this totals to about 27 million RDF triples.
Best Buy uses RDFa and GoodRelations to increase local open box products and store location visibility on the web. All Best Buy store managers have the ability to blog, update store location data, and generate their own open box offers through custom WordPress plugins which generate RDFa for location data and open box product data, right in the HTML. This makes the data available for both humans to read and machines to utilize. Myers says, “We’re not using search engine tricks, we’re using genuine information. It’s easy to sell the top 20 products using keywords, but surfacing the long tail is where the data makes the difference.”
After publishing semantic store data, Best Buy started to see more accurate search results, practically immediately. Myers says, “We saw a pretty big SEO jump when we started using semantic descriptions, way more than we could get using clever keyword strategies.” As for open box product SEO, he isn’t ready to publish his figures yet, as the project is still being rolled out to all stores. He’s gathering data and hopes to have a report at the SemTech conference in June (where both Jay and I will be speaking).
Second problem: how to provide product descriptions in the way I talked about in chapter 7 of my book? It just makes sense that we should have a single reference for each product that everyone can link to, rather than copying and Balkanizing product descriptions across thousands of different web sites. At the moment, Jay and many others support Freebase.com, which was designed to be a central repository of just such information. In my opinion, I think we’re going to need something even more heavy duty, geared at products (I’ll talk more about Freebase another time), but for now, putting product descriptors into Freebase where everyone can get them is a lot better than using the Amazon.com API. And that’s really what Jay and I both want: that people do product research by searching an open web of products, rather than having to go to Amazon.com, cnet, epinions, and a thousand other web sites. Jay believes we’re all better off if we cooperate than if we try to compete on the basic descriptions of products. Better to share the descriptions and then compete on price, availability, service, etc.
I’ve been saying for a long time that we’ll need product descriptors for every kind of product, from polo mallets to computer mice to washing machines to truck transmissions. Eventually, we’ll have a WikiPedia-like tool where we can start aggregating all that data. At the moment, we’re getting by with Freebase and some XML that isn’t really shared among all competitors. Ideally, this huge world of product descriptors would be part of GoodRelations, and Martin Hepp has some people working on a segment of it for consumer electronics. But in the end, as long as everything is linked, it doesn’t matter if the descriptors and the offers reside on the same server or are scattered all across the web.
If you work at Amazon.com, your job would be to resist this open-data movement, because you keep more control by maintaining your own standards. But eventually, the rest of the world wants open data, and the rest of the world is much larger than Amazon.com. I expect that within five years we’ll see several very active communities of people building the kinds of open repositories I talk about in my book, and that it will take another ten years to get to the point where 90% of the world’s products are described once, in a single place, and everyone refers to them with links, rather than copying and introducing errors. I give Jay a lot of credit for wanting to work with his competitors to make these open resources possible. As Jay puts it: “I get excited about better defining products. I got pretty good at making a web page people could read to learn about a product. Now I want that same page to be readable by any software that can find it using rich data, and I want those descriptions to be findable by any search engine that understands RDFa, Microformats or Microdata. Let’s give search engines better access to descriptions, to make products more visible.”
Jay goes on to describe his hopes for the industry: “At the moment, the product web is lacking substance. There’s no reason that semantic product descriptions should be proprietary. We want to develop ontologies that are open and universal that everyone can use. That’s one of the challenges – anyone can develop an ontology, but it’s better to share them.
I asked Jay what his biggest challenge was, and he said getting business people to understand something new. That, he said, is where my book has come in very useful. I hope to report on more excellent progress from Jay in the next few years.