Optimizing Websites With Machine Learning: BloomReach CEO Raj De Datta (Part 5)

Sramana: Would you give me a use case of how a company would employ BloomReach?

Raj De Datta: Search is by far the most-used method to find content. People put a query into a search engine, which in turn presents information to users, who click on links to visit. From the perspective of the website owner, they need to realize there are millions of ways people are looking for things. They need to know how to structure their content so that it increases the probability that users find their content.

Let’s assume our customer has 1,000 products. Our customers have to determine how to present those 1,000 products on their website. How many pages does it take? If you lay it out ten different ways, you actually create a candidate set of 10 to the 28th power of phrases. If each of those products can be described in 10 different ways, you have now created 10 to the 30th power different phrases, which must be optimized for. That is not a problem that a human can optimize. Our system will identify the minimum set of relative content that must be augmented, or have new pages created for, which maximizes the probability of the content being found via web search.

Now move to social media. Users discover items via social media using news feeds. If you look at traffic on websites, you will see an increasing proportion of traffic coming from social media. Of course, it is industry specific. Content sites and entertainment sites get a lot, while commerce sites get very little. Sharing by the news feed, or Tweeting is a new method of discovery. The interesting thing with social media is that most web content is not a good structure for social discovery – look at the number of pages that have a reasonable number of likes  The only way that Facebook knows a page exists is if somebody has liked it and shared it. Yet only a portion of the pages on the Internet address that.

We create persistent experiences on the site. A lot of our customers are e-commerce websites. They want to know why people bought their products. A customer may have bought a table, but what they are really working on is redecorating their living room. We enable users, as they are browsing a site, to group products into an experience. That group will be a persistent experience that lives on the site. When a new user comes to the site, he or she can find an existing experience and can follow, share or modify it.

Sramana: What does that look like from the consumer side?

Raj De Datta: It is a pitch. It is a dynamically created collection of objects and information that is amalgamated around a feed. That becomes a living node on the website. A week later, if a new item comes in that new object can be propagated to the customers Facebook page the user is following. This marries two of the most important graphs on the Internet, the behavioral graph and the social graph. The social graph is easy to understand, while the behavioral graph is more complex. When I am looking at an item on a website, most websites have lost the knowledge of what the previous users were looking at and why. We aim to transfer knowledge between the thousands of people who viewed, browsed, or looked at products online.

Sramana: You are dynamically generating pages based on the browsing behavior of the user?

Raj De Datta: Yes, and then we share those pages with people’s social graphs and with search engines. Provided they are good quality, we will make sure they are pages that can be indexed.

This segment is part 5 in the series : Optimizing Websites With Machine Learning: BloomReach CEO Raj De Datta
1 2 3 4 5 6 7