The Internet is mostly noise, with small kernels of insight thrown into the technological soup. There are the rare Tweets and blog posts that transform a fact into an idea, that turn a random concept into something meaningful and personal. It’s the same commentary that is so valuable for advertisers, so rare for readers, and so hard to find. The biggest problem, though, is that it is almost impossible to teach a computer the difference between valuable content and content not worth its space on the hard drive. It’s even harder when trying to compute it on a personal and individual level.
But big problems don’t scare everyone away, and Prismatic thinks it has finally cracked the nut in a way that Pulse, Twitter’s Discover service, and news aggregators like Google News have failed to in the past. I’m inclined to agree.
Launching to the public today with $1.2 million in funding from Battery Ventures, Prismatic is looking to be the one-stop source for a single feed of personalized, relevant news. No more need to sift through the gigatons of data that sweep by us everyday.
When the user first signs up for the service, they’re asked to authenticate themselves using Twitter, Facebook, or Google Reader. Once signed up, Prismatic begins to crawl the user’s social data and analyzes it. Using the aggregated data and the information it gleans from machine learning techniques, Prismatic then begins to tailor recommendations for news articles, Tweets, blog posts, and insightful commentary.
The company also provides users with incredibly specific interest categories that tell the service what to focus on. For example, users can subscribe to everything from Technology to US Politics to local area news to recipes. It covers almost anything and everything you’d want to read about, and it does it accurately.
In fact, the service is so accurate and learns to generate categories so quickly that it sometimes gets away from the company. According to company cofounder Bradford Cross, one user wrote in to the company explaining how he loved the recipes section of Prismatic. Cross wasn’t even aware that the category was being used or existed, but he quickly marked it as one of his interests. Now, the recipe section is getting more attention, and the algorithm is learning even more from the additional users.
The real meat of Prismatic, though, is how it decides what is and isn’t relevant for users. There’s a fine balance for the algorithm, with many false flags that other services consider as evidence of importance. For example, a story could be Tweeted 10,000 times, but that doesn’t mean it is an amazing story or that someone specifically should read it. What does matter, though, is if a story has real commentary in a few hundred Tweets, the commentary was insightful, and the commentary ties it in with the user’s interests.
If this sounds like a hard problem to solve, that’s because it is. The company has been hard at work building Prismatic, through ups and downs over time. The company makes sure that it only hires the types of people who can really deal with any number of problems, and swfitly fires people who aren’t up to the challenge. It may sound harsh, but it also means that the small team is laser-focused on building Prismatic to be the company that it needs and wants to be.
For the past few years, Prismatic has been slowly tweaking its algorithms and collecting as much data as it can. In fact, the company has acquired so much data and continues to do so, that Cross expects the company to have more data on its servers by the end of the year than Twitter does, if all goes according to plan.
What’s all of this data for? As Cross explains it, Prismatic copies some of the backend functionality of search engines. As a replacement for search engines like Google News, the company needs as wide of a data set as it can get its hands on. This means collecting data, crawling the Web, and pulling information from public APIs. With this information already analyzed and categorized, when a user does finally join Prismatic, the chances are that the data is already there, waiting for you to utilize it.
The data has reached such a large volume though, that it creates a problem of noise. After all, is a Tweet more valuable than a +1, and where does a “Like” land when compared to an auto-share on the timeline? As Cross explained it to me, the company gets more accurate and weighs the sharing of articles differently for different people. As Cross says, “It all depends on your social network and social interactions, your topic and publisher preferences that we learn over time, and any other signals we gather.”
It may seem like holding onto a cache of such a large data set is an unneccessary infrastructural burden, but whatever the costs of the service, users seem to be liking the end result. The company has already shared one million articles, and 25 percent of Prismatic’s users come back to the site more than 5 times per week. For a private, invite-only, under the radar startup that is entering a dense and competitive market, that’s a big deal.
All of this data, growth and engagement is good enough for the company to tout on its own, but the fact is that as more people use the service, the service itself improves. It’s a healthy cycle, and one that ends up benefitting all users. This means that even though the service is good now, it will be better by orders of magnitude in only a few months.
With such a firm grasp on the recommendation analysis system, Prismatic is already looking to the future of media and how it can impact it. When Cross initially began working on the idea after he sold his previous startup, FlightCaster, he was thinking about what company he would want to work at next. Cross wanted to work on such a big problem, that he wouldn’t solve it in his lifetime. By all appearances, Cross is taking on a gigantic problem: media and investigative journalism.
Cross believes that investigative journalism is at the point where it needs to seriously begin taking advantage of technological tools. Not in an effort to replace writers with robots or algorithms, but more along the lines of helping them overcome mountains of data to make stories even better. For the future, Cross has a rough plan to create a “Newsroom 2.0”, that would be made up of half experienced reporters and half of statisticians, all working on large, important exposes. It’s risky, but Cross says that the feedback he’s gotten from the media industry — including heavyweights like CBS, News Corp. and the Washington Post — indicates that the larger players want to see how the strategy plays out.
At this point in time, though, Prismatic can only be seen as a social discovery service. One that hasn’t even released its iPhone application (more on that later this summer). If it can’t deliver timely, relevant, and interesting articles to people substantially better than older services can, then it will be dead in the water regardless of the technology and effort behind it.
That being said, I wouldn’t bet against the company, because I’ve already found plenty of stories I wouldn’t have found otherwise. And that’s saying something, considering that my job is finding interesting stories.
[Illustration by Hallie Bateman]