Hatched in the same laboratory as Siri, the voice-based assistant acquired by Apple, Trapit is a personalized search and discovery engine that runs on the premise that your love for your friends may not always extend to their taste in reading material.
“Personalization has become nothing more than a buzzword,” said Hank Nothhaft Jr, Chief Product Officer and Co-Founder of Trapit, which has been on the web since November, and unveiled its free iPad app today. “Personalization should be about you, the individual, and your unique tastes and interests, not about what your friends are buzzing about on Facebook and Twitter.”
Compared with other “personalized” news aggregators such as Flipboard, Trapit takes a distinctly antisocial approach to content discovery. Its goal is to capture the rich, long-tail content that is burrowing in hidden niches of the “deep web,” and serve it to you exactly as you like it, based on a thorough, algorithmic understanding your appetite (that we will get to in a moment).
“We don’t care what your friends are talking about, we don’t care what’s buzzing, we don’t care what’s trending today; we care about what you’re interested in,” said Nothhaft. “Look at social networks like Twitter: what you see is groupthink and action, right? Seventy percent of the content shared on Twitter is really from 100 to 200 of the top sources on the web. There is a vast web out there that is just simply what we call ‘undiscovered’ because it is not able to penetrate the echo chamber of Twitter or Facebook.”
“Search is basically corrupted by SEO,” Nothhaft asserted, “so a lot of hobbyists, passionate people and activists are producing great original content with a lot of care, and adding photography and multimedia etc., and a lot of them are never finding an audience. In fact a lot of the content we see shared from Trapit to networks like Twitter, it’s the first time that that article has been tweeted.”
Trapit hunts for this fresh, tasty material within 120,000 sites that it has vetted— it claims to have blacklisted 10,000 unworthy sources— aiming to surprise you with a constant feed of content that is right up your alley.
This involves a de-emphasis on newspaper stories, which Nothhaft considers “ubiquitous.” “You walk out the front door in the morning and get smacked in the face with general news,” Nothhaft said. “What we’re looking for is more long-tail, high-quality, what I call ‘magazine and journal content.’ ”
He explained that this variety of content often does not yet exist in a digital format that is consumable on the web. “We’re trying to make inroads there, and having some success,” he said, adding, without naming names, that Trapit has just signed its first contract with a publisher and has several more on the go. “In terms of the partnerships, we feel that’s where there still is a lot of value to be unlocked.”
But if it prides itself on scouring broad swathes of the deep web and ignores what your friends are doing, how can it possibly “trap” exactly what your heart desires, you may wonder. The answer lies in $200 million worth of artificial intelligence research. Trapit is the second independent company to spin out of the CALO (cognitive assistance that learns and organizes) project- the first was Siri.
Based in Palo Alto, California, Trapit launched its web app in November. At under a year old, it serves nearly 5 million suggestions into user traps per day and captivates users for an average of 16-18 minutes.
The idea behind the iPad version is to “really to nail the lean-back experience” and “bring browsing to the next level.”
How to trap with your bare hands
1. Log on
You can accomplish this crucial first step using your Twitter or Facebook account (to avoid filling in another form; your contact lists will be safe and nothing will be posted without your explicit consent, Nothhaft assured us), or with an email address if you prefer.
2. Create “traps” based on keywords or URLs
If you use a keyword or series of keywords, Trapit will hunt and gather between 30 and 100 documents from the past 30 days from its vetted sources that contain these word(s), and deliver the spoils to your app.
How? “There’s a pretty complex cocktail involved, but we do something called query expansion. So we make a concerted attempt to understand what that keyword really means— we’re not just doing a simple analysis of in what quantity does that word appear in the article you’re looking at— we actually go and look at some reference material. We might look at the Wikipedia page for that keyword, in an effort to really understand what is this keyword really about,” Nothhaft said.
What if you search with a URL? “Let’s say you’re reading an interesting story in The New York Times and you want to follow that topic as it evolves for that storyline,” explains Notthaft. “Trapit will go visit that site and use semantic technology and natural language processing [to] …extract the key text from the article, and really break it down and understand the DNA of the article – what is it about.”
It will then build a trap for you, which constantly checks the approved sources as they publish new content and traps relevant articles, delivering them to you.
3. Window shop
Glide through the “carousel” of content in your traps (displaying headlines, sources, timestamps, images and summaries), and when something catches your eye…
4. Tap to “drop into the trap”
This will load the original webpage, including ads and links, right in the application. The app is very “publisher-friendly,” Nothhaft said; “they get the impressions and all the metrics related to this visit, and Trapit is the referrer.”
4. Share or save
You can share an article via Twitter, Facebook, email, etc. allowing the app to populate the message and then modifying it to your heart's content, or save it in the embedded reading list. (*Note: on the web version, copying a URL to share outside of social media has proven difficult; apparently the iPad version allows you to copy the shortened link directly. Also, expect each Tweet to be followed by @Trapit)
5. Train your app to get to know you better
You do this by rating the stories it serves you using thumbs up and thumbs down icons.
“These are not social proclamations, so when I touch thumbs up I am not telling my friends that I like this article; I am actually teaching Trapit. Trapit is adaptive and its able to learn, and we do this on a user-by-user basis,” Nothhaft clarified. “And what we do with that information is we take the complete text from the articles [you thumbs up] and we break it down to understand what are the words in this article, how many times to they occur, what is the density…” They even have an algorithm that tries to determine the quality of the writing (something that we would be very curious to experiment with).
“In the end what we really do is we determine what those articles are about, and how they are different from everything else that we’ve shown you,” he said. From that, they build a “rich statistical model” for each user that is akin to an “interest graph.”
“We use that to judge every new article that comes in from our sources, and determine, is that article relevant for that topic for you? And as you continue to look at new content, rating it thumbs up or thumbs down, we adjust that model in real time,” Nothhaft said.
Assuming one were to use Trapit regularly over an extended period, this could amount to a powerful heap of personal user data. Asked whether that information might ever be shared with publishers, Nothhaft is adamant: “No. I don’t ever foresee sharing user data on that level.”
How about on any level? “I don’t think we would share this information with publishers. If we were to bring advertising, [which is] not a high priority, but if we were to begin placing advertising in traps we would use the data to place ad content that is highly relevant to a topic area, but we wouldn’t do that by sharing user data with advertisers; we would do it internally and place those ads ourselves,” he said.
“We definitely see ourselves as the arbiters. If we’re distributing publishers’ content etc. inside the application we will maintain control of the user data; no need to share that with publishers,” Nothhaft said. “We take privacy very seriously.”
Advertising is fourth on Trapit’s list of revenue priorities, after licensing the personalization technology to media and publishing companies, doing partnerships where they sell access to premium content, and selling an enhanced version of the app to “those 5-10 percent of users who are looking for more and are willing to pay for it” such as “power users, social media mavens, brands who are looking for more features.”
The company may not be trapping any revenue at the moment, but the chief product officer is unconcerned. “We’ve not only taken this unique technology, but we’ve coupled it with a really unique approach to finding users and content, and selecting that content on a user-by-user basis, and I think the fact that users are staying in order of magnitude longer on our site than other comparable sites, or sites that we’re compared to, speaks to the fact that we’re doing something different and very valuable,” he says.
“This is a just radically different approach, and almost selfish in a way, but we think that’s good.”
Image courtesy of Trapit