It’s the Day of the Dolphin. Bottlenose (previously known as, which we initially covered here, moves out of stealth and into private beta mode. The service lassoes your Twitter, Facebook and Yammer streams, and drives real-time understanding and surfacing of personally relevant content so that you don’t have to read everything (not that you ever could!). It debuts with a new architecture for leveraging “crowd computing” for enabling scale and for creating more and more “semantic stream” smarts around the flood of information on social networks.

Nova Spivack and co-founder and CTO Dominiek ter Heide (formerly CTO of Cerego Japan who has long been tackling the issue of distilling interest profiles behind social streams) are the minds behind the service. Spivack has essentially referred to Bottlenose as everything, and more, that Twitter Annotations never was.

Annotations, if you recall, was meant to let developers add type and attribute metadata to any tweet, for better search and analytics. While Twitter last week had a major upgrade designed to make it easier to see personally relevant information, with features such as Discover for identifying stories and trends based on a user’s connections, location and language, Bottlenose is thinking bigger about tackling the noise and finding the music on the social media stream.


Bottlenose aims to be the smartest dashboard for social media, to, as Spivack says, “solve the problem of more. Right now more is bad. If you follow more people or more interests, you get too much information. We want to make more good. If something sees more of your interests and knows more about you, it just seems natural that it should do a better job of filtering for you.” Its understanding of your streams leads it to draw a clearer picture of your interests, so that messages can be assigned an importance score according to what it has learned about you.

The focus for the personal version of Bottlenose debuting now is on individual professionals and influencers rather than average consumers (a high Klout score may get you in and invites also are going out in waves to those in its Beta list). And it’s gearing up to support professional and enterprise use for team productivity and collaboration (including integration with more social enterprise applications, helping to unify external and internal message streams and support centralized administration), through to enabling social support and sales and marketing efforts. It will open things up to developers in the near future so that they will be able to craft plug-ins for full vertical applications to run on the platform, such as customer support solutions to filter issues and incidents.

Social networks and the social enterprise are in danger of collapsing under the weight of what Spivack calls the Sharepocalypse. The latter segment is growing quickly — Gartner says that by 2014 20 percent of business users will use social network services inside and outside the firewall rather than email for interpersonal communications. “The total number of messages pumped into the stream exceeds the total amount of attention of the entire human race,” Spivack says. “People can’t keep up. They get disheartened and quit. That’s a problem for the social enterprise and social media. If users stop using the stream because it’s not effective, marketers and advertisers won’t pay to reach them because they won’t get noticed.” He adds that it’s also a problem for applications that to learn to speak to the stream, as well.

As discussed in the last article, the service is building on a completely rewritten natural language engine, created in JavaScript, that understands the less-formatted and hashtag-heavy language of the social network sphere, and it suits entity-extraction to the real-time nature of the environment. It has extended the JSON Activity Streams standard to include more metadata, and that metadata both is available and displayed.  Says ter Heide, “With real-time streams you need to know that something [mentioned] is the new album by Lady Gaga. There’s no database for that [new title]. You need a whole different entity extractor for detecting something that no one’s heard of before.”

Get to the Edge

Thanks to its JavaScript architecting approach, the service is moving computation to the edge of the network. Instead of Bottlenose building a big server farm, users’ browsers run the engine that loads transparently. “We call that crowd computing – we push your desktop to become part of our cloud when you use the site. So we can scale this to a massive scale with hardly any servers,” Spivack says. The tons of users effectively add up to a huge data center and a supercomputer. “And, as we learn things from your computer, that is then shared back with the network. So everyone using Bottlenose is helping it get smarter.”

There’s no need for Bottlenose to have a firehose license, either. “When you get enough people in here you have a statistically valid sample and have basically sampled the stream. And it’s a metadata-rich stream. So the crowd gives us the same thing you get from a firehose and more metadata, too,” Spivack says. He says there’s no noticeable performance hit, either, having testing the system at over 3,000 messages per second per browser – more than what anyone even gets close to on Twitter or Facebook. The architecture ultimately is translatable to other systems, ter Heide notes.

As for what those crowd-contributors can leverage, Bottlenose’s appropriately dolphin-named Sonar technology reads the overwhelming message stream for you, figures out trending concepts for tags, topics search query or individual, corporate or brand identity, according to what’s personally important for specific individuals to pay attention to and with the capability of letting users drill deeper into any of those components. “Normally you don’t read the thousands of messages” that have been traversing the social network since you last checked in, Spivack says. “As more stuff goes through the stream and the rate of change increases, the chance of your seeing something decreases.” Sonar will deliver the relevant information personalized to the user going back so far as a day in the productivity version that’s debuting as a freemium product now, and extend back even further than that for the professional version that will offer more storage and will be available for a monthly subscription sometime in the next couple of quarters. (On the horizon is the enterprise version with more app integration and centralized admin tools.)

While not based on semantic web technologies such as RDF, SPARQL, or OWL, “Bottlenose is semantic because it’s all about understanding the meaning of things [through its linguistics processing technology and metadata]. It knows how to find different things, like only news or video or picture posts,” Spivack says. Thanks to its framework that carries more metadata, messages can be richly authored and filtered, carrying tags and type attributes that also include opinions, Q&As, and even gossip.

Pre-configured filters called Assistants are there to further help users harness the power of semantic understanding, or users can customize rules to their own ends, to message content or media types, for instance. For instance, you can have it filter on event posts in the social network space by specific sub-types, such as technology posts coming from sources with 5,000 or more followers. Users can visually pick their way through their streams, zooming into a conversation about Twitter, say, and seeing how that’s connected to other conversations about iPhones, for instance. Or, they can subscribe to only particular types of posts from someone in the social network, like their Q&A streams, or share something Bottlenose alerts them to via email, other social networks or ultimately enterprise collaboration tools. And more.

Says Spivack, “Twitter said they would do Twitter Annoations and they never did, so we are doing it. And, by the way, the metadata we create is still something we will share thru APIs in the future. We do want to make a contribution to the ecosystem.”

If you’re interested in seeing Bottlenose in action for yourself, the doors are open to the first 1,000 readers who enter the code “semanticweb”at Bottlenose.