In April 2010, the U.S. Library of Congress and Twitter announced that Twitter will donate its public Twitter conversation data to the Library of Congress archives, making it available for researchers to read and preserving it for generations to come. This was an historic moment: for the first time, a communication channel in the social media space had been identified as a relic of our culture, emblematic of the role technology has come to play in our interactions with one another.
Today we’re honored to announce that Gnip is playing a key role in the this project by delivering past and future Twitter data to the Library of Congress for historic preservation in the archives.
“By providing a complete and reliable stream of Twitter data, Gnip is playing a critical role in helping the Library of Congress build a stable, sustainable archive for future generations,” said a representative from the Library of Congress.
On a technical level, the magnitude of this project is enormous: with over a billion Tweets being generated weekly, the scale of data delivery involved in this project is historic. We’ve learned a lot from this project and it is clear that some of the outcomes from this project will benefit Gnip’s future product offerings. We look forward to sharing some of the technical details in the future at the appropriate time.
As a company, we couldn’t be more pleased to be a part of this momentous project. We’re eager to see what kind of research develops from the public archive made accessible by the Library of Congress. We’ve watched with awe as news of world-changing events like the Arab Spring and Osama Bin Laden’s death have flowed through our servers, and now we’re proud to be part of ensuring these historic events will be preserved for our posterity.
The archive of Twitter data which Gnip delivers to the Library of Congress will cater to the needs of researchers who wish to use limited amounts of public Twitter data for non-commercial purposes. Gnip will continue to serve those seeking realtime data, full-coverage data, and commercial use cases.
Thanks to Twitter and the Library of Congress for making this announcement possible. We couldn’t be more excited to be working with you both.
Gnip’s business is growing heartily. As a result, we need to field current demand, refine our existing product offering, and expand into completely new areas in order to deliver the web’s data. From a business standpoint we need to grow our existing sales team in order to capture as much of our traditional market as possible, as fast as possible. We also need to leverage established footholds in new verticals, and turn those into businesses as big as, or hopefully bigger than, our current primary market. The sales and business-line expansion at Gnip is in full swing, and we need more people on the sales and business team to help us achieve our goals.
From a technical standpoint I don’t know where to begin. We have a large existing customer base that we need to keep informed, help optimize, and generally support; we’re hiring technical support engineers. Our existing system scales just fine, but software was meant to iterate, and we have learned a lot about handling large volumes of real-time data streams, across many protocols and formats, for ultimate delivery to large numbers of customers. We want to evolve the current system to even better leverage computing resources, and provide a more streamlined customer experience. We’ve also bit off a historical data set indexing challenge that is well… of true historical proportion. The historical beast needs feeding, and it needs big brains to feast on. We need folks who know Java very well, have search, indexing, and large data-set management backgrounds.
On the system administration side of things… if you like to twiddle IP tables, tune MTUs for broad geographic region high-bandwidth data flow optimization, handle high-volume/bandwidth streaming content, then we’d like to hear from you. We need even more sys admin firepower.
Gnip is a technical product, with a technical sale. Our growth has us looking to offload a lot of the Sales Engineering support that the dev team currently takes on. Subsequently we’re looking to hire a Sales Engineer as well.
Gnip has a thriving business. We have a dedicated, passionate, intelligent team that knows how to execute. We’re building hard technology that has become a critical piece of the social media ecosystem. Gnip is also located in downtown Boulder, CO.
My favorite requirement from customers is the “I want all the data, from all the sources, for all of history, and for all of future” one. You’re never going to get it, from anyone, so reset your expectations. A few constructs fall out of this request.
Two Types of ‘feeds’
These are aggregate sources of data for a given publisher. They may, or may not, be a complete representation of that publisher’s data set. Everyone wants firehoses, but truth be told, there are very few of them in the wild, and those that do exist are of less “valuable” data. Consider access to firehoses to be at statistically relevant access levels, rather than truly “complete” sets of data.
These encompass the majority of data sources, and they require that you know what you’re looking for. Be it a keyword, a tag, a user name, a user id, or a geo-location.
In either case you need to know what it is you’re after. Blind, unfettered, access to a given publisher’s feed is a rarity and actually isn’t all that interesting in the end; you just think it is because someone else had the product idea first (e.g. the publisher you want all the data from… e.g. Twitter).
Storing and indexing lots of data is conceptually simple, yet hard to implement at scale; just ask any of the big-three search engines. You can stuff as much data as possible into a database, and “search” it offline, in order to meet most historical data access requirements, but weaving that into a variably accessed consumer application isn’t always easy. While storage costs are generally nil for today’s highly compressible data, the operational management costs of your locally stored data aren’t.
Processing data in a manner other than which it originated causes an impedance miss-match. Stream-to-offline processing implies that you’ll have gaps in data due to queuing problems. Offline-to-stream suggests the same. Offline-to-offline and stream-to-stream are generally easy to get your head and code around, but be wary of overloading stream processing with too much work as it then starts to feel like stream-to-offline. Once you enter that world, you need to solve parallel processing problems; in real-time.
Regardless of access pattern, you can only introspect and access the data you initially seeded your sources with. If your seed was wrong, for example you used the wrong set of users or keywords, processing the data doesn’t matter. Full circle to garbage in, garbage out.
If you find yourself asking for the introductory requirement with your team, and/or a vendor, I suggest you actually don’t have the focus on your product or idea that you’ll ultimately need in order to be successful. Batten down the hatches, and get crisp about precisely what it is you want to build, and precisely what data you need to do so. If you can do that, you will have a shot at success.
John Battelle wrote a great post last week titled “What Marketers Want from Twitter Metrics” in which he recounts a conversation with Twitter COO Dick Costolo and lists some data he hopes we’ll soon see from Twitter. These metrics include:
- How many people *really* see a tweet. Even though @gnipsupport has 150 followers, it’s unlikely that they all saw our tweet about this post.
- Better information around engagement, such as retweets and co-incidence data. There’s a classic VC saying: “the first time I hear about something I don’t notice; the second time, I take an interest and the third time I take action.”
For me, marketing is about sending a signal into the marketplace and then measuring how effectively it is received. For instance, Gnip is trying to better engage with companies that use third-party APIs, and since we’re a startup, low cost matters. One mechanism is this blog and the article you’re reading now. That’s the “sending a signal” part. While you’re reading this, I’m likely logged into Google Analytics, monitoring how people find this article, and watching Twitter to see if anyone mentions this post. That’s the “measuring effectiveness” part. And this isn’t a static, one-time cycle. Based upon the feedback I get (some direct, some inferred), I’ll write and promote future posts a little differently.
I am positive that Twitter and other forms of social media will be hugely beneficial to marketing and the surrounding fields of sales, advertising and customer service. Highly measurable and disintermediated low-friction customer interactions with the marketplace is a wonderful thing. However, if five years from now we’re still primarily talking about social media in terms of marketing, then an opportunity has been squandered.
If marketing is a company sending a signal to the marketplace and measuring how it is received, then business intelligence (from a product perspective) is the process of measuring and acting on the signal that the marketplace itself is sending. For instance, last holiday season, a major discount chain wanted to know why, in the midst of the a recession, many of their traditional customers were opting to shop at more expensive competitors. After examining Twitter, Facebook and other social services, they discovered that customers were unhappy with their stores’ lack of parking and cashiers. Apparently, even in a financial crunch, convenience trumps price. The store took steps to increase the number of cashiers and sales immediately increased. THIS is where I’d like to see more emphasis in social media.
It’s a function of magnitude
With marketing, the product or service has already been created and success is now predicated on successfully engaging as many people as possible with your pitch. The primary question is “How do we take this product and make it sound as appealing as possible to the market?” Great marketing can create far greater demand than a shoddy one, but in the end, the product is fairly static by that point. Sales is plotted on a continuum defined as “customer need multiplied by customer awareness,” where need is static and awareness if a variable. What if you could change the scale of customer need?
When the product or service is still being defined, the size of the opportunity is extremely fluid. A product that doesn’t address a customer need isn’t going to sell a ton, regardless of how well it’s marketed. A product that addresses a massive customer need can still fail with poor marketing, but it will be a game changer with the right guidance. Business intelligence is crucial to the process of identifying the biggest need in a market and building the appropriate solution.
Steve Ballmer is very vocal about how he only cares about ideas that will move his stock price a dollar. But to move his stock price by even $0.10 at today’s P/E is to increase earnings (earnings!) by almost $100MM annually. In other words, if you’re a startup whose product can’t generate a billion dollars, then it’s not worth Microsoft’s time to talk to you. And if you’re a MS product manager who isn’t working on a billion dollar product, you might want to put in a transfer request. Or better yet, listen to the market and retool what you’re currently building, because no amount of marketing is going to save you.
Yeah, “billion” with a “b”
Typically, entrepreneurs use personal experience and anecdotal evidence to design their offering. Larger companies may conduct market research panels or send out surveys to better understand a market. We are now blessed with the ability to directly interact with the marketplace at a scale never previously imagined. The market is broadcasting desire and intent though a billion antennae every day, yet product managers are still casting a deaf ear. Maybe we need better tools and data so that the business world to start tuning in.
First off, when you’re launching a product, you ought to know what the market looks like. We need better access to user demographics, both at the service level (who uses Twitter) as well as the individual level (who just tweeted X). A number of companies are starting to serve this need (folks like Klout, who offers reputation data for Twitter users, and Rapleaf, who offers social and demographic data based on email address) but there is a massive way to go. I would kill for the ability to derive aggregated demographics — tell me about all the people who tweeted Y in the last year.
Secondly, access to historical data is critical. When deciding whether to even begin planning a new product, it’s important to know whether the marketplace’s need is acute or a long-standing problem. Right now, it’s nearly impossible to access data about something from before the moment you realize you should be tracking it. This has led to all sorts of “data hoarding” as social media monitoring services attempt to squirrel away as much data as possible just in case they should need it in the future. The world would be so much better with mature search interfaces. Think about your average OLAP interface and then think about Facebook Search. Twitter has already said that they are taking steps to increase the size of their search corpus; let’s make sure they know this is important and let’s encourage other social services to make historical data available as well.
The best part of all this is that Marketers and Product Managers need many of the same things — they’re in the same universe, you might say. The best companies engage marketing as the product is being defined, and a result, a lot of these metrics will be benefit product managers and marketers alike.
Dell selling $6 million of computers on Twitter? That’s pretty great. Dell identifying a new $600M market because of signals sent on Twitter… that’s simply amazing. And that’s the level of impact I hope to see social media have in the next few years. Got your own ideas on how we can get there from here? Post ‘em in the comments.
Long, long ago, in a galaxy far, far away, Twitter provided a firehose of data to a few of partners and the world was happy. These startups were awash in real-time data and they got spoiled, some might say, by the embarrassment of riches that came through the real-time feed. Over time, numerous factors caused Twitter to cease offering the firehose. There was much wailing and gnashing of teeth on that day, I can tell you!
At roughly the same time, Twitter bought real-time search company Summize and began offering to everyone access to what is now known as the Search API. Unlike Twitter’s existing REST API, which was based around usernames, the Search API enabled companies to query for recent data about a specific keyword. Because of the nature of polling, companies had to contend with latency (the time between when someone performs an action and when an API consumer learns about it) and Twitter had to deal with a constantly-growing number of developers connected to an inherently inefficient interface.
Last year, Twitter announced that they were developing the spiritual successor to the firehose — a real-time stream that could be filtered on a per-customer basis and provide the real-time, zero latency results people wanted. By August of last year, alpha customers had access to various components of the firehose (spritzer, the gardenhose, track, birddog, etc) and provided feedback that helped shape and solidify Twitter’s Streaming API.
A month ago Twitter Engineer John Kalucki (@jkalucki) posted on the Twitter API Announcements group that “High-Volume and Repeated Queries Should Migrate to Streaming API“. In the post, he detailed several reasons why the move is beneficial to developers. Two weeks later, another Twitter developer announced a new error code, 420, to let developers identify when they are getting rate limited by the Search API. Thus, both the carrot and the stick have been laid out.
The streaming API is going to be a boon for companies who collect keyword-relevant content from the Twitter stream, but it does require some work on the part of developers. In this post, we’ll help explain who will benefit from using Twitter’s new Streaming API and some ways to make the migration easier.
Question 1: Do I need to make the switch?
Let me answer your question with another question — Do you have a predictable set of keywords that you habitually query? If you don’t, keep using the Search API. If you do, get thee to the Streaming API.
- Use the Streaming API any time you are tracking a keyword over time or sending notifications / summaries to a subscriber.
- Use the Streaming API if you need to get *all* the tweets about a specific keyword.
- Use the Search API for visualization and search tools where a user enters a non-predictable search query for a one-time view of results.
- What if you offer a configurable blog-based search widget? You may have gotten away with beating up the Search API so far, but I’d suggest setting up a centralized data store and using it as your first look-up location when loading content — it’s bad karma to force a data provider to act as your edge cache.
Question 2: Why should I make the switch?
- First and foremost, you’ll get relevant tweets significantly faster. Linearly polling an API or RSS feed for a given set of keywords automatically creates latency which increases at a linear rate. Assuming one query per second, the average latency for 1,000 keywords is a little over eight minutes; the average latency for 100,000 keywords is almost 14 hours! With the Streaming API, you get near-real-time (usually within one second) results, regardless of the number of keywords you track.
- With traditional API polling, each query returns N results regardless of whether any results are new since your last request. This puts the onus of deduping squarely on your shoulders. This sounds like it should be simple — cache the last N resultIDs in memory and ignore anything that’s been seen before. At scale, high-frequency keywords will consume the cache and low frequency keywords quickly age out. This means you’ll invariably have to hit the disk and begin thrashing your database. Thankfully, Twitter has already obviated much of this in the Search API with an optional “since_id” query parameter, but plenty of folks either ignore the option or have never read the docs and end up with serious deduplication work. With Twitter’s Streaming API, you get a stream of tweets with very little duplication.
- You will no longer be able to get full fidelity (aka all the tweets for a given keyword) from the Search API. Twitter is placing increased weight on relevance, which means that, among other things, the Search API’s results will no longer be chronologically ordered. This is great news from a user-facing functionality perspective, but it also means that if you query the Search API for a given keyword every N seconds, you’re no longer guaranteed to receive the new tweets each time.
- We all complain about the limited backwards view of Twitter’s search corpus. On any given day, you’ll have access to somewhere between seven and 14 days worth of historical data (somewhere between one quarter to one half billion tweets), which is of limited value when trying to discover historical trends. Additionally, for high volume keywords (think Obama or iPhone or Toyota), you may only have access to an hour of historical data, due to the limited number of results accessible through Twitter’s paging system. While there is no direct correlation between the number of queries against a database and the amount of data that can be indexed, there IS a direct correlation between devoting resources to handle ever-growing query demands and not having resources to work on growing the index. As persistent queries move to the Streaming API, Twitter will be able to devote more resources to growing the index of data available via the Search API (see Question 4, below).
- Lastly, you don’t really have a choice. While Twitter has not yet begun to heavily enforce rate limiting (Gnip’s customers currently see few errors at 3,600 queries per hour), you should expect the Search API’s performance profile to eventually align with the REST API (currently 150 queries per hour, reportedly moving to 1,500 in the near future).
Question 3: Will I have to change my API integration?
Twitter’s Streaming API uses streaming HTTP
- With traditional HTTP requests, you initiate a connection to a web server, the server sends results and the connection is closed. With streaming HTTP, the connection is maintained and new data gets sent over a single long-held response. It’s not unusual to see a Streaming API connection last for two or three days before it gets reset.
- That said, you’ll need to reset the connection every time you change keywords. With the Streaming API , you upload the entire set of keywords when establishing a connection. If you have a large number of keywords, it can take several minutes to upload all of them and during the duration you won’t get any streaming results. The way to work around this is to initiate a second Streaming API connection, then terminate the original connection once the new one starts receiving data. In order to adhere to Twitter’s request that you not initiate a connection more than once every couple of minutes, highly volatile rule sets will need to batch changes into two minute chunks.
- You’ll need to decouple data collection from data processing. If you fall behind in reading data from the stream, there is no way to go back and get it (barring making a request from the Search API). The best way to ensure that you are always able to keep up with the flow of streaming data is to place incoming data into a separate process for transformation, indexing and other work. As a bonus, decoupling enables you to more accurately measure the size of your backlog.
Streaming API consumers need to perform more filtering on their end
- Twitter’s Streaming API only accepts single-term rules; no more complex queries. Say goodbye to ANDs, ORs and NOTs. This means that if you previously hit the Search API looking for “Avatar Movie -Game”, you’ve got some serious filtering to do on your end. From now on, you’ll add to the Streaming API one or more of the required keywords (Avatar and/or Movie) and filter out from the results anything without both keywords and containing the word “Game”.
- You may have previously relied on the query terms you sent to Twitter’s Search API to help you route the results internally, but now the onus is 100% on you. Think of it this way: Twitter is sending you a personalized firehose based upon your one-word rules. Twitter’s schema doesn’t include a <keyword> element, so you don’t know which of your keywords are contained in a given Tweet. You’ll have to inspect the content of the tweet in order to route appropriately.
- And remember, duplicates are the exception, not the rule, with the Streaming API, so if a given tweet matches multiple keywords, you’ll still only receive it once. It’s important that you don’t terminate your filtering algo on your first keyword or filter match; test against every keyword, every time.
Throttling is performed differently
- Twitter throttles their Search API by IP address based upon the number of queries per second. In a world of real-time streaming results, this whole concept is moot. Instead, throttling is defined by the number of keywords a given account can track and the overall percentage of the firehose you can receive.
- The default access to the Streaming API is 200 keywords; just plug in your username and password and off you go. Currently, Twitter offers approved customers access to 10,000 keywords (restricted track) and 200,000 keywords (partner track). If you need to track more than 200,000 keywords, Twitter may bind “partner track” access to multiple accounts, giving you access to 400,000 keywords or even more.
- In addition to keyword-based streams, Twitter makes available several specific-use streams, including the link stream (All tweets with a URL) and the retweet stream (all retweets). There are also various levels of userid-based streams (follow, shadow and birddog) and the overall firehose (spritzer, gardenhose and firehose), but they are outside the bounds of this post.
- The best place to begin your quest for increased Streaming API is an email to firstname.lastname@example.org — briefly describe your company and use case along with the requested access levels. (This process will likely change for coming Commercial Accounts.)
- Twitter’s Streaming API is throttled at the overall stream level. Imagine that you’ve decided to try to get as many tweets as you can using track. I know, I know, who would do such a thing? Not you, certainly. But imagine that you did — you entered 200 stop words, like “and”, “or”, “the” and “it” in order to get a ton of tweets flowing to you. You would be sorely disappointed, because twitter enforces a secondary throttle, a percentage of firehose available to each access level. The higher the access level (partner track vs. restricted track vs. default track), the greater the percentage you can consume. Once you reach that amount, you will be momentarily throttled and all matching tweets will be dropped on the floor. No soup for you! You should monitor this by watching for “limit” notifications. If you find yourself regularly receiving these, either tighten up your keywords are request greater access from Twitter.
Start tracking deletes
- Twitter sends deletion notices down the pipe when a user deletes one of their own tweets. While Twitter does not enforce adoption of this feature, please do the right thing and implement it. When a user deletes a tweet, they want it stricken from the public record. Remember, “it ain’t complete if you don’t delete.” We just made that up. Just now. We’re pretty excited about it.
Question 4: What if I want historical data too?
Twitter’s Streaming API is forward-looking, so you’ll only get new tweets when you add a new keyword. Depending on your use case you may need some historical data to kick things off. If so, you’ll want to make one simultaneous query to the Search API. This means that you’ll need to maintain two integrations with Twitter APIs (three, if you’re taking advantage of Twitter’s REST API for tracking specific users), but the benefit is historical data + low-latency / high-reliability future data.
And as described before, the general migration to the Streaming API should result in deeper results from the Search API, but even now you can get around 1,500 results for a keyword if you get acquainted with the “page” query parameter.
Questions 5: What if I need more help?
- Twitter Streaming API docs: http://apiwiki.twitter.com/Streaming-API-Documentation
- Twitter Search API docs: http://apiwiki.twitter.com/Twitter-API-Documentation
- Twitter API Announcements: http://groups.google.com/group/twitter-api-announce
- Twitter Development Talk: http://groups.google.com/group/twitter-development-talk
- Twitter IP Address Whitelist Request Form: http://twitter.com/help/request_whitelisting
Streaming HTTP resources:
- Wikipedia entry: http://en.wikipedia.org/wiki/Comet_%28programming%29
- cURL for debugging
- Ask questions in the comments below and we’ll respond inline
- Send email to email@example.com to ask the Gnip team direct questions