Unveiled this week, Twitter’s newly upgraded search engine promises to deliver more relevant “who to follow” suggestions for users. But in my examination, the results appeared inconsistent, if not biased, when searching for the best Twitter sources in specific U.S. cities.
For example, when testing the new feature using 20 different city names as search terms, Groupon placed near the top of the new recommendations with amazing consistency given its relatively modest popularity and influence on Twitter.
By comparison, many accounts owned by major media organizations often did not rank as high in the search results, despite having significantly larger and engaged audiences.
The results raise questions about the criteria Twitter is using to rank recommended accounts, and what — if any — optimizations media companies can implement to improve their search rankings.
Twitter improves its user recommendations
The social network announced its improved search tool on Monday. In the past, searching for “iPad” or “Chicago” would have led only to accounts containing those words in their usernames. The enhanced “people results” recommends Twitter accounts deemed relevant to the term being searched:
“This new approach helps you find the Twitter users that will best help you follow your interests. For example, if you’re interested in hip hop, chances are that you’d like to follow hip hop artists. Searching for “hip hop” now surfaces accounts like @common and @questlove.”
I began digging into the functionality of the new search tool after reading Dylan Stableford’s “Top 25 Newspapers on Twitter” list. Given the new recommendations feature, how would these top media outlets appear in a search for their respective hometowns? For instance, where does the Chicago Tribune appear in a Twitter search for “Chicago”?
Using Stableford’s effort as a rough guide (itself based on work by Jeremy Porter at Journalistics) I searched for the cities of the top dozen papers noted.
In my initial searches, The New York Times and Washington Post were the only media properties to appear as top results. And overall, New York City was the most media-friendly with the @NewYorkPost and @NewYorkObserver also in the top five.
(To replicate my findings, enter a city name in the search box on Twitter, then when the results are returned, click on the “People” tab.)
Similarly in Denver, The Denver Post — with 40,239 followers — placed outside of the top 20 recommended accounts. But surprisingly, several accounts with far fewer followers — including @HuffPostDenver (2,677) and @GrouponDenver (2,989) — did appear in the top 10.
How are the search rankings calculated?
Twitter has not revealed the calculations behind its recommendations, but it is apparent the number of followers is only one criteria. I have asked them to clarify what criteria are used and I will update with a response.
The service has access to a wide array of data with which to analyze and rank user accounts. These signals include username, location, biographical text, number and content of tweets, number and content of lists, and various influence and authority calculations.
Some “misses,” such as @ColonelTribune, might be due to simple SEO (or TRO — Twitter Recommendations Optimization) issues. The Colonel’s account does not have the word “Chicago” in the username, though it is mentioned twice in the bio. The same might be said for the Atlanta Journal-Constitution’s @AJC account. Even with its 34,625 followers and recent Shorty Award nomination, it falls outside the top 20 results for “Atlanta.”
Twitter’s improvements to the search tool were meant to allow account discovery beyond simple name matching. So what types of accounts are being surfaced prominently?
Politicians and individual athletes also make a regular appearance in the top results, including @GavinNewsom in San Francisco and @JohnElway in Denver. In most cases those accounts have between 50,000 and one million followers.
Groupon, Huffington Post fare disproportionately well
But several accounts stood out during my review due to their small audience size. In Washington, D.C., @PopSugarDC was the fifth search result, though it has only 381 followers. And, in Denver @DealRodDenver also placed fifth with 375 followers.
Across the 20 different cities I eventually tested, Groupon stood out for its stunning consistency.
Here is a list of where Groupon’s Twitter account for these 20 cities ranked when searched by location name:
|City search on Twitter||Groupon’s People tab search results rank||Groupon’s # of Twitter followers|
Groupon is a popular service, but many of its local Twitter accounts have relatively modest followings. The company runs a separate Twitter account in each market — @GrouponPhilly and @GrouponNYC, for instance. Among those I examined, the average number of followers was about 6,500 with @GrouponLA, @GrouponDC and @GrouponSF all around that amount.
Despite that, Groupon placed in the top five results more than 75 percent of the time (15 times out of 20). Only twice did it fall outside the top 20 results, in Hartford and Honolulu. Even more amazingly, a Groupon-related Twitter account appeared among the top four results 11 times.
That top four placement is especially valuable given Twitter’s search results page. While each search recommends up to 20 accounts to follow, only the top four are prominently featured and are therefore more likely to gain followers.
Groupon’s success is not the result of a promotional deal with Twitter. A spokesperson for the company confirmed that it has not paid for search placement, nor is such placement a promoted product Twitter has announced. So, Groupon’s placement appears to be the organic result of Twitter’s recommendation algorithm.
Twitter has made previous attempts at developing an effective user recommendation tool. Its “Suggested User List” was discontinued in January 2010 after some debate. In 2009, Ben Lorica analyzed that feature and reported that being on the list was worth about 50,000 followers a week.
The new system is targeted more closely by interests and topics, so it is unlikely to generate follows of that magnitude. But, it does give some indication of the value of promotion and visibility on the site.
So the question is, if Twitter is not specifically tweaking its search results to promote specific accounts and companies, what characteristics are shared by those highly-ranked accounts?
Search Engine Optimization for recommendations
Twitter is likely using an engagement metric to help judge relevance, but I could not identify the logic behind it. As a proxy for Twitter’s internally developed metric, I compared several dozen accounts by their Klout scores. Klout bases its rankings on followers, retweets and general engagement calculations. Groupon’s many accounts ranked from 41 to 59. But @ChicagoTribune gets a 68 and @AJC a 76 on the same scale, and they landed outside the top 20 results.
But in general, each account had a specific location attached, most had a specific place name included in the bio, and almost all were verified accounts.
That last point is worth considering. Of the four news organizations I found ranked in the top five for their cities, each — The Washington Post, New York Times, New York Observer and New York Post — had its city in its name and each had a verified account.
Perhaps location and verification are important criteria for a social networking service like Twitter, which could begin to explain Groupon’s advantage.