Social Ranking in Search -- Opportunities and Potential Problems

Search results are useful to the degree that they’re ranked correctly from the user’s point of view. Since the dawn of search, people have thought about how to customize ranking to the needs of the user that issued the query. DirectHit was an early and successful effort in this regard, where clicks on search results were tracked and used to adjust the ranking formula for each specific user.

Major search services, including Google and Bing, now both do this sort of personal ranking adjustment to some degree, and much has been written about how search results vary as a function of an individual’s search and browsing behavior over time. How we interact with the search results we get helps determine the results that we get in the future.

Flavored Search

The more general idea here is what is often called “flavored search” — that is, search that adjusts the ranking of results based on some specific set of situational parameters. The search may be flavored by a domain restriction (e.g., only return results from the domain cnn.com), by a content type restriction (e.g., only return results that are classified as science), or by any other constraint that can be clearly and algorithmically defined.

Facebook brings an entirely new opportunity for flavored search: ranking search results using the social connections between people, as captured in the social graph.

Facebook is already providing content from public posts and pages shared on the site. Bing is making use of this, and so is OneRiot. These search engines capture and index this public content via the Facebook API and then return that content as part of their ranked search results.

This is a great step forward, as Facebook represents an excellent source of data (URLs, in particular), especially with the relatively new “Like” button generating a significant volume of information over time. As each Facebook user “likes” a page on the web, that information is registered by Facebook, and the public Facebook API makes the relevant page available to external search services.

But this isn’t really “social search.” An individual’s social graph isn’t being used to provide a unique experience. The real opportunity for social search over Facebook-provided data is much deeper, more personal, and more interesting.

Tribal Search

Jonathan Allen, Director of Search Engine Watch, has referred to this use of the Facebook social graph in search results ranking as implementing “tribal search.” I really like this term.

All Facebook users have their own unique social graph. None of us has exactly the same set of friends, so ranking search results according to our Facebook friends should deliver unique and useful results.

The key question here is, how can the social graph be used to improve ranking?

Recall that a big breakthrough of PageRank was the scalable way in which the “link as vote” idea was algorithmically assembled. PageRank doesn’t just say that the more inbound links that a given page has, the higher the rank of that page. PageRank mathematically combines “link votes,” essentially calculating a sequence of probabilities along a chain.

The power here comes from the leverage provided by any given link. Change one link somewhere in the graph of web pages, and the leverage provided by that change is massive.

This is what I mean when I say that PageRank is scalable. The algorithm leverages local information in a way that scales up to take on the challenge of a web-sized graph.

The real opportunity for tribal search is to leverage the social graph in much the same way that PageRank leverages the web graph. In fact, “link-based calculations” from both graphs need to be combined in some appropriate way, so that votes for page content can be suitably weighted and combined, using both web graph links and social graph links. Doing this correctly will deliver real tribal search results.

Simply counting inbound links in the social graph and using that count as a weighting factor isn’t good enough. Some insight parallel to the one provided by PageRank will need to be applied to the social graph as combined with the web graph.

Benefits and Concerns

True tribal search could have significant benefits, but there are also potential concerns.

One defining benefit is that we will be able to search the web and have our search results ranked according to what our particular “tribe” thinks is good. For example, if I’ve paid a lot of money to go to a top school, and am now well-linked in the Facebook social graph to fellow graduates of that school, then being able to search the web according to the interests and preferences of my fellow alumni is tremendously useful.

Ranking search results based on the recommendation of my tribe has a potential problem: we each can become more easily locked into our existing worldview. We connect to like-minded people on Facebook. And with tribal search of the sort that I’m discussing, our search results will be delivered through the attention lens of those people.

Tribal search could easily become a way of reinforcing existing preferences and biases. We could end up searching for exactly what we expect to find, and then find only things that don’t surprise us.

However, the benefits of well-implemented social search will outweigh potential problems.

The truth is, we each already have many filters through which we obtain and value information. We listen to certain radio stations, we watch only specific TV shows, and we tend to visit certain web sites. We already have a rich set of information-distortion tools available, and we eagerly make use of them in an effort to simplify the world and to make sense of the information we get (or at least, to get only information that’s consistent with what we want to believe).

Tribal search will just bring that same sort of information bias to search. But properly and clearly implemented, tribal search will be potentially powerful and personal. Truly personalized search of Facebook and of the entire web will revolutionize the way people interact with the online world.

Join us for SES San Francisco August 16-20, 2010 during ClickZ’s Connected Marketing Week. The festival is packed with sessions covering PPC management, keyword research, search engine optimization (SEO), social media, ad networks and exchanges, e-mail marketing, the real time web, local search, mobile, duplicate content, multiple site issues, video optimization, site optimization and usability, while offering high-level strategy, keynotes, an expo floor with 100+ companies, networking events, parties and more!