Social Search: Intimacy Over Authority

I’ve been using Aardvark, a social search engine, since last summer. I use it a good number of times every week, and continue to be amazed by the quality and speed of answers, and the helpfulness of strangers in my extended social network.

What is social search? It’s search that relies on your social graph. Whereas, in a traditional search engine the challenge is to find the right link, in Aardvark it’s to find the right person. Their analogy: Google follows the “library paradigm of information retrieval,” Aardvark is like the “village paradigm.” In Google the algorithms determine authority (PageRank), in Aardvark the algorithms determine intimacy.

Aardvark works best when you ask questions of a highly contextualized or subjective nature – the average question on Aardvark is 18.6 words compared to 2.2-2.9 words on a web search. Questions and answers are received via IM, email, and SMS – modeling the real-world process of asking questions to friends.

Here’s a look at the type of questions asked on Aardvark:

Today, Aardvark blogged about a conference paper they submitted, The Anatomy of a Large-Scale Social Search Engine. For inquiring users like myself, it was an incredible look at how a social search engine works.

How It Works (Roughly)

A crawler and indexer indexes “topics,” which are submitted by users or extracted from users’ Facebook page, Twitter messages, Facebook news feed items, or messages sent to other users.

Aardvark creates an index with scored lists of user IDs and topic IDs. When a user enters a question into their chat window, for example, a question is sent to Aardvark’s conversation manager, which sends the question to its question analyzer, which issues a routing suggestion request to the routing engine, which searches the index and social graph of suggested user IDs.

What’s Interesting

Users aren’t just ranked by topic knowledge, but also by the “degree of social connectedness and profile similarity.” Profile similarity is determined by demographic data and social graph overlap, and also by matching vocabulary, chattiness, speed, verbosity, and politeness (e.g. Thanks!). Aardvark has found that if users have friends in common or match demographically, for instance, the questioner is more likely to receive a good answer. Also, if a user has friends with expertise in the topic, than Aardvark will judge that user as having higher expertise.

I continue to be surprised by the quality of answers I get every time I use Aardvark. I once asked how to create a pivot table in Excel, only I didn’t know it was called a pivot table and had to explain in detail what I wanted to do. I received a step-by-step answer from a gentlemen in San Francisco (read the entire thread).

According to Aardvark, users answer questions from strangers because it’s a “very gratifying experience: they’ve been selected by Aardvark because of their expertise, they were able to help someone who had a need in the moment, and they are frequently thanked for their help by the asker.”

The Impact

As of October 2009, more than 90,000 users have created accounts on Aardvark from just 2000 users in March. Over half of its users have asked a question on Aardvark and the average volume of questions per day is 3,167.2. The numbers are not insignificant and indicate a need that hasn’t been satisfied very well up to this point.

Social search like Aardvark helps tease out information from the right people – people in our social graphs who have some degree of connectedness and profile similarity to us. It’s the recognition of intimacy as just as important, if not more, than authority.