The Feat of Aardvark and Matching People
Who knew my previous post on Aardvark and the future of social networking would be the straw that broke the camel’s back? Yesterday, Google announced that it purchased Aardvark for a rumored $50 million (see TechCrunch and Aardvark’s Blog). More on that later.
First I’d like to review a paper written by the aforementioned Damon Horowitz and Google’s former head of personalization, Sep Kamvar (Damon’s post and the paper itself). Their paper outlines, for all intents and purposes, exactly how Aardvark’s system works. There are a number of fascinating details which the technorati or general public previously did not know.
Before reading Horowitz’s paper, we knew that Aardvark matched an answerer to a questioner based on the ability to infer that the answerer had sufficient knowledge to answer the question. What we didn’t know is the extent to which Aardvark tries to match the actual answerer to the questioner herself. Indeed, Horowitz, et al. write,
the Routing Engine scores each user according to the degree to which she herself—as a person, independently of her topical expertise—is a good “match” for the asker for this information query. The goal of this scoring is to optimize the degree to which the asker and the answerer feel kinship and trust, arising from their sense of connection and similarity
This philosophical viewpoint–that the user who is most similar to the questioner will be the best person to answer the question–is a fascinating approach to developing a search system. The way Aardvark designed their system reflects how psychological and sociological theories of communication and developing relationships should inform the architecture of a social networking system, in an overtly top-down manner. The next two sections will examine how Aardvark has applied psychological or sociological theories in their design.
Matching People by Communication Style
Aardvark computes the connectedness between users using a number of variables. Some are easy to guess, such as number of common friends and demographic similarity. Others are more intuitively clever of human communications. Aardvark matches users by:
Vocabulary match (e.g., IM shortcuts)
Chattiness match (frequency of follow-up messages)
Verbosity match (the average length of messages)
Politeness match (e.g., use of “Thanks!”)
Speed match (responsiveness to other users)
People communicate in different ways. Pop culture has shaped the cultural norms for many teenagers and young adults who communicate with IM shortcuts (e.g., “lol”, “omg”) but there are still many young adults that don’t communicate that way. And if Aardvark is trying to induce a 1-on-1 conversation between users, it is wise to pare apart or match people by how they communicate. The rest of the factors also reflect this principle.
Aardvark also matches people by profile similarity (e.g., common favorite movies) and they also extract “topics from users’ status message updates and from the messages they send to other users on Aardvark.” Beyond being able to match users by information disclosed in profiles and on status messages, services like Aardvark can use the aggregate of user data they pull to determine normative interests, movies, and music. What do I mean by this? Many users hew to the normative: interests they declare, movies they like, and music they listen to are largely the same music, movies, and interests as most other users on Facebook.
Services like Aardvark can compare one user’s profile data against the aggregate profile data of their friends, to understand how much they deviate. What would this deviation tell Aardvark? It might be a signal as to how strong their interests are. For example, if the great majority of Joe’s friends specify interests of “off-roading, bmx, motocross, kenny chesney,” and Joe lists interests of “politics, philosophy, classical music,” it signals to Aardvark that Joe’s interests are in the 5% of users that deviate from the normative, relative to their immediate social network.
This deviation could serve as a signal to a number of different things. It could say that Joe is so passionate about his interests that he’s willing to be a social outlier. It could reflect on Joe’s opposition to herd mentality and his interest in individuality, both being important personality variables. By taking profile data and status messages and putting them to use, Aardvark is continually assembling personality profiles for all of their users. This, I should point out, is not being done by many Facebook applications, to their loss.
Indeed, the language we use reflects who we are. James Pennebaker and now Sam Gosling, two psychologists from UT, have taken research from psycholinguistics and have started to give it to computer scientists. A number of interesting papers can be seen on ICWSM’s website.
Horowitz and Kamvar’s paper is a great read and despite the possibility that it gives their competitors (such as Adam D’Angelo’s Quora) a head start, it does a great service toward advancing the direction that social networking ought to go in.
A Note on the Acquisition
As I have previously said, I think services such as Aardvark represent the right direction for social networking. I want to briefly discuss the price of acquiring Aardvark only as a proxy of valuing the service. That being said, I don’t know how Yelp managed to cull a rumored $500 million offer from Google, especially because I think there are many limits to the service. Yelp has the user-base, Aardvark had 90,000 as of September ’09, but I don’t know that user-base has or should have a firm correlation with price. It may seem like I’m comparing apples and oranges, what with Yelp being a way to inform consumers about shopping and dining, and Aardvark being a general recommendation service, but I think that Aardvark could potentially expand into the domain of informing purchases without many of the biases and flaws in trust apparent in Yelp. If Aardvark can manage to continue to improve their service (and avoid any possible symptoms of “participation fatigue“) and seriously cut into Facebook’s 400 million-strong user-base, the future will be very bright.
Subscribe to comments with RSS.