In this part 2 we review the sociograph vs the tastegraph as a means for recommending. Recommendation through the sociograph means drawing advice and responses from those in the same social network as a user. The tastegraph refers to basing recommendations and search results on the preferences of and feedback from people who like the same things as the user regardless of the existence or not of a social connection.
I asked Medros what he thought about these two trends and whether or not he thought the tastegraph was a better recommendation process than the sociograph. Here is what he had to say
Medros of TripAdvisor: When we talk recommendations at TripAdvisor we talk about a hierarchy of advice. The top of the hierarchy is the wisdom of friends and this is what Trip Friends enables. Trip Friends does more than allowing you to ask your friends, it lets you know which friends to ask. At the other end of the hierarchy is what TripAdvisor has been doing for 10 years – wisdom of crowds. [The crowd providing] A broader set of advice and opinions than available from a transaction provider. In the middle there is the tastegraph – we call it “people like me”. Where you can filter down the recommendations and wisdom of friends through “people like you” filters. The tastegraph is better than wisdom of crowd but not as good as wisdom of friends.
There are three approaches to collecting data for recommendation engines. Two don’t work.
- Give to Get: Asking consumers to do a survey of who they are and use the results of the survey to show recommendations. It doesn't work and no one does it;
- Black box: Take all of the click steam data – put it into the "black box" and produce an answer for the consumer. Does not work. No matter how much consumers tell you they want recommendations this way, they (the consumers) do not believe the results when they come out. This method is easier for a transaction site that has only transaction data to rely on. But for research site consumers are all over the site for many different reasons. Too noisy to trust the click data; and
- People Like Me and Finger Printing: Try to finger print a hotel or travel experience by asking people to choose similar experiences. The best version of this is, "If you like this hotel in city X you will like hotel Z in city Y".
I see the value in having access to friends to complement the crowd content on TripAdvisor. This is also a very good implementation by TripAdvisor. That said I feel that the path to recommendations is through the tastegraph/people like me approach more than the sociograph. I have lots of friends, close friends, that have very different interests to me and whose recommendations I won't take.
The first time I used Trip Friends it took a few clicks more than I expected to get logged in and up and running. Once in, I found the product easy to use and easy to integrate into a search. But looking down the list of friends connected for the destination I was interested in (Hong Kong) I was not convinced that any of them knew any more than I did. The one friend I do have that lives and grew up in Hong Kong did not have that in her profile so was missed.
I think the "like button" is going to rule the world one day. By that I mean that tools and products to allow content consumers (online or off) to positively or negatively vote for an item or content (including Facebook, Digg, StumbledUpon and more) will drive the future of content and transactional consumption. But I think this is more likely to happen in a "people like me" or tastegraph environment than an environment driven by "what do my friends think".
Now to you - tastegraph or sociograph - which one will rule?