What is "Social Search"? SMX Interview

Santa Clara, CA – 2008

Allfacebook interviewed Facebook’s Aditya Agarwal at Search Marketing Expo’s SMX West. Aditya is Facebook’s Director of Search Engineering, responsible for search strategy, search product management and overseeing the engineering team responsible for new product development related to search.

AF: Please explain “social search” for the layman reader…

Aditya: Social search involves combining social graph information with pure algorithmic search results. The two are really not that different. The real question is how to use the information in the graph. There is a great variety of information that can be found in the social graph. For instance, you have friend connections, information you post about yourself and the apps that are found on your page. My team develops methods for combining that information in useful ways to help rank the results users get when they perform searches on Facebook. The core issue is this: As opposed to set of data that is returned as a search result, we extra data on top of that related to the user and use this to make the data returned more valuable. The exact details of how we do this are proprietary to Facebook. Here is an example of the way it works for a user: If you do a search of people for “Matt”, we will show you results from within your network, from your geographic area, etc. This helps ensure a high probably relevant result for you. This use of the graph to filter results can be used in searches for groups to join, events to attend and so on.

AF: What is Facebook’s unique advantage in social search?:

Aditya: Facebook’s best advantage is our complete and accurate social graph. Out competitors suffer from inaccurate graphs. Think about it, if a social network’s graph is not accurate, where 20% of the nodes and 20% of the edges are not real, then the social search results are confused and therefore not as valuable. Its hard to tell the full extent damage when a large part of your social graph is not real but it certainly devalues their graph. Facebook’s graph is full of real people, taking real actions and using real semantics.

AF: What do you think Facebook’s long term influence on social search will be?:

Aditya: Facebook will give users most accurate search results based on the most accurate social graph.

AF: What are second order effects? Why are they so important for marketers participating in social media?:

Aditya: Second order effects are influences that reach you from beyond your immediate network. Marketers can think about it from the perspective of a hotel operator: You will encounter a hotel by virtue of the fact that a friend reviewed the hotel, not because you are friends with the hotel. Also, you will be influenced by the fact your friend had a positive interaction with the hotel – this is a second order effect when this gets filtered through to you on Facebook. Marketers themselves cannot always make a direct connection with a consumer. My team works to understand a marketer’s placement in the entire graph, and how they might fit in search results that are influenced by the “localized area” of the social graph for a particular user. Keep in mind that the localized area is different for every user, and therefore the results for a particular query are different for each user. For search marketers, the difference between being a 1st and 2nd result can be how important a friend’s interaction is to the user’s query.

AF: Are there unique technology challenges in working with the social graph?

Aditya: Facebook maintains massively distributed databases. My team is responsible for determining best way to store, access and understand this data for social search.

AF: You said on your panel that at the end of the day, real world relationships do not change. What do you mean by this?:

Aditya: A lot of people think that “real world” means “physical” relationship. This is not how we think about it. People have interactions in the physical and the online worlds that form basis of relationships. We are challenged with capturing the subtlety of these interactions in systems that can inform social search results. It’s important to be able to take IM, Emails, posts, etc. and couple this with physical interactions. Facebook is able to capture the intersection of these two in a way that is both accurate and subtle.

AF: It seems like quite an exciting challenge to make sense of all the user generated content generated on Facebook, particularly non-text content like images and video.

Aditya: My hypothesis is this: As the number of content types, publishers, and volume of ratings increases, traditional measurements used by current search engines to determine global ranking cannot scale. Facebook understands that finding the most relevant results is really about your social graph. In a way, it’s similar to Google’s Page Rank for referring sites, which also makes intuitive sense. However, you will probably find that the content generated by people in your social graph a more accurate way of determining search results.