Follow the technology business news and barely a day goes by that somebody doesn't announce a new or refined social commerce or recommendation product. The concept is quite simple. Use my social graph to filter goods, services, and content making it a bit easier to get through the overwhelming volumes of each. The premise is that if my friends like it, then so will I. That premise however seems dubious to me. Even if you limit it to my closest friends, the reality is that the overlap in our taste across a broad variety of content is very small.
For example, a small group of us may really enjoy taquerias. In fact, I have a circle of friends that share notes on taquerias. And if they recommend one, I trust their recommendation implicitly. So far, so good. But now let's move on to music. This same small group has diverse musical taste. Some of it works, but some of it wouldn't make any of my play lists. So even in a small social graph, the ability to leverage it for recommendations can break down quickly.
How do you solve this problem? Clearly there needs to be another element to these filters if they are going to actually provide better answers instead of just different answers. I believe the answer lies in applying semantics. Semantics adds meaning to content and makes it possible to do more accurate and relevant matching.
Semantics focuses and associating concepts with content and defining the relationship between concepts. These relationships become critical because people are often ambiguous or use different terminology when looking for content. I might look for a taqueria but somebody else may simply look for tacos. Semantics allow the concept that a taqueria serves tacos to be established, thereby making it possible for both of us to find appropriate restaurants. The transformation from simple keyword searches (which I like to call "clumps of letters searches") to concept searches changes the relevance dramatically.
The difference between semantic and social is most clearly demonstrated by Pandora and Genius or Spotify. Pandora classifies music on over 400 distinct attributes. The music genome is one of the richer semantic systems available for a specific domain. When you select music you like or don't like, Pandora uses the concepts associated with each song to build a profile of what music you like. Genius on the other hand compares the songs you like with the songs that other people like who also like your songs. This works reasonably well if the people involved have a pretty narrow taste in music. But if their taste is broad and goes in a different direction than yours, it is less effective.
It isn't social vs semantic though. The intersection of the two is incredibly powerful. Going back to my earlier statement about taquerias, if a friend who likes taquerias recommends one, I will absolutely go there. The power is in using semantics to narrow the content I see, then expose members of my social graph that like the same content. Based on how well I know and trust them, it can boost the my confidence in the recommendation. In fact, when semantic relevance on two pieces of content are equivalent, the social graph can serve as a tie breaker.
The opportunity isn't purely semantic or purely social, but rather apply semantics to provide contextual relevance and then leverage the social graph to increase confidence in the results.