The digital marketing world is aflutter with talk of entities and structured data, and how both are contributing to the rise of semantic search. Google has rolled out its Hummingbird update, while Facebook has its Graph Search. Both represent big leaps forward in easing searchers’ access to the information they’re seeking.
The creators of social media platforms are amassing some of the largest databases ever known. All of the data is being input on a volunteer basis by all of us who participate in social media. Much of the time, that data is being input in a way that has structure; a structure that defines where I am (check-ins), whom I’m with (tags), or that I’ve indicated an affinity (likes). Ultimately, in selling advertising, social platform companies have an incredible opportunity to create wealth based on that structured data.
In March 2013 Facebook launched its new search system, based on its proprietary search systems called Unicorn and Typehead.
How Graph Search Is Constructed
With Graph Search, Facebook is showing its cards on the data it possesses. While the fundamental user interface uses wizards that guide the user, once you learn the syntax, you can easily type queries directly into the search box. The more you play with Graph Search, the more you can see how Facebook is putting information together, and perhaps even imagine new ways to use the data to help your own marketing efforts.
First, there’s the initial search box, asking for you to search on “people,” “places,” and “things.”
The first element of Facebook Graph Search:
People, Places, and Things are entities. Entities then have a relationship, such as “Liked,” “posted,” “by,” “of,” “taken by,” and so on. In network theory, these entities and relationships are called nodes and ties, while in graph theory the latter are more commonly referred to as edges. We can see a query that is built like this:
{Entity} {Relationship} {Entity}
For instance,
“Photos Liked by Me”
You can think of the Entities as the nouns, and the Relationship as the verb. Both Entities and Relationships have metadata associated with them
The queries can build on one another, for instance, I can have, [friends of my son who liked my photos]. I can also have a query like this, [Friends of my son’s friends who liked my photos].
Some relationships can only be possessed by certain types of objects. I can, for instance, take a photo but not a book. Taken by would have a relationship to a person, whereas Taken In would have a relationship to a location or a date.
You can also use Boolean expressions to build more complex queries, such as, [Friends of my son’s friends who liked my photos and live in Kingston, New York].
The system is sophisticated enough to be able to receive queries in different order. For instance, it can parse [my son’s friends] as [friends of my son]. After all, an apostrophe with the letter “s” does connote “of his” or “of mine,” depending on its adjacency.
According to FB representatives, these “objects” are only a beginning — they wanted to roll out Graph Search early and get feedback. For the future, the engineers are planning an even greater richness in what Graph Search can determine.
Results
The results returned on any search seem to be returned in order – as related to by the searcher’s connections. It doesn’t only show results that are related to the searcher, though. For instance, if I search on [People who like New Found Glory] (a punk band), I will see a list of people and their connection to my own friends. If I search for [pages liked by People who like New Found Glory], it will return results not liked by anyone in my own network.
How It Can Be Used As A Research Tool
Tom Stocky, one of the two former Google employees who helped develop Graph Search, said that like Newsfeed and Timeline, Graph Search comprises one of the major pillars of Facebook. It’s very probable that Graph Search is going to become a lot more than something for marketers to play with; it will likely become a valuable way for users to find relevant content.
One way Graph Search can be used by marketers is as a research tool for community research in performing what has been called online ethnography or netnography. I could, for example, be studying people who love elephants.
This query leads me to a list of countless other pages liked by those elephant savers. What else can I discover about them? How about [books ready by people who like save the elephants]?
Unfortunately, the order in which Facebook returns the results doesn’t necessarily tell me which results are most liked by my audience.
I could bifurcate my query like this: [People who like Save the Elephants}} and who like [We Are Human Angels]. I can continue to stack the queries until I identify particular people, and see if I can glean anything particular about those individuals.
Or, I can go to groups that those people belong to, and study the conversations there. If it’s pure numbers I’m after, I’ll use the ad tools instead, to break up the queries and get numerical results.
Community Research
Analyzing posts and comments from a group is not as simple as working with pages. If you want to study past conversations, you’ll need to write a script for importing that data so that you can run it through more advanced tools. If you want to import current conversations, you can have those group posts emailed to you, whereupon they can be automatically delivered to a database.
Depending on the scale of your efforts, performing online ethnography may merit a dozen hours, or hundreds. If, in your own work, the longer time commitments make sense, someone might manually study the posts to determine the overall theme of the post and the sentiment of the comments.
When you’re dealing with group posts that have hundreds of comments, and now comments on comments, the manual review may be unwieldy. You can also just look for the main elements that are meaningful to your efforts. For instance, in studying the “Save the Elephants” group, I was able to review to major comments like this:
One of the problems with automated sentiment analysis is that the vocal majority may skew the results. At the same time, you’d want to take note of those individuals to determine if they are exerting an influence on the group as well.
Human relationships are complex and messy. It’s rather amazing that in social networks we’re able to quantify and structure some of that turmoil, and hopefully, acquire insights that add value to our work.
Commentaires