Facebook will soon launch their new Graph Search to the general public, and it has received a lot of interest lately.

With graph search, the users will be able to query the social graph that millions of people have constructed over the years when friending each other and putting in more and more personal information about themselves and their friends in the vast Facebook database. It will be possible to query for friends of friends who have similar interests as you, and invite them to a party, or to query for companies where people with similar beliefs as you work, and so on and so forth. The information that is already available, will all the sudden become much more accessible through the power of graph search.

How can we bring this to an enterprise search environment? Well, there are lots of graphs in the enterprise as well to query, both social and other types. For example, how about being able to query for people that have been members of a project in the last three years that involved putting a new product successfully to the market. This would be an interesting list of people to know about, if you’re a marketing director that want to assemble a team in the company, to create a new product and make sure it succeeds in the market.

If we dissect graph search, we will find three important concepts:

1. The information we want to query against don’t only need to be indexed into one central search engine, but also the relations and attributes of all information objects need to be normalized to create the relational graph and have standard attributes to query against. We could use the Open Graph Protocol as the foundation.
2. We need a parser that take human language and converts it to a formal query language that a search engine understands. We might want to query in different human languages as well.
3. The presentation of results should be adapted to the kind of information sought for. In Facebook’s example, if you query for people you will get a list of people with their pictures and some relevant personal information in the result list, and if you query for pictures you will get a collage of pictures (similar to the Google image search).

So the recipe to success is to give the information management part of the project a big focus, making sure to create a unified information model of the content to be indexed. Then create a query parser for natural language based on actual user behavior, and the same user studies would also give us information on how to visualize the different result set types.

I believe we will see more of these kind of solutions in the coming years in the enterprise search market, and look forward exploring the possibilities together with our clients.