A latest development of the social graph functionality of Facebook, the latest graph search features highly organic inter-site search engine functionality where ‘natural question’ queries are processed to reveal personal data from the database of billions of users on Facebook. The social graph functionality refers to the core of Facebook’s mode of operation where users are mapped on a social graph based on the interaction and relationships between them.
The graph search makes use of personal data like never before and is anticipated to revolutionize the way online searches are performed. It is interesting to note that Facebook has partnered with Microsoft Bing and not Google to retrieve user information beyond the platform of Facebook. Yes, that is right. Facebook is planning to use its ‘in house’ information as well as external information (with the help of Bing) to display the results of graph search. The results from Bing would be based on traditional keyword match search engine results. The program is still in the Beta stage and is selectively rolled out to a few thousand users from the US (and also from elsewhere, if you are enrolling to the graph search feature on the site). The reason for the slow rollout is the utmost privacy concern of the graph search utility raised by bloggers, journalists, Facebookusers and internet activists in general. In fact, keeping this in mind, Zuckerberg’s company had also rolled out a host of new privacy settings so that users get accustomed to them before embarking on the graph search platform.
The engineering team that developed the graph search utility for Facebook was headed by former Google engineers Lars Rasmussen and Tom Stocky. Rasmussen, describing himself as the ‘pointy haired engineering director’ recently appeared in Reddit for an AMA (Ask Me Anything) session as part of the promotional process of Graph Search. This AMA session revealed some very crucial info about the graph search platform, and is in fact the only source of comprehensive information as of now regarding the design, development and working of the semantic search engine platform inside Facebook.
Here’s how Graph Search works.
The core of the utility consists of a framework of context-free-grammar which essentially means that it is a data center for all types of questions the system can interpret. Different probable questions and different versions of the different probable questions are stored at this core. When a user types a particular ‘natural question’ query in the search box, the parser finds the questions from the grammar that closely matches the users query and displays the results accordingly in a drop down menu. The parsing obviously includes searching for both peoples and entities. When the user decides to click on any of the queries suggested by the parser, the corresponding semantic is resolved in three stages. The three stages are,
- Retrieving the answers of the candidates from an inverted index
- Filtering any information that the searcher cannot access due to the privacy settings
- Order the information according to several criteria that will be most relevant for the searcher. These criteria are determined by an evaluation of the activity of the user on Facebook and in the internet in general.
When this is resolved, the results are displayed.
However, Lars chose not to answer certain questions regarding the Graph Search platform especially those related to the use of graph search in advertising.