By now you have probably heard of the “social graph.” But what is it exactly? Well, it’s simply a fancy term for all of the connections among people in a social network. It was a term originally introduced by Facebook, but it applies to any system in which people connect with one another, and those connections can be symmetric (e.g., “friending”) or asymmetric (e.g., “following”).


The word “graph” is indicative of the fact that these connections among people can be represented as mathematical constructs derived from the field of graph theory. The individual people in social networks can be represented as nodes in a graph and the connections can be represented as “edges” in graph theory speak.  The connections or edges can have additional attributes, such as distinct directions (e.g., following) and even weightings. Encoded in this convenient form, various insights can be derived. For example, common connections among two or more people can be identified, as well as the well-known degree of separation between any two people. Although more sophisticated network characteristics can be derived from the social graph, even these simple constructs can be quite usefully applied. For example, a system can suggest to you that you connect with someone who is connected to one or more people in common with you. This is the basic approach behind the people suggestions you receive on Facebook, LinkedIn, and Twitter, as well as with many enterprise social platforms.


More recently another graph, the interest graph, has also come to the fore. An interest graph is a set of interest-based relationships between a person and something else, typically a topical area. These interests can be explicitly indicated by the person, but more often are inferred from various user actions such as postings, views, likes, commenting, following, etc. Whether explicitly indicated or implicitly inferred, it is clear that interests are generally a matter of degree, and so the interest relationships are represented as a continuum (typically normalized to a 0-1 range) rather than as a black or white, yes or no, interest relationship.


The interest graph, often in combination with the social graph, enables a whole set of new and very useful functionality. For example, an item of content can be suggested to you based on the specific degree of interest you have with respect to topics that are associated with the item of content. Or another person can be suggested to you, based not only on just your social graph, but also by taking into account a similarity of interests between you and that other person. Most fundamentally, the interest graph enables a capability for automatic personalization, which is pretty much the core of where all user-oriented computing is headed.


There is actually at least one other relationship between people and topical areas besides interests that can be valuably applied, namely expertise. As in the case of interests, the expertise level of a person with respect to a topic can be explicitly indicated by the person or by other people, or it can be implicitly inferred from the person and/or other people’s interactions with the person or the person’s works. Similar to interests, levels of expertise are a matter of degree and are represented as a continuum. The resulting “expertise graph” can be put to work on tasks, such as automatically matching in real-time the people who are predicted to have expertise in one or more areas to the people or projects or jobs that need that expertise.


Just as the social graph is continuously updated as people add or modify connections, the interest and expertise graphs are continuously updated based on people’s actions. These updates are made by machine learning-based algorithms that operate on the social big data that are generated by social networking and other collaborative systems. Taken together, the social, interest, and expertise graphs lie at the very heart of the systems that we can expect to be working with from here on out, whether in the consumer world or in the enterprise—systems that automatically and continuously learn from us, thereby anticipating and delivering to us the knowledge and expertise we need right when we need it.

[guest post from Steve Flinn]