Facebook’s frictionless sharing has kicked off an interesting discussion about using graphs to filter signal from the ever-increasing noise. There are three types of graphs we learned to use for filtering:

  1.  The Social Graph
  2.  The Social Interest Graph
  3. The Interest Graph

The Social Graph

In its minimal form the Social Graph only contains people (nodes) and the relationships between them (edges). Usually the relationship between nodes is symmetric: both have to agree on being linked to each other. The by far largest social graph existing today is Facebook’s Open Graph, which also includes virtual objects people shared (photos, videos, links, wall entries …) and the relationship with these objects (posted, commented, liked, disliked, read …).

When using the Social Graph as a filter a node will be presented with content other nodes have entered into the graph.  Quantitative measures can be used such as how close a node entering the content is and how many other nodes related to the reader also shared the content.

The Social Interest Graph

While relationships in the Social Graph are symmetric and emerge out of a bi-directional conversation the Social Interest Graph is asymmetric and initiated by one person starting to follow another one. Usually this asymmetric relationship is based on common interests whereby the follower expects to learn relevant content from the person she follows.

Using the Social Interest Graph as a filter delivers more relevant results than the Social Graph as the scope of content is more focused on the follower’s interests (provided the followed person maintains this scope).

The Interest Graph

We define the Interest Graph as the set of relationships one person has to a number of terms she is interested in. These terms can relate to real-world items (car) as well as virtual items (quality) and it’s meaning entirely depends on the person’s individual perception (a person interested in SUV will perceive quality different than a person interested in sports cars). As such the Interest Graph resides within the person itself and is highly individual.

When using the Interest Graph as a filter only content relevant to the individual reader’s interest at that point in time is delivered – independently of the popularity of the content within the Social Graph.

How to Measure Interest

To measure interest we need to give it a price tag. The most precious currency today is attention and therefor the attention a person spends on a term can be used as the measure of relevance. So to identify the terms of interest we need to analyze the content the person has paid attention to.

There are three ways content is analyzed for personalization purposes:

  1. Human Tagging and Categorizing: The writer, an editor or a curator categorizes content and assigns tags based on a set of rules.
  2. Algorithmic Tagging: An algorithm analyzes the content and generates a list of tags. Usually this is based on statistical methods.
  3. Semantic Analysis: An algorithm analyzes the content, detects relationships between terms and uses an ontology to generate a semantic profile of the content.

By nature the first way is labor intensive and inevitably influenced by the categorizer’s perception. This becomes especially problematic when a piece of content can be associated to more than one category. A reader subscribing to one category will miss relevant content put into another one by the categorizer.

The second alternative delivers a deterministic result based on its algorithm. However, missing out on the semantics of the text only terms will be associated explicitly contained in the content. A text on iPad will not be tagged Steve Jobs unless he is explicitly mentioned in there.

The third way will deliver a semantic profile of the content based on an ontology. Searching for iPad will also deliver content on Steve Jobs as the ontology links him to iPad directly and indirectly via Apple.

Human tagging and categorizing alone will not sufficiently reduce noise as there are only so many topics an editor can manage consistently. The second alternative delivers sufficient filtering, but will inevitably create an echo chamber. Semantic analysis delivers better results and is dynamic provided the ontology used for the analysis is maintained.

How to learn the Interest Graph

Lets come back to the iPad example we used before: A reader interested in the term iPad will also be interested in Tablet PC, iOS, iPhone, Apple, Steve Jobs, Android, Google, Samsung, Apps, all the terms directly or indirectly related to iPad. But how do we get to know these relationships?

The answer lies in the content base we have at hand: provided we have enough text to understand the relationships between terms we can build the ontology around iPad from that base. And with the content explosion we witness there is enough text on almost any topic these days free of charge. So all we need to do is aggregate content from trustworthy sources and use it as the base for our ontology, which will then be built in real-time whenever the reader calls for content.

Every new piece of content will influence the ontology and thus make it dynamic: while Amazon’s Kindle used to have a week relationship with iPad via eReader, the new Kindle Fire all of a sudden established a very strong link to it and as a consequence also the relationship between the terms Apple and Amazon was strengthened significantly.

So now that we got our ontology, how do we get to the interest graph? As we stated above attention is the currency: the Interest Graph is build from the content the person paid attention to. Reading an article on iPad will result in an Interest Graph containing iPad and its related terms via the ontology.

Using the Interest Graph for Marketing

One application of the Interest Graph is offering relevant content to a buyer. To do so we score content against his Interest Graph. The more overlap the semantic profile of the content has with the buyer’s Interest Graph the more relevance it carries for him. The interest graph is updated with every new piece of content consumed by the reader. This way our content recommendations follow the reader, click-through rates are improved, spam effects are eliminated and we become the go-to source of relevant, valuable content.

We can also use the Interest Graph for content creation. Running new content against the Interest Graphs we collected will tell us, how the new content will score versus other content and provide feedback to the writer even without having to display the content to anybody. This way alternative drafts can be tested in real-time using the entire recipient base or selected segments of it.

Another application is to use the Interest Graph to detect a buyer’s intent and to automatically drive lead nurturing. Rather than deterministically defining lots of “if-then-else” rules trying to automate marketing we can use the development of hot spots within the buyer’s Interest Graph to trigger action: after the value of a specific term exceeded a threshold we approach him to take our relationship to the next level.

We can also use the Interest Graph to peer buyers with our own resources: if the Interest Graph shows a very narrow, yet very deep profile we are likely to talk to a specialist and want to peer him with one of our own best resources carrying a similar Interest Graph.

These are just a couple of ideas on how to use Interest Graphs within sales and marketing. Managers will have to become familiar with the concept to be able to capitalize on it.


There is no killer approach to Relevance. Henry Nothhaft, Jr., CMO of TrapIt, described it as “the myth of the sweet spot”. The competitive edge will be with services that support multiple discovery methods, multiple filtering approaches, have flexibility, and support multiple mobile platforms.

Mahendra Palsule, The Age Of Relevance, March 3rd, 2011

Marketing will have to use a combination of Social and Interest Graphs to drive lead generation and nurturing. By default using 3rd party Social (Interest) Graphs (Facebook, Google+) will not provide competitive differentiation, as they are available to anybody willing to pay for their usage. In contrast a marketing organization can build her own customers’ Interest Graphs by providing valuable and relevant content and tracking its consumption.

Combined with PLM, CRM and Marketing Automation these Interest Graphs can then become the source of sustainable competitive advantage from product and service definition all the way through aftersales. Especially for B2B marketing Interest Graphs will provide deep insight into buyer’s expectations, decision criteria and buying cycle progress.

(This post was originally posted on Relevancer Blog on Dec. 8, 2011)