Our Science

Peerset supports a well-known thesis that human interests tend to cluster together. We use statistical knowledge of how human interests are related to pinpoint the consumers most likely to have a deep engagement with a product or service.

Other targeting methods miss potential consumers by not utilizing implicit relationships between consumers’ stated interests and interests that are highly correlated with the advertisements. Peerset broadens targeting opportunities by identifying likely connections between advertisements and users.

On June 8, 2010 a patent was issued to Peerset which describes methods whereby an input (e.g., Duran Duran) produces a set of highly related keywords (e.g., Tears for Fears, Erasure) found in the social web as well as methods used to holistically analyze complex user profiles to provide these keyword sets. We call this core component of Peerset Audience Targeting our Interest Correlation Analyzer (ICA).

Interest Correlation Analyzer (ICA)

ICA leverages the almost limitless mine of information pertaining to people’s interests found in social media in such places as social networking sites, social bookmarking sites, dating sites, blogs, and the like. The technology uses natural language processing and sentiment detection to analyze sources such as these and extracts words that are associated with anonymous yet unique individuals. Words in specific focus are interests, opinions, characteristics and attitudes about things, services, and Brands. ICA then uses this data to train a neural net to recognize which words tend to occur together across many different profiles. It can thereby come to realize that people who express one or a set of words will tend to share an affinity with other words that commonly cluster together.

A simple representation of this at work can be seen in a correlation index indicating the likelihood that people interested in one or a set of words will also be interested in other words. Employing the ICA in this manner provides an extremely powerful tool that can be used to expand on any individually selected interests chosen by an advertiser with additional terms related to the anchor term, as determined by the correlation ratio. Below is an example of a search for interests related to “Nature.”

One can see that given the single word “Nature” the ICA is able to expand this into a large series of words, all of which have been attested in thousands of profiles as occurring together. There is no guesswork here, the system follows well established statistical rules to generate the word expansions. The far right column “correlation” represents a score against an average index, such that for example the second word, “ecology” is 33.46 times more likely to occur in conjunction with the word “nature” than without it.

For the purposes of ad and content recommendation, we employ the ICA in a more complex manner. By creating a vector representing an individual’s position in our ICA’s interest space of over 250,000 interests and characteristics (in our English language database), we can compare the whole person’s profile to potential ad or content matches.

ICA does not retain any personally identifiable information and does not need to track an individual user across multiple domains. It is a purely anonymous and aggregated database reflecting human interests and characteristics.