Understanding User Interest Graphs for Personalization

Understanding user interest is a crucial component in enabling personalization in the next generation of data-driven consumer experiences.

Building an explicit model of a user which is both accurate and actionable (meaning it can be easily used to personalize a consumer experience) is hard. Most personalization technologies rely on standard techniques like collaborative filtering and don’t explicitly model users. Collaborative filtering relies on correlations between users to recommend content – the idea being if two users ‘liked’ similar items they should be shown similar items. Collaborative filtering, although relatively easy to implement, has strong limitations. You need user models to reach the true potential of personalizing every online experience.

Generating User Profiles

Below is an example of a user profile generated from a Vidora deployment. Vidora leverages unstructured metadata associated with content like video speech-to-text, keywords, textual semantic analysis, to create explicit models of users. In other words, Vidora starts with an unstructured mess of keywords and content information and transforms that information into an explicit representation of a user’s interests.

In the graph below, we group similar concepts in similar areas. Larger words indicate more interesting areas of the graph for this user. Vidora creates tens of millions of interest graphs on a monthly basis by analyzing millions of unique keywords and billions of behavioral events. These Interest Graphs are very powerful and enable a whole slew of experiences.

Benefits of Explicit Interest Graphs

Here are some of the benefits for an explicit User Interest Graph for personalization –

1. Traveling Profiles – An explicit Interest Graph allows a user to ‘carry’ their interests from one consumer experience to another. Each consumer experience can leverage the same information to personalize the consumer experience. For instance a profile could travel from one Newscorp property to another enabling a better personalization across both – e.g. WSJ to Fox Video App.

2. More Personalized Communication – A profile allows one to speak to the user by leveraging an explicit understanding of their interests. Netflix innovated in this area by ‘motivating’ recommendations – e.g. ‘Because of your interest in dark dramas set in the 1950s we recommend Mad Men’. Netflix is able to engage and communicate with a user more directly. This type of direct communication will only increase moving forward and Interest Graphs will be a key component of enabling this functionality.

3. Better Personalization Performance – Our tests on live deployments have shown that combining explicit models with behavioral-based models increases key metrics like pay-transactions or click-through-rate by over 20%.

4. Integrating Prior Information – A flexible explicit user model allows a company to integrate prior information. For example, existing DMP information is easily integrated to leverage multiple data-sources for better personalization.

5. Contextual Advertising – The advent of mobile in particular is resulting in contextual advertising increasing in importance. Interest Graphs allow for the easy targeting of contextual ads to each user.

More powerful user models will become increasingly important as personalization becomes the norm for every consumer experience. We are in the early innings of creating accurate and actionable models of users – but the technology has the opportunity to truly revolutionize online experiences.