Increasing Relevancy of Recommendations with Human Pyramid – Part 1

When building a recommendation site, product or service, there are six human layers you can pile on top of pure data-based relevancy. To define the scope of ‘recommendation product’ more clearly, any search service is a recommendation service. Google is recommending websites to the user based on the search term and the page ranks, meta data and inbound links to the relevant websites. If a user gives input to a website and the website then returns output based on that input, then it is a recommendation service. The 6 layers are illustrated in the pyramid below, with the top level being the most accurate, but also usually the hardest to accurately achieve.  In this post, I’ll discuss the bottom 3 layers, while saving the top 3 layers for tomorrow’s post.

Recommendation Pyramid - Personalized, Social, Community

I’ll use the example of a ‘Recent Releases’ dvd movie site to illustrate the 6 pyramid levels. Without the pyramid, the site would show the 90 most recently released dvds, either alphabetically or chronologically. Instead of making the user sift through all 90 titles and make a decision based solely on name, plot, genre, cast and crew, we can add some or all of the layers below to filter the list to the most relevant releases and give the user more data to use when making a decision. This increases the signal:noise ratio, which is an important meta strategy for the web as a whole, as the internet is full of noise and getting noisier each day with tweets and status updates and blogs and microblogs, etc. Filtering through all of the content on the web to find what is relevant and useful to you will be of growing importance and solutions to this problem will be a big part of the next phase of the web.

Experts – The site can use critic reviews of the 90 movies to filter out the poorly rated movies, which perhaps will lower the list from 90 to 50. Of course, the user won’t always agree with the critics, so there will be some duds on the list and also some movies the user would have liked that got filtered out. Rotten Tomatoes is an example of a site that is built around the Experts layer of the pyramid.

Community – The site can allow users to review the movies and then filter out any movies rated below 3 stars by the community.  Wisdom of the Crowds is a very common layer added to websites to improve the relevancy of results.  It can be harder to build out this layer than the previous one since it requires building a community of users that are willing to rate a wide variety of items in your database.  Flixster is an example of a site that is built around the Community layer.   Note: Another method of using Community data for relevance would be to use Twitter or other real-time data streams to give each movie a score based on how much it’s being discussed, and with what sentiment, at the moment.  So the standard ratings and reviews is not the only method of gathering community feedback on data for relevancy purposes.

Friends – Gaining traction lately with the rise of social networks, like the behemoth Facebook, is the use of data from friends of the user.  By layering the user’s social graph over the data set, sites can give what is often more accurate results.  Your friends are more likely to have similar tastes to you than the overall community.    Flixster also connects to Facebook to give you recommendations based on your friends, not just the entire community.

Tags: , , ,

3 Responses to “Increasing Relevancy of Recommendations with Human Pyramid – Part 1”

  1. […] Sweet Spot Strategy User Needs + Competitor Gaps + Our Capability = Competitive Advantage « Increasing Relevancy of Recommendations with Human Pyramid – Part 1 […]

  2. […] This post was mentioned on Twitter by Nicholas P. Nicholas P said: Increasing Relevancy of Recommendations with Human Pyramid – Part 1 […]

  3. You got a really useful blog I have been here reading for about half an hour. I am a newbie and your post is valuable for me.

Leave a Reply