Pinterest.com is an interesting website from the perspective of recommender system. Most of the sites who employs Recommendation engine are e-commerce or textual content based sites but Pinterest analyses the visual picture. The Visual analysis of user activities forms the staple diet of Pinterest’s Recommendation Engine.
Pinterest has an varied user base of people who have interest in creative design. The product can be anything, as long as its visually appealing to the user, he or she can pin the item on his/her Pinterest board.
Pinterest key goal is to collect User ideas based on the Pins collected by its users and then suggest products which can be bought or advertised on Pinterest.
The user can help them by providing a feedback to the products suggested by Pinterest. The feedback will refine the Recommendation Engine which in turn will provide better recommendation.
Pinterest recommendation engine is really a top notch system. The reason that I say that is because of the fact that their suggestion is nearly very similar to the pin you save in their board. The attention to detail is unparalleled , dare I say better than Amazon. Let me give you an example of how fine their recommendation engine works. I have a pinboard where I have collection of wrist watches. Now I like simplicity in design, hence I have collected a collection of watches which are based on Bauhaus design philosophy. So watches from brands like Nomos, Junghans , Junkers are some of the brands which I like and this is where the recommender system of Pinterest shines. It simply does not suggest any wrist watch but watches which are similar to my collection. Here I must mention that I started my collection with just one watch and with this scant data, the recommender system of Pinterest was able to suggest valid items as per my liking. It seems, like mentioned in NYT article, Pinterest’s recommendation engines is inspired by algorithm similar to Collaborative Topic Modelling (CTM) where with just one item its able to suggest a relevant suggestion.
Its tough to suggest recommendation to an system which works so well but there is one area where I think there is some scope of improvement. Although the recommendation system works very well but I think it intentionally ignores the price component of the items collected. Now this could be a non issue since Pinterest is not a conventional e-commerce site but I think person who has a collection of Rolex watches might not be interested in Invicta, although visually both might appear very similar and this is where I feel the recommendation engine of Pinterest needs improvement.