Collaborative Filtering

There are several things we buy or consume on an everyday basis, which are based on our individual tastes - the songs we listen to, the movies we watch, the TV channels we subscribe to, the books we read, to name a few.

Since most of our research and discovery these days is online, it sometimes gets overwhelming to pick out a book or a movie that you would really enjoy from the endless options available today. Very often, we end up going with the 'safe choice' because it is simpler, less time-consuming, and the risk of investing time and money over a bad choice is minimized. Think about this: if you like reading books and visit Flipkart or Amazon to buy a book, how often do you simply search for books by well known authors, or authors whose books you have liked before? Pretty much most of the time, isn't it? This is where we might end up ignoring a lot of really well written books, just because the author is not well known. A similar thing happens to low budget movies with little known actors. Sometimes, we realize, much later, that a movie which flopped, was actually a very good one. Jaane Bhi Do Yaaron, Andaz Apna Apna and Bheja Fry are some examples that immediately come to mind.

Now imagine this: You visit a website to buy a book. The website has 'profiled' you  basis your past purchases and books that you have reviewed or rated on the website. Like you, there are several others who have bought or reviewed books on the same site, and the site has profiled them too. Based on people with similar profiles, the site now knows people who have a similar taste as you in books, and it knows what other books those people have read or liked. If those books are recommended to you, there's a good chance you would actually find them interesting. The site may need a significant sample size of books you have liked or reviewed before profiling you accurately and showing you really relevant recommendations, but once you are profiled accurately, the recommendations would really make sense. This way, you need not stick to the tried & tested authors, but can explore several new books and authors who might actually be creating good stuff. This also reduces the disadvantage that a lesser known artist or author has as compared to a bestselling author or a popular musician.

This process is called 'Collaborative Filtering'. Back in the old days, when we visited the neighbourhood video rental store, the person there usually suggested a movie we could rent, knowing what kind of movies we usually watch. He would recommend a completely different set of movies to our neighbours or friends. Collaborative Filtering takes this personalization to a much larger scale, allowing us to be profiled with people all over the world. 

There are several possible opportunities to exploit this capability. As our world gets more and more 'social networked', anything that can be customized can actually be marketed to a target audience. In the next decade, I see several online businesses centred around this - Vacations, Gifting, Beauty Parlours, Live Events, Jewellery, Hotels, Restaurants, etc.