Tell me the company you keep and I'll tell you who you are... and what you want. Web applications are becoming everyday smarter: Amazon guesses books we might like, YouTube proposes us videos of our interest and Facebook finds our friends before us. Nowadays the web is plenty of start-ups using its 2.0 interactive power to create new recommendation-oriented products. Teaching a machine to learn is definitely the new frontier of online implementation.
Nothing to do with education or pedagogy, the machine-learning field of research develops algorithms allowing computer to evolve a statistical-based behavior using external data, just like the one who has made Google's fortunes: a simple search form able to give back a huge amount of filtered results. Developing Recommendation System today means take advantage of the collective use of internet to improve the filtering information system.
Two are the approaches, mixed together, to build People-You-May-Know or Who-Bought-this-Item-also-Bought applications. The older content-based system recommends items similar to those an user has liked in the past, whereas a collaborative filtering identifies user tastes, similar to those of other users, overviewing social activities. Supposing Jack has rated Inception with 8 stars and Source Code with 9 on Internet Movie Database, while John respectively with 6 and 5, there are a few ways to determine how similar their tastes are. The easier one is to calculate an Euclidean Distance with the two movies as axis:
The function returns a value between 0 and 1, where 1 means that two users have identical preferences. Obviously the example is a simplify version but the principle is the same for bigger and different ranking.
All that means that the Facebook I Like button is more than a funny toy. Today targeting people is the first goal to monetize the web revolution. The recent Google suite unification, taking down all redundant data and building a single user account for every services, is an other interesting example and a giant step over this path.
Two are the approaches, mixed together, to build People-You-May-Know or Who-Bought-this-Item-also-Bought applications. The older content-based system recommends items similar to those an user has liked in the past, whereas a collaborative filtering identifies user tastes, similar to those of other users, overviewing social activities. Supposing Jack has rated Inception with 8 stars and Source Code with 9 on Internet Movie Database, while John respectively with 6 and 5, there are a few ways to determine how similar their tastes are. The easier one is to calculate an Euclidean Distance with the two movies as axis:
The function returns a value between 0 and 1, where 1 means that two users have identical preferences. Obviously the example is a simplify version but the principle is the same for bigger and different ranking.
All that means that the Facebook I Like button is more than a funny toy. Today targeting people is the first goal to monetize the web revolution. The recent Google suite unification, taking down all redundant data and building a single user account for every services, is an other interesting example and a giant step over this path.
Previous
