>From Mahout in Action book, chapter 4.8 "...Another future direction for Mahout is model-based recommendation.... The model might be a probabilistic picture of users’ preferences, in the form of a Bayesian network, for example. The algorithm then attempts to judge the probability of liking an item given its knowledge of all user preferences, and it ranks the recommendations accordingly..."
It looks similar to the bayesian model I presented below for predicting my favorite drink. I think that predicting my favorite drink could be interpreted as a model/content based recommendation similar to : from Mahout in action, chapter 4.7.2: "...For example, if your friend told you she likes the albums Led Zeppelin I, Led Zeppelin II, and Led Zeppelin III, you might well guess she is actually expressing a preference for an attribute of these items: the band Led Zeppelin. By discovering these associations, and discovering attributes of items, it’s possible to construct recommender engines based on these more nuanced understandings of user-item associations..." Pepsi is like a Led Zeppelin, a can of Pepsi you can buy is like a Led Zeppelin record. Taste of drink is like a music genre, e.g. Pepsi and Coke taste similar. Thanks Sean for your comments and your book. They both opened my mind for new ideas. Daniel. 2012/1/19 Sean Owen <[email protected]> > On Thu, Jan 19, 2012 at 12:26 PM, Daniel Korzekwa > <[email protected]> wrote: > > I came up with the same conclusions you did below, it's neither pure > > recommendation nor classification. It sounds similar to clustering with a > > fixed number of classes, e.g. one class per drink and one additional > class > > for non_fan. However I haven't achieved any satisfying results with this > > method yet. > > Maybe I'm oversimplifying what you suggest, but it sounds like you're > defining a "my favorite drink is X" class, where favorite means lots > of purchases. And then classifying people as to whether their favorite > drink is X. But if that means lots of purchases, then this is just a > roundabout way of asking, which drink they purchased a lot, which can > be computed straight from the data. > > > > > The reasoning behind this 'my favorite drink' question is: If significant > > number of customers are fans of particular drink, then maybe it's better > to > > organize shelves (or web shelves) with drinks in a more fans driven way > to > > achieve better selling results?. > > Maybe but this sounds like you're asking one of two much simpler > questions, which don't need machine learning: > > - What is the best-selling drink? so I can emphasize it, etc > - What drink is the 'favorite' (most purchased) for the largest number > of customers? so I can emphasize it, etc > > > > Or maybe if we know that there are many fans of unpopular drink CopaCopa > > and then if we advert it more strongly we may get a big jump in selling. > > Those fans may be customers, who buy more Pepsi than CopaCopa, because > > Pepsi is really well advertised and easily available. Looking only at > > individual customer transactions, it may seem he is a fan of Pepsi, > whereas > > if we look at all transactions for all customers with .e.g. Bayesian > model. > > we can classify more customers as fans of PopoPopa, which may push us to > > advert PopaPopa more aggressively? > > Same, given the way that popular is defined as just a function of > purchases, isn't this just something like looking at purchases per > marketing dollar? Finding the one with the most return on marketing, > under the assumption that the marginal additional marketing spend will > be most effective? > -- Daniel Korzekwa Software Engineer priv: http://danmachine.com blog: http://blog.danmachine.com
