We have already seen that the listening behavious of males and females are different and hence they should be targetted separately. For example, if a user is a male and he happend to be listenning to Rihanna then we can target him by an add of Beyoncé, because the rule involving these two artists is associated with a high confidence. However if the user is a female and she happens to be listening to Rihanna she can be targetted by three different artists - Beyoncé, Britney Spears and Lady Gaga because all these artist have high conficence with the artist Rihanna. However it may be that the males who are listening to Rihanna might not be so much interested in Britney Spears and/or Lady Gaga (at least thats what our rules suggests).

Now that we have identified the rules we should be asking two very important questions -

1. Given that a user has listened to a particular artist, say {t.i.}, how probable is he to listen to the other artist/s?

2. Given that a user has already listened to a number of artists how probable is he to listen to a particular artist, say {t.i.}

We are going to study this separately for the males and females in the last fm population. Note that the purpose of this project is not to build a recommendation system but to suggest how a recommendation system can be built using the technique of association rules mining.

A. FOR MALE

Let us see how we figure out how we can find solutions for the two questions listed above when we have a male user. So first we will try to figure out that if the user is a male and he is listening to a music of {t.i.} then what else can we expect him to be interested in in the next instance so that appropriate ads can be recommended to him. The following figure gives the necessary R codes along with the output.

The output lists all possible artists that the male user might like to listen to given that he had been listening to t.i. along with the suppor, confidence and lift. The following figure is also very handy to understand the output.

#install.packages("arulesViz")

library(arulesViz)

plot(rules,method="graph",interactive=TRUE,shading=NA)

The given figure quantifies the association rules of t.i. with the other artists in terms of confidence. The size round orange circles quantifies confidence. This figure clearly potrays that t.i. has the highest confidence with Kayne West. Therefore if the user is a male and he is listening to T.I. then it would be a good idea to recommend him with an add of some new albums of Kayne West. Or if the user had already responded to Kayne West previously then this time we can also recommend him with 50 Cent or Jay-Z. 

Now we will try to answer the second question, i.e. if the artist is a male and had been listening to a number of songs, how likely is he to listen to T.I., because we need to decide if at all that user should be recommended with a T.I. ad.

We can see that if a particular male user had listened to the following set of songs given at the right side of the rules then he would love to listen to songs of T.I. almost with certainty. The corresponding lift  values are also very high indicating there is a high chance of the user to lift a T.I. album if he is recommended the artist given that he has been listening to the artists listed on the left.

B. FOR FEMALE

We can do the same thing for a female user also. So let us look at the solutions of the above two listed questions in case of female users.

The output and the visualization suggests that female user has also a high chance of choosing Kayne West given that she had already listened to T.I. followed by Beyoncé. Note here. This is a change for the females. The next recommendation option for females who are listening to T.I. is Beyoncé where as for males it is 50 Cents.

For the second question the output looks as below. (Only five rules are listed below).

We can see that if a particular female user had listened to the following set of songs given at the right side of the rules then he would love to listen to songs of T.I. almost with certainty. The corresponding lift  values are also very high indicating there is a high chance of the user to lift a T.I. album if he is recommended the artist given that he has been listening to the artists listed on the left.

FINAL NOTES:

This is the idea of using association rules mining for building a recommenders system. We will use association rules mining to understand the listening behaviour of the user and recommend them accordingly. Note that in this project we have segregated the data by gender. However we could also have segregated the data by countries. This would have enabled us to capture the information of the listening tastes of each countries and would enable us to target more effectively. However limitation of time did not permit us to go for that step.