Now all that we want to do over here is to find rules by which people are listening to the music. Unerstand that a proper understanding of this behavious of the used will enable us to send appropriate advertisement to the customer. We have used the apiori algorithm to mine the association rules. To know about association rules mining and spriori algorithm please refer to the appendix.

I. TOP 10 Rules (By considering the entire population) - sorted by lift

To find out the rules we have have taken the minimun support and confidence thresholds to be 0.01 and 0.5 respectively and came across 50 rules. Only 10 of these rules are displayed below by the decreasing order of lift.

Interpreting the Rules:

The first rule with a lift of 13.416 is

                {the pussycat dolls}  =>  {rihanna}

The support of the rule is 0.01, i.e. 1% of the total Last Fm world population who has lisened to The Pussycat Doll has also listened to Rihanna.

The confidence of the rule is 0.578. i.e. 58% of the user has listened to Rihanna given that he/she has already listened to The Pussycat Dolls.

The lift is 13.416! This indicated that there is a strong positive association between the choise of these two artists. Infact listening to Rihanna actualy increases the chance of a user listening to the The Pussycat Doll by 13.416 times.

The fifth rule in the list is

                {pink floyd, the doors}  =>  {led zeppelin}

The support of the rule is 0.011, i.e. 1.1% of the total Last Fm world population who has lisened to Pink Floyd and The Doors has also listened to Led Zeppelin.

The confidence of the rule is 0.539. i.e. 53.9% of the user has listened to Led Zeppelin  given that he/she has already listened to Pink Floyd and The Doors.

The lift is 6.802.! This indicated that there is a strong positive association between the choise of these two artists. Infact listening to Pink Floyd and The Doors actualy increases the chance of a user listening to the Led Zeppelin by 6.802 times.

And so on...

II. TOP 10 Rules (For Men) - sorted by lift

We have subset the last fm data by gender and taken the male data into consideration. To find out the rules we have have taken the minimun support and confidence thresholds to be 0.01 and 0.5 respectively and came across 64 rules. Only 10 of these rules are displayed below by the decreasing order of lift.

We can obsrve some ups and dowms of the support, confidence and the lift values for the male last fm population compared to that of the entire population. Some of the rules which were not there in the top 10 are sees to be elevating up and taking up the first 3 placed due to their high lift values. Of course the idea of segregatting the data by gender is to identify the differences in the listening habits due to gender so that approriate target of advertisements can be made to a particular user given the gender. Now let us take a look at the next table showing the top rules for the female population.

III. TOP 10 Rules (For Female) - sorted by lift

We have subset the last fm data by gender and taken the female data into consideration. To find out the rules we have have taken the minimun support and confidence thresholds to be 0.01 and 0.5 respectively and came across 240 rules. Only 10 of these rules are displayed below by the decreasing order of lift.

Its definite! Females are more attracted to the female artists. There is almost no similarities in the top 10 rules for male and female.This identification was important, because if we would have recommended adds based of the rules potrayed by the entire population combined then definitely we would have lost fome information accuracy and lost money. Its not that the rules of the segregated data (segregated by gender) do not capture the the rules obtained for the entire population. Its just that the preference differs. The priority of atrist preference differed by gender.

Based on these captured information we would now figure out a way to suggest recommended artist to the users of last fm.