# Popularity algorithm

Hi all,

I have created a small web page for my benefit that will list shortcuts to my favourite websites.  So far so good!  It has a database backend which stores various data including the date and time each time I access a site listed on the page.  The reason for this is that I want to alter the order of the links in each category so that my most popular links appear higher up the list.  Of course I could just count the number of times I have accessed each site but this has two flaws that I can see (perhaps more!):

1.  If I add a new site that I access a lot, it would still appear lower down than some more established, less-popular sites

2. A certain site may become less popular over time but remain nearer the top of the list despite me not accessing so often.

Obviously I could simply count the number of times the sites are accessed in a given period, e.g. over the last 4 weeks, but I was wondering if there is a more efficient algorithm?  I am sure a similar algorithm must exist as this could be used in a wide variety of circumstances, e.g. most played songs in a personal juke-box.

So if you know of any algorithm that solves this, and perhaps gives added weight to more recent 'accesses', and if there are any other issues I should consider, please let me know,

Cheers,

Nev
###### Who is Participating?

Commented:
You have to let things develop over a longer period of time.

Say the initial scores for sites (A,  B,  C)  on Day_1 are  (20, 10, 0)

You never visit A again.  You isit B once  a day.  You visit C twice a day.

The scores will evolve like this.  It takes 8 days for the algorithm to realize that A isn't your favorite any more.
You can make it faster or slower by changeing the .95 up or down.

Day      A      B      C

1      20      10      0
2      19.00      10.50      2.00
3      18.05      10.98      3.90
4      17.15      11.43      5.71
5      16.29      11.85      7.42
6      15.48      12.26      9.05
7      14.70      12.65      10.60
8      13.97      13.02      12.07
9      13.27      13.37      13.46
10      12.60      13.70      14.79
11      11.97      14.01      16.05
12      11.38      14.31      17.25
13      10.81      14.60      18.39
14      10.27      14.87      19.47
15      9.75      15.12      20.49
16      9.27      15.37      21.47
17      8.80      15.60      22.39
18      8.36      15.82      23.28
19      7.94      16.03      24.11
20      7.55      16.23      24.91
21      7.17      16.42      25.66
22      6.81      16.59      26.38
23      6.47      16.76      27.06
24      6.15      16.93      27.71
25      5.84      17.08      28.32
26      5.55      17.23      28.90
27      5.27      17.36      29.46
28      5.01      17.50      29.99
29      4.76      17.62      30.49
30      4.52      17.74      30.96
0

Commented:
You could increment the popularity score by one every time you visit a site.

And then let the score decay over time:  Once a day (or once  a week) multiply all the scores by 0.95 (or some other fraction).

That would give you a weighted average with a memory of 20 days (or 20 weeks).      0.95  =  1 - 1/20
0

Author Commented:
Thanks for that, it makes a lot of sense.  But can I run an example by you.  There are 2 sites initially, namely S1 and S2.  After one day their scores are as follows:

S1 - 20
S2 - 40

They are multiplied by 0.95 to get:
S1 - 19
S2 - 38

This continues but the scores are still cumulative so although this seems fair for sites that are added at the same time, it does not allow for newly added sites.

This is certainly along the right lines but I think it needs a little modification.

Cheers,

Nev
0

Author Commented:
Thanks for that, I was being lazy by not going further with my example!  That solution will work a treat!
0

Commented:
Already finished but got another solution you could consider:

Add not 1 for each visit but
exp (- (days since visit)).
The exponent will be ~1 for recently visited sites and going towards 0 for visits long ago.
0

Commented:
SteH's solution is pretty good.       For one thing it updates automatically.  So you don't have to do the daily decay adjustment.

It also saves information long term.  With my solution all the sites you stop visiting eventually
windup tied at zero.

Try simulating the algorithm in Excel.  Some combination of the two might work out best.

0

Author Commented:
Thanks to you both, I have now got a good system that seems to work a treat.

Is it possible to give SteH some points for your contribution?
0

Commented:
You can post a new question "points for SteH" were you refer to this one. Only requirement of EE for this is that the points for this Q and the new Q added are 500 or less.
0
Question has a verified solution.

Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.

Have a better answer? Share it in a comment.