Compilation Techniques

For the compilation of the Internet Entertainment Charts a set of rules and algorithms is used. Each one has its own purpose, and they are all described here.
Step 1	Exponential points for the previous week
	The chart from the previous week is always used as a starting point from which the next chart is being compiled. Titles get points corresponding to their position in this previous chart. These points are exponential. The lowest entry in the chart gets 1 point. Every other title gets 1.05 times the number of points of the one below. So in a Top 100 list we get these figures:
	number 1: 125.2393 points number 2: 119.2755 points number 3: 113.5957 points ... number 97: 1.1576 points number 98: 1.1025 points number 99: 1.0500 points number 100: 1.0000 points
You can see that the difference between the points of high positioned titles is much bigger than the difference between those in the lower regions of the chart. This makes sure that titles can climb and drop rapidly in the lower regions, but have to fight a much tougher battle at the top of the list. The intervals are bigger at the top, and therefore harder to overtake.
	The higher the position of a title on the list, the harder it is to suddenly drop from the chart. Even if a title suddenly gets very few points from voters in one week, which happens every so often which such small numbers of voters, it will still have its 'bonus' for being high on last weeks' list. This means that sudden weird behaviour of titles in the charts, due to big but illogical fluctuations in the email votes, is minimized.
Step 2	Counting the votes
	The points sent by email in the last 5 weeks are summed up for each title. The weight of these votes decreases over time. Votes that came in for the current week are used at 100%; votes that are one week old are used for 80%, then 60%, 40% and votes from 4 weeks ago are used with a weight of 20%. Only the most recent votes from each person are being used. So, only if someone hasn't voted for three weeks, then his votes from three weeks ago are being used at 40%.
	In the first three years of the Games Charts we used the votes from four weeks (not five), all at 100%. A bonus was given to new games to achieve a same kind of effect, in a much less elegant way.
You can see that the emphasis is on the newest votes, which helps the charts to react to new hits as quickly as possible. Giving exponentional points to the titles in last weeks' chart, and using votes from the last five weeks, would make a chart slow and not responsive enough to hot new titles. By giving the most weight to the newest votes, we are able to counterweight this effect.
Step 3	Limiting the oldest titles
	History has proven that people like to vote for their most favourite titles ever. The Internet Entertainment Charts are not supposed to be all-time favourites lists, but weekly updates of the current most popular titles in the world. Of course, a title can still be popular long after it has been commercially available. The charts should reflect that, and they do. But on the other hand, as long as old titles keep occupying high positions, new titles will suffer from it. Therefore, once a title has proven its massive popularity over time, it will be forced to retreat, very gently.
When a title has been in the list for a certain number of weeks (different for each chart), a mechanism will start that reduces the points from voters little by little. Over a period of many weeks (or even years) the points from voters for this title will be reduced to zero.
	This helps to make a little bit more room near the top for newer titles that the readers are most interested in, while no title is expected to last many years anyway, so we do not really force a title out of the list. The parameters may be adjusted whenever we feel it becomes necessary.
	For the Games Charts another more drastic method was used, where people could vote for the same title for only 26 weeks. This method was abolished when it got too time-consuming for our software.
Step 4	Normalizing all the points and adding them up
	We now have two sets of points: one set of exponential points from Step 1, and one set of votes points from Steps 2+3. Since you can't add 'apples' to 'pears' without normalizing them to 'fruit' first, we have to do some normalization first. This means that all points are multiplied equally, in such a way that the total of all these points is a constant number, used every week.
The exponential points are normalized to a certain total, and the votes points are also normalized to a certain total. Only then they are added up. This ensures that the number of votes received in a week has no influence on how much changes there are in the chart. More votes results in a more accurate chart, but not in a chart with more movement. The exponential points and votes points are therefore always used in the same ratio, every week.
	You can now clearly see the effect of votes on the titles in the charts: When a title gets few or no points from voters, it will automatically drop in the new chart. When it gets many points by voters (compared to the titles around it) it will climb. Votes determine how titles will move in the list, not what their absolute positions will be.
Step 5	Limiting the number of re-entries
	With the set of methods explained above, we have accounted for all the effects of having only a small numbers of voters for big weekly lists. One effect remains however, which is the ease with which titles are able to re-enter the charts again and again. The titles at the bottom of the list, or just below it, have such small numbers of voters, that sudden jumps are very common. We feel that once a title has dropped out of the list, it should only return when necessary.
However, instead of checking each re-entry with logical factors, we decided to just limit the number of re-entries. Only one or two titles can re-appear. The least powerful re-entries will be removed and will have to try again in the next week. (History proves that this rule has no effect most of the time. It only needs to do its work in an occasional quiet week, when there are suddenly several re-entries, for a lack of real new titles.)
Conclusion
	As you should have noticed, it is not the case that the title that gets the most points from our voters, will automatically be number one. Only when it already has a very high position it could be number one in the new chart. Otherwise it will have prove itself, and do serious climbing first, which will automatically be the case when it gets many points. This compilation system introduces a chart that is robust. Titles can only reach the top when they get many points for several weeks in a row. The more usual 'most points is highest position' system would cause very strange things to happen, when the number of voters is limited, like in this case. Another advantage of the adopted system is that it is more interesting. Voters can carefully watch the movements of their favourite titles, and influence the movements by balancing their votes. Of course, the influence of a single person on the chart decreases, when the number of voters increases.
Several comments
	Each chart needs a previous list to start with and to give exponential points to. But with the first edition of a new chart, we don't have a previous week. For that reason we compile a list that is used as the 'initial list' in week 0. For the Games Charts we compiled a list from many magazine review percentages, and retail charts. For the Albums Charts we used a mixture of the most important albums retail charts in the world. For the Movies Charts we made an initial list from Box Office lists and movie ratings from Internet people. Such initial lists are needed to get the algorithms and the chart rolling. They are also needed because the number of voters needs to grow to a reasonable amount, and it takes time to achieve that. Normally, after one or two months, the effects of having one initial (non-voters) list in week 0 will have died out, and the chart will live on its own, fed by voters only.
The points as showed in the last column of the published charts are a simple total of all the votes that have been received over the last five weeks. These points have not been subject to reduction or normalization. This means that you can see some strange things happening every now and then. For example, a title can enter the chart at a position above a second title that seems to have more points. This happens when the new title received all or most of its points in the last weeks, while many of the second title's points have been received in previous weeks, and have therefore been reduced (which is not visible in the published chart). For people who understand the way the lists are made, these unmodified points can be an interesting indication of what will happen in the next few weeks. A title that shows many more points in the last column of the list than many other titles above it, is expected to climb drastically in the next week.
	Only votes for the current week are stored in our database. When we make a chart, the program scans for votes in the previous weeks for people that have not yet voted again. These votes are used at a lesser weight. For year-end lists and other periodical totals, only the votes that are actually stored in the database are being added up. This means that someone who votes every week, will have four times as much influence on the year-end lists than someone who only votes every month.
	This was not the case for the Games Charts in the first two years, when votes from people were actually stored four times in the database, for four consecutive weeks. We had to abolish this way of storing votes, to reduce the size of the database and to be more flexible.
And that's it
	We hope you understand the methods we use for the compilation and why we use them. We feel that, if we would just add up votes and present them as they came in, the chart would suggest false information, indicate trends where there aren't any, and would just confuse people and be uninteresting to follow weekly. We hope that the algorithms make the charts look like how they would actually be if we had 1,000,000 voters. :)