To supress evil voters
Posted: 10/10/2003 - 16:01
RKO have some members who want to influence the average scores of tunes to their own liking rather than submit an honest opinion. This has appeared as ugly average scores when only a few votes have been cast, and personal voting records of only bads and outstandings. Or this is how I've understood. I brainstormed some ideas to counter this menace. They are not in any order and can be used alone or together.
1. Reduce the number of choices for a vote. Now six, limit to, say four. This will supress the weight of those members who want to botch the vote.
In the following suggestions, average for ranking purposes means the average score displayed in the long list. Average displayed in the pop-up screen should remain as it is except for 'ranking' field.
2. Currently the average score will be displayed immediately after one vote has been cast. Increase this value so that for ranking purposes, average won't be calculated (or shown) until at least, say five or seven, votes have been cast. Then the heavily variating scores of the first hours will remain hidden.
3. Limit the group of members whose vote is included in the average for ranking purposes. In other words, only the votes from, for example, 100 most or top 20 percent active voters, will be accepted for calculating the average. This way the members with a record of only a few bads won't ruin the rankings.
4. For ranking purposes, use Bayesian or some other estimate instead of simple average. The principle of these formulas is that any number of votes is not the absolute truth but simply a series of random events that hint of what the real score would be. In practice, if the first two votes are bads, the estimate isn't the lowest possible score, but still lower than the combined average of all tunes. And the more votes are cast, the closer to their average the estimate becomes.
1. Reduce the number of choices for a vote. Now six, limit to, say four. This will supress the weight of those members who want to botch the vote.
In the following suggestions, average for ranking purposes means the average score displayed in the long list. Average displayed in the pop-up screen should remain as it is except for 'ranking' field.
2. Currently the average score will be displayed immediately after one vote has been cast. Increase this value so that for ranking purposes, average won't be calculated (or shown) until at least, say five or seven, votes have been cast. Then the heavily variating scores of the first hours will remain hidden.
3. Limit the group of members whose vote is included in the average for ranking purposes. In other words, only the votes from, for example, 100 most or top 20 percent active voters, will be accepted for calculating the average. This way the members with a record of only a few bads won't ruin the rankings.
4. For ranking purposes, use Bayesian or some other estimate instead of simple average. The principle of these formulas is that any number of votes is not the absolute truth but simply a series of random events that hint of what the real score would be. In practice, if the first two votes are bads, the estimate isn't the lowest possible score, but still lower than the combined average of all tunes. And the more votes are cast, the closer to their average the estimate becomes.