FrontPage SiteMap RecentChanges HowTo Blog

Matching Pages:

RSS

Holy See, Anniversary, Pontifical Ministry of His Holiness Pope John Paul II

MarketBasedAttentionAllocation

You may want to read the page InformationDerivativeMarket for background before reading this page.

Summary: It would be interesting to build a discussion board website which used the results of a market as the moderation algorithm. That is, imagine a Scoop which posted stories to the front page based on a price threshold, rather than a vote threshold.

As discussed on InformationDerivativeMarket, futures markets can be used to efficiently pool communal knowledge to answer questions. Not just questions about the future price of a commodity, but any sort of question. Some kinds of questions that markets have been asked include "who will win the next election?", "what will the weather be on tuesday?", "which movies will sell the most tickets?".

One type question that you may ask the market is, "What are people paying attention to? What will they be paying attention to in the future?". In other words, one could build a SharedAwarenessSystem using a market as the moderation algorithm.

Background: attention allocation in SharedAwarenessSystems

Most large-scale SharedAwarenessSystems contain a filter of some sort. Each individual has a limited amount of time per day with which to learn about new happenings (this quantity is called "attention"). Users want the SharedAwarenessSystem to tell them about events that they need to know; the system isn't very useful if it presents many more items per day than the individual can check out; users would prefer the system to separate the wheat from the chaff.

But, a large SharedAwarenessSystem will have access to orders of magnitude more incoming events per day than any single individual has time to check out. In order to be useful to individuals, most SharedAwarenessSystems filter the stream of incoming events somehow.

For example, SlashDot employs a number of editors who decide which submitted stories are posted. KuroShin allows users to vote on which "submitted stories" are "accepted" onto the front page. These are two ways of doing moderation.

Another way to do moderation would be to set up a futures market in attention. That is, each submitted story would be linked to a virtual stock1. Instead of voting for the story, users would buy the story's stock[[real or fictional currency may be used}}. At some point, the story's "success" would assessed, and the stock would pay out currency to its holders2. The payout criterion should be chosen so that stories which users would be most interested in reading make the most money.

With this system in place, users would be able to use the value of the stocks to decide which events to read about. For example, if a certain news item was the most valuable stock, that would mean that the market predicts that this item will be the most interesting story to the site's users. To filter stories, all you have to do is read off the top n items from the list of top stocks.

Comparison with other moderation systems

One difference between a market-based moderation system and a vote-based moderation system is that, instead of one-vote-per-story, in a market users could spend different amount on different stories according to how strongly they feel that the current market price is in error. This increases the accuracy of the system.

Another difference is that, in a vote-based system, users are free to vote based on whatever criteria they like. In a market-based system, users are rewarded for allocating attention to maximize expected "success".

Choosing a payout criterion

The payout criteria for an AttentionFuturesMarket clearly determines the behavior of the market, and hence must be chosen with care.

Vote-based payout criterion

One simple criterion would be the rating of users who have read the story. In other words, users vote on stories, and this vote is the payoff for the market3. Why not just vote, then?

  1. When users vote, they may be voting according to whether they WANT the story on the front page, rather than a prediction of if everyone ELSE wants the story to go on the front page. With a market+vote system, the vote lets people say what THEY want, and the market forces them to predict what OTHERS want.
  2. Gaming the system may become less of an issue. There would be no need for "meta-moderation" to ensure that people vote in the interests of the community. They can (and should) vote however they like. And, the market mechanism would ensure that people have an incentive to execute their trades in a way that would mirror the community's desire. However, if sign-up is free, there would still be a problem with SockPuppets?. By contrast, if stocks can be bought with real currency, then the rich would have a disproportionate influence on the moderation. There could also be problems with collusion, although any moderation system is vulnerable to that.
  3. As described above, simple one-vote-per-story voting is less accurate than a market mechanism because of the lack of granularity.
  4. If there is a massive amount of submitted stories, then it is likely that good stories would get overlooked because few people would find them in the slush pile and vote for them. A market system provides a strong incentive to seek out undervalued, forgotten-about stories.

Link-based payout criterion

Even better than voting would be a system which simply tracks how useful the story is by monitoring natural4 indicators of success. One AttentionFutures market which does this is BlogShares. BlogShares stocks each represent a WebLog. The value of a stock is tied to the number of incoming hyperlinks to that WebLog. Hence, investors have an incentive to predict whether the incoming hyperlinks to each WebLog will increase or decrease, and by how much.

This enables readers to find important blogs; you can find what the BlogShares market considers important just by reading the most valuable blogs. What is valuable is that "importance" in this context does not mean "how many incoming hyperlinks are there today", but rather, "how many incoming hyperlinks are predicted for the future". For example, an unknown but very high quality blog would be expected to have a relatively high price, even though it currently has few incoming hyperlinks5.

This allows readers to choose the blogs that they want to read in a better way than simply looking at a list of the blogs with the currently greatest numbers of incoming hyperlinks.

Discussion

[new]


BayleShanks
I had this idea while looking at BlogShares; in some sense it's a generalization of the BlogShares idea, applying it to any item to which you may want to pay attention, rather than only to blogs. Perhaps everyone else involved with that project has been thinking along those lines for quite some time. I'd be interested in hearing about related discussions.


CategoryRatingSystem CategoryMarketBased? CategoryWikiTechnology

Footnotes:

1. another term for this is “FinancialDeriviative?
2. either once, like a bond, or periodically, in the form of dividends
3. to prevent a conflict of interest, one might discount the user’s own vote when calculating that user’s payoff
4. or “intrinsic” or “endogenous” indicators
5. this dynamic is dampened by BlogShares’ policy of automatically normalizing the P/E ratios of its stocks when they get very high or very low; a very, very, promising blog would quickly hit the P/E ceiling and have its price artificially lowered. Hence, one would expect BlogShares stock prices to only look a short time into the future, not taking into account potentially dramatic gains and losses

Define external redirect: CategoryMarketBased FinancialDeriviative SockPuppets

EditNearLinks: WebLog KuroShin SlashDot

Languages: