KR20050023242A - Displaying paid search listings in proportion to advertiser spending - Google Patents

Abstract

The advertiser 2202 pays to include a search list in a database in which a search result 2212 is provided in response to a query from a searcher 2202, so that an advertiser wants to spend on a search term. Determine the amount. Search provider 2208 displays a list 2212 of advertisers in proportion to the amount each advertiser spends. This allows the advertiser to subscribe to the database search system 2200 and determine how much to pay for the subscription for a period of time. Search provider 2208 may recommend optimal spending 2206 for advertiser 2202.

Description

Display Paid Search List Proportional to Advertiser Spending {DISPLAYING PAID SEARCH LISTINGS IN PROPORTION TO ADVERTISER SPENDING}

<Related application>

This patent application claims the benefit of the filing date of US Provisional Patent Application No. 60 / 369,460, filed April 1, 2002, which is hereby incorporated by reference in its entirety.

A pay-for-placement search engine works like the online version of the yellow page: the user performs a search, and the system displays a list of paying advertisers that match the user's query. Since the actual area of the screen is limited, the user typically sees only some of the lists that match his query. So, to make a viable system, search engine providers need to determine how often to display each list, how much to charge, and the order in which the lists appear.

The current level of technology is Overture's search engine, available at www.overture.com. Overture uses a scheme known as bid-for-rank, which bills advertisers by clicks and orders the list based on the amount each advertiser is willing to pay for each click. The advertiser can bid whatever he wants with a bid of at least 5 cents per click. Many web sites display overture search results, each showing a different number of search results starting at the top of the list, so there is a strong motivation to be near the top. Advertisers with the highest bid always appear, and fewer and fewer advertisers appear as the bid decreases. In theory, the advertiser with the best quality of service bids the best and appears at the top of the list. In reality, the system doesn't work so well. This scheme is simple in theory, but there are a number of problems that frustrate both users and advertisers.

From the user's point of view, the problem is that rank bidding can lead to an unrelated and unwanted search list near the top. Advertisers have absolute control over the order in which the list appears, and clever advertisers can use this freedom to get free exposure at the user's expense. For example, imagine an advertiser selling pet ferrets. Advertisers appear under very specific search terms such as "ferrets" and "pet ferrets" and also under more general terms such as "pets". When a user searches for "pet ferret", it makes sense that the ferret advertiser appears at the top of the list because the user is almost certain that he is looking for what the ferret advertiser is selling. However, for more general terms such as "pet", ferret advertisers do not appear near the top. Users who type "pets" are more likely to find dogs or cats than ferrets.

Unfortunately, if a ferret advertiser is clever, it will generate a list that reads "Outstanding pet ferrets! It's so cheap!" He knows that when a user clicks on this, he thinks it's about ferrets, so he can bid a very high price for this list-the price that brings it to the top of the list. Within the list is enough information the advertiser wants to sell. In a ranking bidding system, an advertiser does not pay for being at the top of the list, but only if the user actually clicks on the list. This list is so specific that a user will only click if he is interested in ferrets, and therefore there is little risk for advertisers even if the list appears under general terms such as "pets." In conclusion, the first search result for "pet" is that the list is relevant only to a small number of users. For everyone else, this list is useless and the overall search experience is poor. This ferret phenomenon is common in PFP search engines. Yellow pages do not have this problem because advertisers must pay based on how much they occupy the page.

From the advertiser's point of view, the problem with ranking bidding is that it is complex and difficult to know what he gets. If an advertiser bids in a particular position, he has no way in advance to calculate how many clicks he will receive, or how much money he will spend, or whether he will have a higher profit in some other positions . Also, other advertisers may come behind with higher bids, so even it may not be certain that he gets the rank he wants. If you have a limited budget, you should keep monitoring your spending so you don't exceed your budget, while keeping enough bids to get the maximum possible number of clicks. Normally, all these uncertainties should be addressed for 50 or 100 different search terms, each with its own bid.

As an example, suppose an advertiser is currently bidding $ 1.00 to be in the second position for the search term "fresh fish". He can stay in the current position, raise the bid to $ 1.20 to rank 1, or lower the bid to $ 0.80 to rank 3. To make this decision, you need to know how many of his clicks are in each of the three rankings. It is almost impossible for him to obtain such information. Given the current state of the art, search engine providers in fact cannot say with certainty that he will get more clicks in rank 1 than rank 3. Even if the advertiser decides that it is reasonable to raise the bid to Rank 1, there is no guarantee that one of the other advertisers will not bid higher than that in a few hours. Or, similarly, an advertiser may raise a bid to rank 1 and find that they are overpaying because the advertisers below it are missing a few days later. The advertiser must continuously monitor his bid and position to get what he wants without overpaying. The current state of the art is to track bids using an electronic bidding agent. Examples are www.gotoast.com, www.did-it.com, and www.pay-per-click-bid-managers.com. However, they only run periodically and are limited in how well they perform and cannot obtain information from search engines such as the number of clicks they need, such as the number of clicks an advertiser wants to receive in a different ranking.

If the advertiser has a fixed budget, he must continuously monitor the expenditure so that he meets the budget. For example, suppose an advertiser has $ 1,000 to spend next month. If he sets the bid to $ 1.00 and 50 users click on the list on the first day, the spending ratio is too high and the bid should be lowered. At $ 50 / day, he will spend his entire budget before the end of the month. Conversely, if only 10 users click on the list on the first day, the spending rate is too low, so the bid should be increased. The interaction between bidding and budgeting is complex and difficult to obtain without constant adjustments. It can also lead to bad search results because instead of looking at the best and most relevant advertisers, the user is just looking at advertisers currently under that budget.

All these problems become even more complicated when an advertiser bids on multiple search terms. Each term receives a different number of searches and clicks and requires different bids. The advertiser must allocate some of the money between them in a way that optimizes the overall profit. There is currently no preferred tool to help advertisers do this, and even the best bidding agent does not attempt to raise or lower bids among multiple terms to match a fixed budget.

Many advertisers prefer an alternative to this system. Sophisticated bidding structures require too much micro-management, which obscures only two problems that an advertiser is really concerned about: the cost he has to pay and what he gets in return. Advertisers can get 1,000 clicks for $ 1,000 next month for search engine providers, or double the number of clicks if they spend twice as much, or half the number of clicks if they spend half I want to say that. This model is much more in line with other kinds of advertisements, such as yellow pages and internet banner advertisements. If the information is extracted, costed and clicked, it is clear that the whole concept of bidding and ranking is unnecessary. Advertisers aren't really interested in which ranking they appear in the search results list. He is interested in how many clicks he will get on the amount he invested and how often these clicks lead to sales. Ideally, he makes this purchase decision based only on this information, the amount he pays, and in return, the number of clicks he receives, and all other details about where and when the list appears are available to the search engine provider. You can leave it. A system based on this idea can provide search engine providers complete freedom in deciding which list to show in response to a user's query, thus making the advertiser much simpler and solving the ferret problem.

1-20 are flowcharts showing detailed algorithms for calculating optimal spending of each advertiser.

21 is a plot showing that the advertiser's overall optimal spending varies as a function of the advertiser.

22 is a block diagram illustrating one embodiment of a network including a computer database search system.

The present embodiments remove bids and rankings while retaining the best features of the ranking bidding scheme, using auctions to automatically set prices and allow advertisers to pick where they should display their search terms. In addition, these embodiments automatically optimize each advertiser's spending on search terms where the advertiser appears, thus eliminating the need for a bid agent.

The idea of these embodiments is simple. Each advertiser determines how much money they want to spend on the search term, and the search provider displays a list of advertisers in proportion to the amount of money each advertiser spends. Suppose there are N advertisers competing for space on a search results page that displays M lists. If there are T searches, the search provider has TM total impressions to distribute to advertisers. Impression is an indication of a search list of one of the search results provided to the searcher. If the amount of money the advertiser i is willing to spend is a _i , the number of impressions he receives is equal to (1).

The number of clicks he can predict is given by Equation 2.

The term r _i is the advertiser's click-through rate, which indicates the probability that the user clicks on the list when it appears in the search.

To distribute the exact number of impressions to everyone, the provider maintains an overall run s _i indicating how many impressions each advertiser receives. When the user performs a search, the provider calculates the difference between this number and the number of impressions the advertiser should receive. This difference And S is the total number of searches that occur. The provider sorts the advertisers by difference and returns the top M. The order of the returned list may be arbitrary, sorted by click-through rate r _i , sorted by the determined quality of the list, or sorted by some other criteria. This algorithm distributes the impression evenly over time, ensuring that the correct number of impressions is received after T full searches.

There is also a limit on the number of impressions advertisers can purchase. In one embodiment, no advertiser may appear more than once in any search result, and therefore may not have an overall impression of T or more. If the total amount of money spent by all advertisers is T, the amount limited to the advertiser is shown in Equation 3.

We refer to this limit as the traffic limit. If the advertiser is above the traffic limit, the search engine provider leaves extra impressions blank. If N ≦ M, there is always at least one advertiser at the traffic limit and the market collapses. In this case, the search engine provider may reduce the number of lists to return so that M <N and the advertisers are motivated to compete. Alternatively, the provider can always set the number of lists as part of the number of advertisers, for example M = 0.5N.

In a preferred embodiment, the advertiser pays for clicks or clickthroughs, not impressions, and the price is confirmed at the point of sale. Click or clickthrough is an action that a searcher selects to view an advertiser's search list and click on its associated hyperlink, or otherwise to view the search list. The searcher's web browser is redirected to the uniform resource locator (URL) associated with the search list. If the advertiser wishes to purchase what he / she clicks on, the search engine provider calculates a cost-per-click based on the estimated number of times the user clicks on the advertiser's list and the total advertisements that are expected to sell. The cost-per-click that the provider charges to be. In this equation, Is an estimate of the search engine provider for the total amount A of the ads the provider intends to sell for T searches, Is an estimate of the click-through rate of the advertiser. The advertiser's CPC goes up as its expected click-through rate decreases. If everyone has the same click-through rate, everyone pays the same CPC. The search engine provider can determine the number of clicks he expects to deliver. Price as To complete the sale, the advertiser determines the amount a _i desired to spend from the displayed CPC. The provider delivers the click at this price until the advertiser reaches the spending limit or the users exit T full searches.

Search engine provider estimated Control the price by adjusting it. If the estimate is as accurate as possible, In the case of = A the highest income. calculation Is too small, the click estimate is too large and the advertiser has not reached individual To spend. In this case, the provider Increase the price by increasing Advertisers tend to spend less, but the total amount the provider makes Goes up. calculation Is too large, the click estimate is too small and the advertiser reaches the full limit A before the end of the T search. In this case, the provider Lower the price by reducing the price. Advertisers tend to spend more, raising the total amount A that providers make. calculation Is correct, each advertiser pays an individual spending limit a _i without any unsold clicks. Similarly, search engine providers are most successful when all estimated click-through rates are as accurate as possible.

One way to make a good estimate of overall spending and click-through rate is to use historical data. For example, suppose the historical click-through rate for the entire market is CTR. Then, the provider We can estimate the click-through rates of individual lists using The term u _i is the number of clicks received by the list and s _i is the number of impressions. For a new list, these numbers are zero and the estimate is initialized to the full click-through rate CTR. For lists that have received a number of clicks and impressions, the estimate approaches the observed frequency. The constant k acts as a scaling factor that determines how quickly the estimate changes between these two values. If the advertiser decides to buy more clicks on the list, the provider estimates the cost per click based on the best estimate r _i so far. The provider can estimate the click-through rate in other ways, for example by comparing the new list with other lists that are historical data.

There are a number of other ways in which a provider executes the process of registering an advertiser. For example, the provider may run a weekly or monthly auction to confirm the amount a _i , or allow the advertiser to continuously adjust it. Such disclosure is not limited to any particular scheme.

Every website that displays search results is different. In the current state of the art, the number of lists varies between 1 and 20, and some web sites display a complete list, while others display only titles. Because of these differences, the preferred embodiment treats all combinations of search terms and web sites as separate markets. An alternative is to treat all the impressions a search term receives anywhere as a single market. In this case, the traffic limit formula is different from equation (3), but the remaining equations and algorithm remain the same. Those skilled in the art have no problem in deriving an appropriate new traffic restriction formula for this alternative formula.

Advertisers typically spend money on many different search terms. One advantage of the present invention is that the provider can recommend the optimal amount the advertiser is spending and can automatically allocate that money among different markets. If an advertiser follows a provider's recommendation, he is guaranteed to maximize the expected benefit. The remainder of this section describes the necessary formulas and algorithms that begin with the case of a single advertiser in a single market and work with all the problems of multiple advertisers in multiple markets.

To optimize an advertiser's spending, a provider can choose from three things: the advertiser's overall budget b _i , the advertiser's profit-per-click p _i , and all advertisers if they spend money elsewhere. Know the available external return rate R. The advertiser's expected profit from buying a paid list in a single market is as shown in Equation 4.

The first term is the advertiser's profit from the advertiser's paid list and the second term is his profit from investing money elsewhere. If the external return rate R is high, the advertiser pays less of his budget for the click. The value of a _i that maximizes its benefit is (5).

The amount d _i is the total amount of money spent by other advertisers in the market, and v _i = r _i p _i is the advertiser's market value. The amount an advertiser spends goes down as its click-through rate or its profit per click decreases. If a _i is less than zero, less than the provider's minimum spend, greater than the traffic limit, or greater than the advertiser's budget, this is limited to an appropriate value.

If there are multiple markets, the optimal solution is to deal with them independently. The advertiser's optimal spending is given by one example of equation (5) for each market. The amounts T, v _i , and d _i may all be different in each market. The only interaction between markets is advertiser budget limits to be. This sum is taken for all markets. To enforce this limit, the advertiser's market value includes a budget scaling factor v _ij = λ _i r _ij p _ij . Note that there is one λ _i that constantly scales the value of the advertiser in all markets. If the advertiser is not budget limited, λ _i = 1. If the advertiser is budget constrained, then lambda _i is set to a value that will spend that budget correctly. Any good numerical equation solver can get this value. Examples include www.ece.nwe.edu/OTC and Numerical Methods for Unconstrained Optimization and Nonlinear Equations , Dennis and Schnabel, ISBN 0898713641, on-line guidance on nonlinear equation solvers.

If there are multiple advertisers in one market, the problem is more complicated. If an advertiser changes the amount he is willing to pay, the optimal solution for other advertisers changes. One possible algorithm is to get iteratively for each advertiser and hope that the iterations converge. If they converge, the final answer is a firm point where no advertiser can increase its profit unless the other advertiser is willing to change the amount they are willing to pay. Using the definition for f _i from equation (4), this determined point satisfies the equation system of equation (6).

As one advertiser changes the amount he spends, the benefit for other advertisers rises, so the determined point is not a comprehensive maximum for any advertiser. Still, this maximizes each advertiser's profits for the amount he can control-his spending-so this is the correct solution to the problem.

The system of Equation 6 can be solved in a closed form. The solution for each advertiser is shown in equation (7).

The total dollar amount A that all advertisers spend together is equal to Equation 8.

Positive v _L is the lowest market value an advertiser can have before Equation 7 becomes negative, and the advertiser does not spend any money. The value is related to the harmonic mean of all advertiser values.

Where the harmonic mean value Is the same as Equation 10.

Another amount of interest is v _U , which is the maximum value an advertiser can have before it becomes a traffic limit. The value is shown in equation (11).

Equations 7-11 are the basic formulas of these embodiments.

When calculating optimal spending using Equation 7, it is possible for some advertisers to have an optimal amount less than zero or traffic limits. These solutions are not possible in the real world. The algorithm for finding a valid solution is to pick one of the advertisers out of bounds, limit it to the appropriate limit, remove it from the problem, and recalculate the solution for everyone else. The recalculated value at each iteration uses a slightly different formula for v _L.

In this equation, m is the number of free spaces on the search results page that are not occupied by advertisers at the traffic limit, and n is the number of free advertisers, which is neither zero limit nor traffic limit. Harmonic mean Is calculated for free advertisers. If all advertisers are free, then m = M and n = N, the solution is the same as the formula in Equation 9 above.

The algorithm for picking up the advertiser to remove sees the total signed excess spend that is negative or exceeds the traffic limit.

If E <0, there are a large number of excessive spending for advertisers below the zero limit, and the algorithm removes the advertiser with the lowest market value. If E≥0, there are a large number of excessive spending for advertisers beyond the traffic limit, and the algorithm limits the advertiser with the maximum market value to the traffic limit. The boundary case E = 0 is determined to add traffic limits to ensure that the market does not collapse into fewer advertisers than the space on the search results page.

If there are multiple markets and multiple advertisers, the optimal solution is to deal with them independently. As with a single advertiser, each advertiser's market value includes a budget scale factor that limits its spending across all markets. If the advertiser is not budget limited, its budget scale factor λ _i = 1. If the advertiser is budget constrained, then λ _i <1. To renumber the advertiser so that the first k is budget bound, the algorithm that satisfies the budget constraint is to solve a simultaneous system of nonlinear equations.

The free variable is λ _i for advertisers at that budget limit. The values of a _ij are given by equations 7 and 12, and are limited by the limits 0 and A _j / M _j . Any good equation solver can solve this system numerically.

There are some practical problems. First, since the advertiser's market value changes as the advertiser's budget scale factor changes, his optimal spending may cross either the zero limit or the traffic limit when solving the system of equation (14). Therefore, even if an advertiser is initially limited in some markets, it is important to always calculate a _ij using equations (7) and (12). By always using the equations, the advertiser's overall optimal spending smoothly varies between zero and its maximum value.

Secondly, since the market limit can be changed when solving for λ _i , the algorithm must be repeated until a stable solution is reached. The algorithm is to first solve for all zero limits and traffic limits in each market, find m and n values, solve budget constraints, and keep m and n values fixed. If any advertiser crosses the zero limit or traffic limit while solving for the budget limit, the process is repeated.

Third, the algorithm must determine which advertiser to include in equation (14) when solving for budget constraints. The algorithm includes any advertiser over budget, and any advertiser with λ _i <1. This combination provides a working set of advertisers that are within their budget limits. This working set may be wrong and some advertisers may actually be below budget even if λ _i = 1. In this case, Equation 14 currently has no solution with the working set. A good equation solver can solve as many budget constraints as possible. In the next iteration, the algorithm removes advertisers that do not belong to the working set and solves equation (14).

Finally, there are areas where the advertiser's overall optimal spending does not change as its budget scale changes. This situation occurs when advertisers are restricted in all markets. 21 illustrates the problem. This figure plots how the advertiser's overall optimal expense varies as a function of its budget scale. Since the advertiser's optimal spending is below the zero limit in all markets, the graph is flat on the left, and since the optimal spending is above the traffic limit in all markets, the graph is flat on the right. These flat areas can cause problems for numerical equation solvers.

The solution is to provide a border to the equation solver limited to two points λ _i 2 min λ _i and max λ _i. These boundaries replace the conventional limits of zero and one. Within a single market, min λ _i is the point at which the advertiser reaches the zero limit and max λ _i is the point at which the advertiser reaches the traffic limit. These values are shown in equation (15).

The value of β is as follows.

β = 0 if the advertiser is limited

/ n the advertiser is free

The desired boundary constraint is the minimum and maximum of these values across all markets. Even with this limitation, the equation solver can encounter a flat area in the middle of the expenditure graph. These flat areas rarely occur when an advertiser is below the zero limit in some markets and above the traffic limit in others. In this case, the algorithm is to restart the equation solver with a value of λ _i outside of the flat area to proceed.

Advertisers can override the automatic optimization and adjust their spending on their own. All equations continue to be valid for the rest of the advertiser except that the formula for v _L is changed. If C is the total amount provided by the advertiser spending a fixed amount, then v _L is as follows.

The amount v _o is the previous zero limit value from (12).

Note that v _L is reduced to v _o when C = 0. Another test of interest is n = 1 and m = M. In this case, v _o = 0, and equation (16) leads back to the original one advertiser result of equation (5).

In one embodiment, the subscription method is implemented for a PFP database search system. The method includes a search provider that provides advertisers with multiple searcher contracts at specific costs. The searcher contract may be an impression, click or clickthrough, post-clickthrough action or some other type of contract. As such, advertisers may select the number or amount of clicks they are willing to pay, and thus subscribers. The subscription is for a set period of time, such as one month, and can be arranged in any mutually agreeable manner. The search provider may provide different ratios for different advertisers, or different ratios depending on the number of search lists or markets in which the advertiser participates.

The method further includes initiating a subscription account with subscribing advertisers. Subscription accounts are paid for by advertisers and subsequently withdrawn automatically as searcher contracts occur.

The PFP database search system subsequently receives a search request from searchers. In response to the search request, the database search system provides the search results. Some search results may be a list of subscribing advertisers when the search query matches the search list. If a subscription advertiser search list is provided as a page of the search list, the subscription account is automatically adjusted. This may be done by counting the number of impressions or clickthroughs paid by the subscribing advertiser in the account and withdrawing one for each search list provided to the searcher. Any other classification of subscription management can be used.

Referring now to the drawings, FIGS. 1-20 form a flow diagram that provides a detailed algorithm for obtaining optimal spending for each advertiser. The input of the algorithm is as follows.

SEARCHES [M]: Number of searches in market M

SPACES [M]: the number of spaces on the search results page

ROM [M]: External Return Rate

PROFIT_PER_CLICK [A, M]: Profit per click for advertiser

CLICK_RATE [A, M]: Advertiser click-through rate

BUDGE [A]: Budget of the advertiser

The output of the algorithm is

SPEND [A, M]: Advertiser's optimal spend

CONSTRAINT [A, M]: Restricted status of the advertiser

LAMBDA [A]: The advertiser's budget scale factor

Throughout the description, variable A always refers to the advertiser and variable M always refers to the market. Since all quantities except BUDGET [A] and LAMBDA [A] depend on the current market M, M is usually omitted in the drawings to improve its readability. Likewise, the figures do not show error conditions or floating point boundary conditions, or opportunities for caching or optimization. Those skilled in the art will have no difficulty in writing efficient computer programs that interpret and implement pseudocode.

The embodiments described herein may be implemented as computer readable program code for operating one or more processing devices and associated data storage equipment. In one particular embodiment, the disclosed methods and apparatus may be implemented as C ++ program code for controlling a database management system or search engine. In other embodiments, the methods and apparatus may be implemented as a data storage medium storing computer readable program code, a data processing apparatus that performs the functions described herein, or any other suitable device.

1 illustrates one embodiment of a higher level method. The method of this embodiment is a loop to solve for zero and traffic limits in each market and then solve the budget limits. The loop terminates when the function IS_SOLVED indicates that the current solution meets all the constraints. In the final step, the algorithm records each advertiser's optimal spending in the output variable SPEND [A, M].

This high level method is a procedure labeled Solve starting at block 100. In block 102, the procedure INITIALIZE_BUDGET_SCALES is called. This procedure initializes the advertiser's budget scale factor, lambda. One embodiment of this procedure is shown in FIG. 9. The loop begins at block 104. In block 106, the looping variable is initialized. In block 108, the procedure SOLVE_MARKET is called. One embodiment of this procedure is shown in FIG. This procedure meets the zero and traffic restrictions in the market. Looping continues at block 110 until all market Ms have been processed. At block 112, a loop is caused and the procedure SOLVE_BUDGET is called. One embodiment of such a procedure is shown in FIG. 10. This procedure satisfies each advertiser's budget limit. The loop comprising blocks 104, 106, 108, 110, 112, and 114 continues processing until the procedure IS_SOLVED returns a true value. One embodiment of this procedure is shown in FIG. 17. This procedure determines whether all zero, traffic and budget restrictions are resolved. Otherwise, control returns to block 104. If so, at block 116, the procedure COMPUTE_SPENDING is called. This procedure calculates the optimal spending for each advertiser. One embodiment of such a procedure is shown in FIG. 18. The procedure Solve then ends at block 118.

2-8 illustrate steps for solving the zero and traffic restrictions. The output is a CONTRAINT [A, M] value that indicates whether the advertiser is zero restricted, traffic restricted or unrestricted in the market. The high level loop is shown in FIG. The algorithm initially initializes all restrictions in the current market to NONE, and then adds the restrictions repeatedly until MARKET_IS_SOLVED indicates that the current solution is valid. During each iteration, the algorithm calls COMPUTE_MARKET_PARAMETERS to calculate the amount of equations 7-12 that characterize the market.

2 is a flow diagram illustrating one embodiment of a procedure SOLVE_MARKET. The procedure begins at block 200. In block 202, the procedure INITIALIZE_CONSTRAINTS is called. One embodiment of this procedure is shown in FIG. 3. This procedure initiates advertiser restrictions in the market. Next, enter the loop at block 204. At block 206, the procedure COMPUTE_MARKET_PARAMETER is called. This procedure calculates the various quantities that make up the market. One embodiment of this procedure is shown in FIG. 4. Also, at block 206, the procedure ADD_CONSTRAINT is called. This procedure adds the most important traffic or zero constant to the market. One embodiment of this procedure is shown in FIG. 6. At block 208, the value returned by procedure MARKET_IS_SOLVED is tested. One embodiment of this procedure is shown in FIG. 8. This procedure determines if all market zero and traffic restrictions are resolved. If this procedure does not return a true value, the looping operation continues. Otherwise, the procedure SOLVE_MARKET ends at block 210.

3 illustrates an algorithm for initializing market constraints. The algorithm is a loop that sets each advertiser's limit to NONE. Procedure INITIALIZE_CONSTRAINTS starts at block 300. The loop enters at block 302. For each advertiser under consideration, at block 304, the value of the array Constraint indexed by the advertiser is initialized to the value NONE. Looping continues until all advertisers are processed at block 306. The procedure ends at block 308.

4 illustrates one embodiment of a method for calculating the various market parameters given by Equations 7-12. The method begins at block 400. At block 402, the procedure COMPUTE_SUMS is called. COMPUTE_SUMS calculates the denominators of the preliminary amounts FREE_SPACES, FREE_ADVERTISERS and Z, i.e., the harmonic mean values given by Equation 10D. These quantities appear in the rest of the formula. The method uses the formula for V_L given by Equation 12, but if some advertisers manually adjust their spending, they use the more general Equation 16 equally well. One embodiment of the procedure COMPUTE_SUMS is shown in FIG. At block 404, the value of V_Bar is calculated and at block 406 the value V_L is calculated. At block 408, the value of the variable TOTAL_MARKET_SPEND is defined. The method ends at block 410.

5 illustrates one embodiment of a procedure for calculating FREE_ADVERTISERS, FREE_SPACES, and Z. FREE_ADVERTISERS counts an unlimited number of advertisers. FREE_SPACES counts the number of free spaces on the search results page that are not occupied by traffic restricted advertisers. Z is the sum of the inverse market values of each free advertiser. The procedure calculates these quantities using a loop for all advertisers.

The procedure begins at block 500. In block 503, FREE_ADVERTISERS, FREE_SPACES and Z are initialized. At block 504, a loop enters using the advertiser as a looping index. At block 606, it is determined whether the restriction for the advertiser is equal to NONE. If so, then at block 508 the values of FREE_ADVERTISERS and Z are incremented. Control then passes to block 514. Otherwise, if the value of Constraint for the advertiser in block 510 is equal to Traffic, then in block 512 the value of Free_Space is reduced. At block 514, the looping operation continues until all advertisers have been processed. The procedure ends at block 516.

6 shows an algorithm that adds the most important traffic or zero limit to the CONTRAINT [A, M] value. If the market is solved, there is nothing to do. Otherwise, the algorithm adds either a zero limit or a traffic limit, depending on the sign of EXCESS_SPEND. If the overspending is negative, the algorithm imposes a zero limit on the free advertiser with the smallest market value. If the overspending is zero or positive, the algorithm adds a traffic limit to the free advertiser with the largest market value.

The procedure begins at block 600. At block 602, the procedure tests whether the market is resolved by calling procedure MARKET_SOLVED. One embodiment of this procedure is shown in FIG. 8. When the market is resolved, the procedure ends and control returns to the calling process. Otherwise, at block 604, the procedure EXCESS_SPEND is called. One embodiment of this procedure is shown in FIG. If the value returned by EXCESS_SPEND is less than zero, then at block 606, the algorithm adds a zero limit to the free advertiser A with the smallest market value. If the value returned by EXCESS_SPEND is zero or positive, then at block 608 the algorithm adds a traffic limit to the free advertiser A with the largest market value. The procedure ends at block 610.

Fig. 7 shows an algorithm for calculating the total signed excess expenditure given by Eq. This maintains the overall execution of the excessive expense and adds to the amount from all advertisers that have a value less than V_L or greater than V_U. The algorithm ignores any advertisers that are zero or traffic limited.

The procedure begins at block 700. In block 702, the value of the variable EXCESS_SPEND is initialized to zero. The looping operation is initiated at block 704 using Advertiser A as the looping index. In a loop, at block 706 it is determined whether the value for the currently indexed advertiser A is less than V_L, the lowest possible free value. If so, at block 708, if there is no restriction on advertiser A, the variable EXCESS_SPEND is increased by the value shown in FIG. 7, block 708. Control then passes to block 714. At block 710, if the value for the currently indexed advertiser A is greater than V_U, the maximum possible free value, control passes to block 712. If there is no limit for advertiser A, the value of EXCESS_SPEND is increased by the value shown in FIG. 7, block 712. FIG. Control then passes to block 714. The looping operation is repeated until all advertisers have been processed. The procedure ends at block 716.

8 shows an algorithm for determining whether the current CONSTRAINT [A, M] value is a valid solution. For each free advertiser, the algorithm checks to see if the value is less than the lowest possible free value V_L or greater than the maximum possible free value V_U. The current solution is only valid if all free advertisers fall within these limits.

The procedure begins at block 800. At block 802, the looping operation is initiated using Advertiser A as an index. At block 804, the value for advertiser A is compared with V_L. If the value is less than V_L, then at block 806, if there is no restriction for advertiser A, the value NO is returned by the procedure. Control passes to block 812. If the value for advertiser A is not less than V_L in block 804, then this value is tested for V_U in block 808. If V_U is exceeded, then at block 810, if there is no limit for advertiser A, the procedure returns the value NO. Control passes to block 812. If additional advertisers remain, the looping operation continues at block 802. If all advertisers were processed without returning the value NO for the procedure, control exits the loop, and at block 814 the procedure returns the value YES, indicating that the market has been resolved. The procedure ends at block 816.

9 illustrates an algorithm for initializing a budget scale factor. The algorithm is a loop that sets the budget scale factor of each advertiser to one. The procedure is called once at the start of the SOLVE algorithm.

The procedure begins at block 900. At block 902, the looping operation enters using advertiser A as a looping index. At block 904, the budget scale factor LAMBDA for the advertiser is initialized with a value of one. At block 906, the looping operation continues until all advertisers have been processed. The procedure ends at block 908.

10-15 illustrate the steps for solving for the budget scale factor LAMBDA [A]. The high level algorithm is shown in FIG. The algorithm starts by calculating an advertiser's working set that is at its budget. The algorithm then generates a variable vector and passes it to the equation solver as one variable for each advertiser in the working set. Each variable has an associated upper and lower bound. The SET_TO_ZERO function is an external equation solver that adjusts LAMDBA [A] so that each advertiser's overall spending exactly matches his budget. Any equation solver works as long as it can solve a system of nonlinear equations with boundary constraints. Inputs to the equation solver are the vector objective function BUDGET_ERROR, the number of variables N, the variable vector to adjust, and the boundary constraints. As it executes, SET_TO_ZERO calls BUDGET_ERROR (I) to evaluate the I-th component of the objective function. Once the solution is found, we exit with BUDGET_ERROR (I) equal to zero for all I's. If there is no solution, terminate with one or more LAMBDA [A] at the upper limit, and the algorithm removes the corresponding advertiser from the working set on the next iteration.

10 illustrates one embodiment of a procedure SOLVE_BUDGETS for solving for a budget scale factor. The procedure begins at block 1000. At block 1002, variable N is initialized to zero and variable WORKING_SET is set equal to the result of procedure BUDGET_LIMITED_ADVERTISERS. One embodiment of this procedure is shown in FIG. 12. At block 1004, the looping operation is initiated using Advertiser A as the looping index. In block 1006, the variable N is incremented and the entry of the vector VARIABLES indexed by the variable N is set equal to the reference to the entry of the vector LAMBDA for the current advertiser. Procedure MIN_LAMBDA is called to determine the value for the variable LOWER_BOUND. MIN_LAMBDA calculates the minimum lambda of the currently indexed advertiser before the advertiser is restricted to zero anywhere. One embodiment of such a procedure is shown in FIG. 13. Then, the procedure MAX_LAMBDA is used to determine the value for the variable UPPER_BOUND. MAX_LAMBDA calculates the maximum lambda of the currently indexed advertiser before the advertiser is restricted from traffic anywhere. One embodiment of this procedure is shown in FIG. 14. At block 1008, the looping operation continues until all advertisers A have been processed.

After completing the looping operation, at block 1010, procedure SET_TO_ZERO is called to solve the mathematical system defined by the input, function BUDGET_ERROR as described above. One embodiment of such a function is shown in FIG. The SOLVE_BUDGETS procedure ends at block 1012.

11 shows a procedure for calculating an amount for an advertiser in a working set that is above or below a budget. The procedure retrieves the I-th advertiser from the working set and returns the difference between its current total spending and its budget limit.

Procedure BUDGET_ERROR begins at block 1100. In block 1102, the I th advertiser is retrieved. In block 1104, the value of BUDGET_ERROR is calculated as the difference between the result returned by procedure TOTAL_SPEND and BUDGET_LIMIT for advertiser A. One embodiment of the procedure TOTAL_SPEND is shown in FIG. 16. The procedure ends at block 1106.

12 illustrates one embodiment of a procedure for calculating a working set of advertisers at their budget limits. The working set includes all advertisers with a LAMBDA [A] that exceeds its budget or is less than the maximum possible value. The algorithm adds to the working set all advertisers that meet one of these two budgetary constraints.

The procedure begins at block 1200. In block 1202, the variable BUDGET_LIMITED_ADVERTISERS is initialized. The looping operation is initialized at block 1204 using Advertiser A as a looping index. In block 1206, it is determined whether the current value of the procedure TOTAL_SPEND for advertiser A exceeds the budget for advertiser A. One embodiment of the procedure TOTAL_SPEND is shown in FIG. 16. If the condition tested in block 1206 is true, then the value of the variable BUDGET_LIMITED_ADVERTISERS is increased by the budget for advertiser A in block 1208. If not, then at block 1210, determine if the lambda value for the currently indexed advertiser is less than the current value of MAX_LAMBDA for advertiser A. One embodiment of the procedure MAX_LAMBDA is shown in FIG. 14. If so, at block 1212 the value of the variable BUDGET_LIMITED_ADVERTISERS is increased by the budget for advertiser A. Otherwise, control passes to block 1214.

At block 1214, control returns to block 1204 if additional advertisers remain. Otherwise, if all advertisers have been processed, the procedure ends at block 1216.

FIG. 13 illustrates one embodiment of a procedure for calculating the minimum LAMBDA [A] an advertiser may have before the advertiser's optimal spending is negative in all markets. For each market, the algorithm calculates the advertiser's minimum LAMBDA [A] using equation (15). For all markets the minimum is the minimum in any market.

The procedure begins at block 1300. At block 1302, the value of variable MIN_LAMBDA is initialized to one. The looping operation enters at block 1304 using market M as the looping index. In blocks 1306 and 1308, Equation 15 is implemented to determine the minimum lambda for the advertiser. At block 1310, control returns to block 1304 if there are additional markets to be processed. Otherwise, the procedure ends at block 1312.

14 shows one procedure for calculating the maximum LAMBDA [A] an advertiser may have before reaching the traffic limit in all markets. For each market, the algorithm calculates the advertiser's maximum LAMBDA [A] using equation (15). The maximum for all markets is the maximum in any market.

The procedure begins at block 1400. At block 1402, the variable MAX_LAMBDA is initialized to zero. The looping operation begins at block 1404 using market M as the looping index. In blocks 1406 and 1408, Equation 15 is implemented to determine the maximum lambda for the advertiser. Looping continues at block 1410 until all markets have been processed. The procedure ends at block 1412.

15 illustrates one embodiment of a procedure BUDGET_ARE_SOLVED to determine if there are any budget constraint violations. The algorithm checks to ensure that all advertisers have a maximum possible value of LAMBDA [A] without exceeding its budget. The algorithm returns YES only if all advertisers satisfy this condition.

The procedure begins at block 1502. The looping starts for each advertiser A in block 1504. At block 1506, the result returned by the procedure TOTAL_SPEND for advertiser A is compared with advertiser A's budget. One embodiment of the procedure TOTAL_SPEND is shown in FIG. 16. If the result is greater than the budget, the procedure returns the value NO at block 1508. Otherwise, if at block 1510 the result returned by procedure TOTAL_SPEND is less than the budget for the advertiser, then at step 1512, if the lambda for the advertiser is less than the value returned by procedure MAX_LAMBDA for the advertiser, the value NO Returned by the procedure. One embodiment of the procedure MAX_LAMBDA is shown in FIG. 15. At block 1512, looping continues until all advertisers have been processed. If no advertiser returns a value NO during iteration through the loop, then at block 1514 the value YES is returned and the procedure ends at block 1516.

FIG. 16 illustrates one embodiment of a procedure TOTAL_SPEND for calculating an advertiser's overall optimal spending for an entire market. It uses OPTIMAL_SPEND (A, M) to calculate the amount in each market and adds the results throughout the run.

The procedure begins at block 1600. At block 1602, the value of variable TOTAL_SPEND is initialized to zero. In the loop containing blocks 1604, 1606, 1608 for each market M, the value of the variable TOTAL_SPEND is increased by the result returned by the procedure OPTIMAL_SPEND. One embodiment of such a procedure is shown in FIG. 19. After all markets have been processed, the procedure ends at block 1610.

FIG. 17 illustrates one embodiment of a procedure IS_SOLVED for determining whether current CONSTRAINT [A, M] and LAMBDA [A] values meet all zero, traffic and budget limits. The algorithm first checks to ensure that all markets meet the zero and traffic limits. The algorithm then checks to ensure that all advertisers meet their budget limits. The algorithm returns YES only if all conditions are true.

The procedure begins at block 1700. At block 1702, a loop enters using market M as a looping variable. At block 1704, the value returned by procedure MARKET_IS_SOLVED is tested. One embodiment of the procedure MARKET_IS_SOLVED is shown in FIG. 8. If the procedure returns a negative value, the procedure IS_SOLVED returns the value NO. Otherwise, at block 1706, the looping continues to test another market M. Once all markets have been tested, the value returned by the procedure BUDGETS_ARE_SOLVED is tested. One embodiment of the procedure BUDGETS_ARE_SOLVED is shown in FIG. 15. If this procedure does not return a positive value, then procedure IS_SOLVED returns the value NO in step 1708. Otherwise, at block 1710, the procedure returns the value YES. The procedure ends at block 1712.

18 illustrates one embodiment of a procedure COMPUTE_SPENDING for recording the final optimal spending amount for each advertiser in each market. The algorithm is a loop that sets each SPEND [A, M] value to the current value of OPTIMAL_SPEND (A, M).

The procedure begins at block 1700. In block 1702, the outer loop enters by using advertiser A as a looping variable. At block 1704, the inner loop enters by using market M as a looping variable. At block 1806, an entry of the array SPEND is set to current values returned by the procedure OPTIMAL_SPEND. One embodiment of such a procedure is shown in FIG. 19. After all markets have been processed for advertiser A's value, the advertiser's value as a looping variable for the referral loop is increased. After all advertisers have been processed, the procedure ends at block 1812.

FIG. 19 illustrates one embodiment of a procedure OPTIMAL_SPEND for calculating an advertiser's optimal spending in a single market using Equation 7. FIG. If the value is less than zero or greater than the traffic limit, the algorithm limits it to an appropriate value.

The procedure begins at block 1900. At block 1902, the value for the variable OPTIMAL_SPEND is determined based on TOTAL_SPEND for the advertiser in the market. One embodiment of the procedure TOTAL_SPEND for this operation is shown in FIG. At block 1904, the result of OPTIMAL_SPEND is set to the greater of OPTIMAL_SPEND and 0 and the minimum of OPTIMAL_SPEND and TARFFIC_LIMIT. The procedure ends at block 1906.

20 illustrates one embodiment of a procedure for calculating an advertiser's market value. The procedure begins at block 2000. The value VALUE is the product of the advertiser's budget scale factor lambda, its profit per click, and its clickthrough rate. The profit per click and clickthrough rate can be obtained from any convenient source.

Therefore, the block diagram of FIG. 22 shows a plurality of client computers 2202, a plurality of advertiser web servers 2204, an account management server 2206, and a search engine web server 2208, all of which are connected to the network 22100. A distributed system 2200 is shown that includes. Network 2210 is hereinafter referred to generally as the Internet. Although the disclosed systems and methods are particularly useful for the Internet, client computer 2202, advertiser web server 2204, account management server 2206, and search engine web server 2208 may be one of many different types of networks. It is obvious that they can be connected together via. Such networks include local area networks (LANs), other wide area networks (WANs), and local networks accessed through telephone lines, such as commercial information services. Client and server processes may include other programs running simultaneously on one computer.

Client computer 2202 may be a conventional personal computer (PC), workstation, or any other sized computer system. Each client 2202 typically includes one or more processors, memory, input / output devices, and network interfaces such as conventional modems. The advertiser web server 2204, the account management server 2206, and the search engine web server 2208 may be configured identically. However, advertiser web server 2204, account management server 2206 and search engine web server 2208 may each include multiple computers connected by separate private networks. Indeed, the network 2210 may include hundreds of thousands of individual computer networks.

Client computer 2202 can execute a web browser program 2212, such as a NAVIGATOR, EXPLORER, or MOSAIC browser program, to locate a web page or record 2214 stored in advertiser server 2204. Browser program 2212 allows a user to enter the address of a particular web page 2214 to be searched. These addresses are referred to as uniform resource locators or URLs. In addition, once one page has been retrieved, the browser program 2212 can provide access to another page or record when the user “clicks” on a hyperlink to another web page. Such hyperlinks are placed within web page 2214 and provide an automated way for the user to enter the URL of another page and retrieve that page. A page may be a content record containing content as plain text information or more complex digitally encoded multimedia content such as software programs, graphics, audio signals, video, and the like.

In one embodiment, client computer 2202 may utilize FTP, SNMP, TELNET, and many other protocols well known in the art, but with the functionality provided by Hypertext Transfer Protocol (HTTP). And communicate over a network 2210 with various network information providers, including an account management server 2206, a search engine server 2208, and an advertiser server 2204. Preferably, search engine server 2208, account management server 2206, and advertiser server 2204 are deployed on the world wide web.

As described above, at least two tie servers are contemplated in this embodiment. A first server contemplated is an account management server 2206 that includes a computer storage medium 2220 and a processing system 2222. The database 2224 is stored on the storage medium 2220 of the account management server 2206. Database 2224 includes advertiser account information, including advertiser subscription account information in one embodiment. It will be understood from the above description that the systems and methods disclosed herein are implemented in software stored as executable instructions on a computer storage medium, such as a memory or mass storage device, on the account management server 2206. Conventional browser program 2212 running on client computer 2202 may be used to access advertiser account information stored in account management server 2206. Preferably, access to the account management server 2206 may be accomplished through an unshown firewall that protects account management and search result batch programs and account information from external changes. Additional security is provided through enhancements to standard communication protocols such as Secure HTTP or Secure Sockets Layer.

The second server type considered is a search engine web server 2208. When a search engine program navigates to a site on another web server that can submit a query to a search engine web server 2208 via a search engine web server URL or browser program 2212, a network user or searchers may search for a keyword query. Typing allows us to identify pages of interest from the millions of pages available on the World Wide Web. In one embodiment, search engine web server 2208 generates a search results list that includes at least partially related entries that are obtained and formatted by the results of the bidding process performed by account management server 2206. Search engine web server 2208 generates a list of hypertext links from client computer 2202 to documents containing information related to the search terms entered by the user. The search engine web server sends this list to the network user in the form of a web page, which is displayed on a browser 2212 running on the client computer 2202. One embodiment of a search engine web server can be found by navigating a web page at the URL http://www.overtue.com .

Search engine web server 2208 is connected to the Internet 2210. In one embodiment, search engine web server 2208 includes a search database 230 that includes a search list record used to generate search results in response to a user query. In addition, the search engine web server 2208 may be connected to the account management server 2206. The account management server 2206 is connected to the internet. Search engine web server 2208 and account management server 2206 meet different information needs of users deployed at client computer 2202.

A class of users located on client computer 2202 may be a network information provider, such as an advertiser with a web site promoter advertising or an advertiser web page 2214 located on an advertiser web server 2204. These advertising web site promoters or advertisers may wish to access account information residing on storage 2220 on account management server 2206. The advertising web site promoter participates in a competitive bidding process with other advertisers through an account residing at the account management server 2206. The advertiser may bid for any number of search terms related to the content of the advertiser's web site. In one embodiment, the relevance of a search term bid in a search list for a corresponding web site is evaluated using a computer program running on the processor 2222 of the account management server 2206, where the computer program is selected. Evaluate search terms and corresponding websites according to a set of editing rules.

When a search is performed using a search term bid by an advertiser, the higher bid on the search result list page generated by the search engine 2208 receives a better position. In one embodiment, the amount bid by the advertiser consists of the amount withdrawn from the advertiser's account each time the advertiser's web site is accessed via a hyperlink on a search results list page. In another embodiment, an advertiser's subscription account is withdrawn a predetermined amount each time an advertiser's search list is served or displayed to a searcher in response to a search query. The searcher "clicks" the hyperlink with the computer input device to initiate the search request and retrieve the information associated with the advertiser's hyperlink. Preferably, the "click" on each access or search result list hyperlink is redirected to search engine web server 2208 to associate the "click" with an account identifier for the advertiser. This redirect action, which is not apparent to the searcher, will access the account identification information coded into the search results page prior to accessing the advertiser's URL using the search result list hyperlink clicked by the searcher. In another embodiment, it is this click-through action to the advertiser's URL that causes a certain amount of withdrawal from the advertiser's subscription account. The account identification information is recorded in the account of the advertiser along with the information from the search request as the search request event. Information obtained through this mechanism conclusively matches an account identifier with a URL in a way that would not be possible using conventional server systems or logs well known in the art, so that accurate account payment records are maintained. In some embodiments, the advertiser's website description and hyperlink on the search results list page is accompanied by an indication that the advertiser's list is a paid list.

The second class of users at client computer 2202 consists of searchers looking for specific information on the web. Searchers access the search engine web page 2232 residing on web server 2208 through its browser 2212. Search engine web page 2232 includes a query box in which the searcher types a search term that includes one or more keywords. Alternatively, the searcher queries the search engine web server 2208 via a query box hyperlinked to the search engine web server 2208 and located on a web page stored on the remote web server. When the searcher has finished entering the search term, the searcher may send a query to the search engine web server 2208 by clicking on the provided hyperlink. The search engine web server 2208 generates a search result list page and sends this page to the searcher of the client computer 2202.

The searcher clicks on the hypertext link associated with each list on the search results page to access the corresponding web page. The hypertext link accesses the web page from anywhere on the internet and includes a paid list to advertiser web page 2214 located on advertiser web server 2204. In one embodiment, the search result list is not placed as a result of advertiser bidding and is not INKTOMI, LYCOS or YAHOO! It includes a free list generated by a conventional world wide web search engine, such as a search engine. Free hypertext links may include links that have been manually indexed in the database 2230 by the editorial team.

From the above description, it can be seen that the present embodiments provide a method and apparatus for displaying a list of advertisers in proportion to the amount spent by the advertiser. Each advertiser decides how much they want to spend on one search term, so the search provider displays a list of advertisers.

Although specific embodiments of the invention have been shown and described, changes may be made. Therefore, it is intended that the appended claims fall within the true spirit and scope of this invention to cover such modifications and variations.

Therefore, the above detailed description should be considered as illustrative rather than limiting, and it is obvious that the following claims and their equivalents are intended to define the spirit and scope of the invention.

Claims (24)

A sales method for a database search system in which searchers submit search queries to a database and receive search lists comprising at least some advertiser-sponsored search listings. Selling a designated amount of searcher engagements to each advertiser for a designated price; And Then responding to search queries and providing search lists of each advertiser in proportion to a specified amount of each advertiser of the searcher contracts. Sales method comprising a. The method of claim 1, wherein selling the designated amount of searcher contracts comprises: Selling searcher impressions of each advertiser's search lists. The method of claim 1, wherein selling the designated amount of searcher contracts comprises: Selling searcher clickthroughs of each advertiser's search lists. 2. The method of claim 1, wherein selling a specified amount of searcher contracts comprises selling post-clickthrough searcher actions at each advertiser's web site after a click through of each advertiser's search list. How to do. 2. The method of claim 1, wherein selling a specified amount of searcher contracts comprises selling one of the searcher impressions and clickthroughs of the respective advertiser's search lists for a predetermined period of time. The method of claim 1, Receiving a search request; For each advertiser, determining each number of searcher agreements received and each difference associated with each number of searcher agreements each advertiser should receive; And Returning search lists according to the determined difference value Sales method comprising more. The method of claim 1, wherein selling the designated amount of searcher contracts comprises: Determining at least one of a cost per click for each advertiser, and an estimated number of clicks for each advertiser. 8. The method of claim 7, further comprising determining a clickthrough rate for each advertiser based on historical data for the market. The method of claim 8, wherein determining the clickthrough rate comprises: Estimating the clickthrough rate as a ratio of the number of clicks received by the search list of each advertiser to the number of impressions received by the search list. The method of claim 1, further comprising: recommending an optimal advertising cost to each advertiser; And Allocating the advertiser advertising cost among different markets. A computer readable medium storing computer readable computer code configured to implement a method for a database search system for which searchers submit search queries to a database and receive search lists comprising at least some advertiser-supported search lists. The computer readable computer code, First computer readable code for providing a specified amount of searcher agreements to each advertiser for a specified price, the searcher agreements comprising at least one of searcher impressions and searcher clickthroughs; And Second computer readable code for responding to subsequent received search queries and for providing each advertiser's search lists in proportion to the specified amount of each advertiser of searcher agreements. Computer-readable storage medium comprising. The method of claim 11, Third program code for determining a cost per clickthrough for each advertiser; And Fourth program code for determining an estimated number of clickthroughs deliverable to each advertiser The computer readable storage medium further comprising. 13. The computer readable storage medium of claim 12, wherein the third computer program code is configured to determine the cost per clickthrough as a function of estimated total ad sales and estimated clickthrough rate for each advertiser. 13. The computer readable storage medium of claim 12, further comprising fifth computer program code configured to determine the clickthrough rate and the estimated total advertisement sales based on historical data for a market. 13. The computer readable storage medium of claim 12, wherein the fourth computer program code is configured to determine the estimated number of clickthroughs as a function of the ratio of each advertiser's advertising cost to an estimated total advertising sale. 12. The computer readable storage medium of claim 11, further comprising third program code for optimizing each advertiser's advertising cost based on the advertiser's overall advertising budget, the advertiser's profit per clickthrough, and an external return rate. . 17. The computer readable storage medium of claim 16, further comprising fourth computer program code for determining a budget scale factor for each advertiser. The computer program code of claim 17 wherein the fourth computer program code comprises: First we solve for all zero constraints and traffic constraints, Determine the number of free spaces on the page of search results provided in response to a search query not occupied by advertisers at the traffic limit, and the number of advertisers that are not currently at the zero limit or the traffic limit, Computer readable storage medium configured to solve a system of equations that determine a budget constraint of each advertiser. An apparatus operating on a computer network to generate a result list in response to a keyword entered through a remote input device by a user, the apparatus comprising: The apparatus comprises a computer system connected to the computer network, A database including a plurality of lists, each list associated with an advertiser; Programming code for generating a result list in response to a user entering a keyword, the result list including lists having associated keywords that generate a match with the keyword entered by the user; And Advertiser subscriptions include accepting advertiser subscription orders that specify a number of user agreements, managing advertiser subscription accounts, and adjusting each advertiser's subscription account when the advertiser list is included in the generated search list. Programming code to manage Device to store it. 20. The apparatus of claim 19, wherein the computer system stores the number of user contracts in response to a reward provided by each advertiser. 21. The computer-readable medium of claim 20, wherein the computer system determines an estimated number of clickthroughs provided in return for rewards provided by each advertiser, wherein the estimated number of clickthroughs is based on the total rewards offered by all advertisers. A device related to a ratio of said reward provided by each said advertiser to said. In a subscription method for a pay-for-placement (PFP) database search system, Providing advertisers with a specified number of searcher contracts at a specified cost; Initiating subscription accounts with one or more affiliate advertisers; Receiving search requests from searchers; In response to the search requests, providing search results including search listings of subscribing advertisers; And Adjusting the subscription accounts of the affiliate advertisers in response to providing the search results Sign up method comprising a. The method of claim 22, Recommending an optimal amount of advertiser cost to each advertiser based on at least one of the advertiser's advertising budget, profit per clickthrough, and external return rate. The method of claim 23, wherein Determining a budget scale factor for each advertiser to limit the advertiser's costs across a plurality of markets.