EP1183630A2

EP1183630A2 - Evolving advertisements via an evolutionary algorithm

Info

Publication number: EP1183630A2
Application number: EP00920561A
Authority: EP
Inventors: James D. Schaffer
Original assignee: Koninklijke Philips Electronics NV
Current assignee: Koninklijke Philips NV
Priority date: 1999-03-26
Filing date: 2000-03-22
Publication date: 2002-03-06
Also published as: CN1378672A; WO2000058878A8; JP2003529116A; BR0005552A; KR20010052410A; WO2000058878A2

Abstract

An evolutionary algorithm is used to effect a directed trial and error search of alternative advertisement characteristics. An initial population of sample advertisements is provided, and the characteristics of each advertisement is encoded as a set of genes associated with each member of the population. The effectiveness of each member is assessed, using for example the number of times an Internet user clicks on each advertisement. The members of the population generate plurality of offspring ads that inherit characteristics from their parents. The members of the population that exhibit more effectiveness than others are preferentially selected for offspring generation. By continued preferential selection of parents having more effectiveness than other, the likehood of generating offspring that have a higher degree of effectiveness increases. The propagation of particular characteristics or combinations of characteristics in this evolutionary process provides an indication and verification of those ad characteristics that produce effective results.

Description

j

Evolving Advertisements via an Evolutionary Algorithm.

This invention relates to the field of advertising, and in particular to the use of evolutionary algorithms in the generation and evaluation of alternative advertisements.

In a competitive environment, effective advertising is often the deciding factor for a products commercial success or failure. The demand for a share of a viewer's attention is so great that entire industries, such as the television industry, are funded primarily by advertisers. Web page providers receive advertising revenue by including advertising banners on their sites. Internet services, and computing systems to use these services, are being offered free to users who are willing to allow advertisements ("ads") to appear continuously on their screen. Traditionally, evaluating the effectiveness of an ad campaign is a time consuming, costly, and somewhat inefficient process. Before an ad is publicly released, consumer feedback surveys are conducted to determine viewer preferences and impressions, usually via questionnaires before and after viewing a proposed ad or set of ads. In some cases, the proposed ad is released publicly to a small test market, for continued feedback. After the ad is publicly released, surveys are often conducted to determine the ad's effectiveness on a random sample of possible viewers. An ad's effectiveness is typically assessed with regard to retention, appeal, and any actions taken in response to the ad. The surveys are also conducted periodically, to determine the effectiveness of a continuing ad campaign, and may address the detrimental factors associated with a continually repeated ad, such as viewer boredom or annoyance

The characteristics of an ad, such as content, color, action, placement, duration, and so on, have a significant effect on the effectiveness of the ad. The cause and effect relationship, however, is not easily described or quantified. Correlations are assumed to exist between particular ad characteristics and ad effectiveness, and each advertisement developer uses those characteristics that he or she believes are correlated to potential success. The validity of the assumed Correlations, however, cannot be determined directly, nor can the effects of cross-Correlations. That is, even if an ad is determined to be effective, it is often impossible to determine which particular characteristic of the ad, or combination of characteristics, had the greatest impact on the success, which characteristics had the least impact, which characteπstic would have had an even greater impact if another characteπstic changed, and so on. The development of a better advertisement, and the validation of an ad developer's effectiveness beliefs, therefore, are likely to be a matter of conjecture and guesswork, despite the resources devoted to evaluating and assessing the effectiveness of particular advertisements or particular ad characteristics

It is an object of this invention to provide a method and system for assessing the effectiveness of particular characteristics and combinations of characteristics of advertisements. It is a further object of this invention to provide a method and system for improving the effectiveness of advertisements. It is a further object of this invention to provide a method and system for generating alternative advertisements.

These objects and others are achieved by the use of an evolutionary algoπthm that effects a directed trial and error search of alternative advertisement characteristics. An initial population of sample advertisements is provided, and the characteristics of each advertisement is encoded as a set of genes associated with each member of the population. The effectiveness of each member is assessed, using for example the number of times an Internet user clicks on each advertisement. The members of the population generate plurality of offspπng ads that inherit charactenstics from their parents. The members of the population that exhibit more effectiveness than others are preferentially selected for offspπng generation. By continued preferential selection of parents having more effectiveness than others, the likelihood of generating offspπng that have a higher degree of effectiveness increases. The propagation of particular charactenstics or combinations of characteristics in this evolutionary process provides an indication and veπfication of those ad charactenstics that produce effective results.

The invention is explained in further detail, and by way of example, with reference to the accompanying drawings wherein:

Fig. 1 illustrates an example web page containing a vaπety of advertisements. Fig. 2 illustrates an example evolution of an advertisement in accordance with this invention.

Fig. 3 illustrates an example evolution of advertisements having charactenstics that are encoded as chromosomes of an evolutionary algonthm in accordance with this invention.

Fig. 4 illustrates an example flow diagram for evaluating and evolving advertisements in accordance with this invention.

Fig. 1 illustrates an example web page 100 containing a vaπety of advertisements 110-170. Each advertisement 110-170 has charactenstics that distinguish it from every other advertisements 110-170 Each of these advertisements is vying for a viewer's attention, and the designer of each advertisement has selected charactenstics that he or she believes will catch the viewer's attention better than other charactenstics. For example, the designer of ad 110 chose a heavy bold type, while the designer of ad 120 chose a lighter scnpt type. Heretofore, it would have been difficult to ascertain whether a heavy bold type characteπstic is preferable to a lighter scnpt type charactenstic. The designer of ad 110 also chose different words than the designer of ad 120. Heretofore, it would have been difficult to ascertain whether the text content "Look Here!" in ad 110 is preferable to the text "Buy Now'" in ad 120. If it is determined, in some manner, that ad 120 attracted more viewers than ad 110, it would be difficult to ascertain whether it was the scnpt type charactenstic, the text content characteπstic, or a combination of the two charactenstics that led to ad 120 being more effective. The location of ad 120 may also have affected its effectiveness compared to ad 110 In like manner, ad 160 has a distinctive shape charactenstic that may enhance or diminish its effectiveness. Ad 170 has a revolving banner 170A and an associated figure 170B that may or may not affect its effectiveness in terms of attracting a viewer's attention. Ads 140 and 150 contain no text content, while ad 130 has a predominant figure 130A and diminutive text 130B

The Internet environment provides a unique opportunity to evaluate the effectiveness of advertisements. In accordance with one aspect of this invention, the number of times an advertisement is selected by the viewer is used as a measure of effectiveness of the advertisement, and, based on this measure, the effectiveness of particular charactenstic or combinations of charactenstics are assessed. By its nature, this measure of effectiveness is "noisy", in that it may not truly be a measure of the ad's appeal in all cases. A viewer may, for example, be looking specifically for airline tickets. The ad 170 may be the only ad on the web page 100 that addresses airline tickets, and its selection by the viewer will be unrelated to its attention getting abilities. With millions of potential viewers on the Internet daily, however, the number of times an ad is selected can be assumed to be correlated to its attention-getting effectiveness. If ad 120 is selected significantly more often than ad 100, for example, it would be reasonable to assume that there is something about the different charactenstics of ads 110 and 120 that resulted in this significantly different result If the ads 110 and 120 were each selected significantly more often than any of the other ads 130-170, it would be reasonable to assume that there is something about the charactenstics that are common to ads 110 and 120, such as their rectangular shape and text only charactenstics, that resulted in their better performance The Internet environment also provides a unique opportunity to evaluate alternative advertisement charactenstics. Substantially similar advertisements with different charactenstics can be provided to the aforementioned millions of potential viewers weekly, daily, or even hourly. In many cases, the alternative advertisements can be automatically generated using, for example, a rules or knowledge based system, or a simple algonthm. In accordance with another aspect of this invention, vanations of an advertisement are provided to the potential viewers, and the number of times each vaπation is selected is used as the measure of effectiveness for each vanation. Subsequent vanations are generated based upon the measure of effectiveness of pnor vanations. For example, if one of the charactenstics that is vaned is color, and blue ads exhibit a significantly higher selection than red ads, diffenng shades of blue may be subsequently evaluated, and vanations on different shades of red would not necessanly be evaluated As noted above, however, the correlation between particular charactenstics of the ad, or combinations of charactenstics of the ad, and the ads effectiveness is difficult to determine. In the red/blue example above, it might have been the use of a particular font that caused the red advertisement to perform poorly, and the use of a bolder font in red might be more effective than the use of the same bolder font in blue. A class of algoπthms, termed evolutionary algonthms, have been found to be particularly effective in the determination of the most effective combinations of charactenstics to maximize their effectiveness, without requinng a specific determination of each charactenstic's individual or combmatonal effectiveness. In accordance with this invention, an evolutionary algonthm is used to direct the generation and evaluation of alternative advertisement charactenstics.

Evolutionary algonthms operate via an iterative offspπng production process Evolutionary algoπthms include genetic algoπthms, mutation algoπthms, and the like. In a typical evolutionary algonthm, certain attnbutes, or genes, are assumed to be related to an ability to perform a given task, different combinations of genes resulting in different levels of effectiveness for performing that task. The evolutionary algonthm is particularly effective for problems wherein the relation between the combination of attnbutes and the effectiveness for performing the task does not have a closed form solution.

The offspπng production process is used to determine a particular combination of genes that is most effective for performing a given task, using a directed tπal and error search. A combination of genes, or attnbutes, is termed a chromosome. In the genetic algonthm class of evolutionary algoπthms, a reproduction-recombination cycle is used to propagate generations of offspnng. Members of a population having different chromosomes mate and generate offspπng These offspπng have attnbutes passed down from the parent members, typically as some random combination of genes from each parent. In a classic genetic algorithm, the individuals that are more effective than others in performing the given task are provided a higher opportunity to mate and generate offspring. That is, the individuals having preferred chromosomes are given a higher opportunity to generate offspring, in the hope that the offspring will inherit whichever genes allowed the parents to perform the given task effectively. The next generation of parents are selected based on a preference for those exhibiting effectiveness for performing the given task. In this manner, the number of offspring having attributes that are effective for performing the given task will tend to increase with each generation. Paradigms of other methods of generating offspring, such as asexual reproduction, mutation, and the like, are also used to produce generations of offspring having an increasing likelihood of improved abilities to perform the given task.

In the context of this disclosure, the population consists of member advertisements having chromosomes that reflect different ad characteristics. Some combinations of ad characteristics are more effective for attracting a viewer's attention than other combinations. In accordance with this invention, by generating offspring from the member ads that have chromosomes that are more effective for attracting a viewer's attention, the effectiveness of the offspring ads for attracting viewer attention is likely to increase.

Fig. 2 illustrates an example evolution of an advertisement 201 in accordance with this invention. An offspring of ad 201 is illustrated as ad 211. The offspring ad 211 has characteristics of the parent ad 201, such as its shape, message content, and font style. It differs from the parent ad 211 in the size of the text, and thus could be termed a mutation of the parent ad 211. Ad 212 illustrates an offspring ad that is the combination of characteristics from two parent ads 201 and 291; in this example, the parent ad 291 is assumed to have an italics characteristic, such that the offspring ad 212 inherits its shape, message content, font style, and text size characteristics from its parent 201, and its italics characteristic from its parent 291. As in natural evolution, some of the characteristics of an offspring may differ from both parents, or may be a blending of the characteristics of each parent. For example, ad 292 may have an oval shape, and the offspring 222 of ad 292 (oval) and ad 211 (rectangle) is illustrated in Fig. 2 as having a square shape. Ad 241 illustrates the inheritance of its message content from a parent ad 293, and its square appearance from ad 231, as it was passed down from ad 222 via ad 232. Ad 251 illustrates the inheritance of a graphic message content from ad 294 and a shape from ad 242; although ad 251 is a descendant of the original ad 201, it exhibits few, if any, of the characteristics of ad 201. In accordance with this invention, the likelihood of each charactenstic being passed on from generation to generation is dependent upon the success rate of pπor ads that have this charactenstic, as discussed below

A multitude of evolutionary algoπthms are available that may be employed in accordance with this invention. The CHC Adaptive Search Algonthm has been found to be particularly effective for complex combmatonal engineenng tasks U.S. Patent 5,390,283, "Method for Optimizing the Configuration of a Pick and Place Machine", by Larry J. Eshelman and James D. Schaffer, issued 14 Feb 95, presents the use of the CHC algonthm for determining a near-optimal allocation of components in a "pick and place" machine, and is incorporated herein by reference. As compared to other evolutionary algonthms, the CHC algonthm is a genetic algonthm that employs a "survival of the fittest" selection, wherein only the better performing individuals, whether parent or offspπng, are used to generate subsequent offspnng. To counteract the adverse genealogical effects that such selective survival can introduce, the CHC algonthm avoids incestuous matmgs, matings between individuals having very similar attnbutes. As would be evident to one of ordinary skill in the art, each evolutionary algonthm exhibits pros and cons with respect to the schema used to effect an iterative solution, and the particular choice of evolutionary algonthm for use in this invention is optional. For clanty and ease of understanding, the details of this invention are presented using techniques common to the CHC algonthm, although the use of other evolutionary algoπthms would be evident to one of ordinary skill in the art in the context of this disclosure. In accordance with this invention, a chromosome is defined to contain the characteπstic aspects of advertisements. The charactenstics may be any feature or attnbute that an advertising designer deems relevant to the attention getting abilities of an ad, and may include, for example, colors, shape, texture, content, animation, and so on. The charactenstics may also be indirect representations of sets of charactenstics, or the use of particular design rules and guidelines. Fig. 3A illustrates an example set of charactenstics that consists of the message content 391, the text font 392, the size of the text 393, and whether the text is italicized 394. Illustrated in Fig. 3B are 8 member advertisements 301-308. Each of the member ads 301-308 are charactenzed using these charactenstics 391-394. Fig. 3F contains a key, or mapping, of individual charactenstics to the code used to encode the characteπstic. For example, ad 301 contains the message "Buy Now!"; the message table of Fig. 3F shows a code 381 of "00" associated with the message 381' "Buy Now!". Therefore, the message field 391 of the chromosome 301C corresponding to the ad 301 contains the encoding "00" in Fig. 3A. Ads 302 and 304 contain the message "Look Here!", and are encoded with "01" in the message field 391, because Fig. 3F shows the code 381 of "01" associated with the message 381' "Look Here!". In like manner, the text of ad 301 is presented in a script font. Fig. 3F illustrates the code 382 of "010" corresponds to a font characteristic 381' of "script". Thus, the font field 392 of the chromosome 301C corresponding to the ad 301 contains the code "010". The font size 393 of the ad 301 is encoded as "10", corresponding to a 14 point pitch characteristic, as defined in the mapping 393-393' of Fig. 3F. Each characteristic of the eight example ads 301-308 are similarly encoded.

After each of the ads 301-308 are presented to potential viewers, a measure of effectiveness 301E-308E is determined for each. In a preferred embodiment, the measure of effectiveness is based upon the number of times each ad was selected within a given time period, and is normalized to the total number of selections during that period. In the example of Fig. 3A, the measure of effectiveness is given as the total number of times each ad is selected per thousand total selections. Ads 304 and 302 are illustrated as having the highest (201) 304E and lowest (9) 302E measures of effectiveness, respectively.

In accordance with this invention, the chromosomes 301C-308C of each ad 301-308 are pairwise coupled 350 to produce offspring chromosomes 311C-318C. Using the CHC algorithm, each offspring inherits all of the genes that are common to both parents, and a random selection of the genes that differ in each parent. In the example embodiment, each bit value of each field of the chromosome constitutes a gene. Consider the pairwise coupling of chromosomes 307C and 308C to produce offspring chromosomes 317C and 318C. From left to right, the parents 307C, 308C have identical 1^st, 5^th, and 8^th genes; thus, each of the offspring 317C, 318C have identical 1^st, 5^th, and 8^th genes. A random number of differing genes are switched; in this example, the 4^th 361 and 7^th 362 genes are switched in each offspring 317C, 318C. That is, chromosome 317C is identical to 307C except in the 4^th and 7^th genes, and chromosome 318C is also identical to 308C except in the 4^th and 7^th genes. Each of the underlined gene values in Figs. 3B and 3D indicate a randomly switched gene value.

The offspring chromosome values 311C-318C correspond to new offspring ads 311-318. That is, for example, chromosome 317C (code 01-000-00-0) corresponds to a "Look Here!" Message (code 01), an Ariel Font (code 000), an 8 point Size (code 00), and no Italics (code 0), as illustrated by ad 317 in Fig. 3B. Chromosome 318C (code 00-110-11-0) corresponds to a "Buy Now!" Message (00), a Lucita Font (110), a 12 point Size (11), and no Italics (0), illustrated by ad 318 in Fig. 3B. Each of the other offspring chromosomes are similarly decoded to their corresponding offspring ad. The offspring ads 311-318 are presented to potential viewers, and evaluated for effectiveness in the same manner as the member ads 301-308. The normalized scores 311E- 318E corresponding to each offspring are presented in the dashed boxes of Fig. 3B.

From the collection of the original member ads 301-308 and newly generated offspring ads 311-318, the best performing ads are selected to be parents to the next generation of offspring. Fig. 3C illustrates the selection of the eight best performing ads for subsequent pairwise mating 351. The first ad in Fig. 3C is ad 315, which had an effectiveness rating of 208; the next ad in Fig. 3C is ad 304, which had an effectiveness rating of 201; and so on. Note that the lowest scoring member in Fig. 3C has an effectiveness rating of 73. All prior members and offspring with an effectiveness rating less than 73 are not selected for propagation of offspring. In this manner, assuming that the characteristics that the genes represent are correlated to the ad's effectiveness, each generation is produced from a better performing gene pool, thereby increasing the likelihood of producing a high performing offspring. The ads are ordered in Fig. 3C so as to provide a high degree of diversity between the mating parent pairs. For example, ads 315 and 304 are selected for mating because their chromosomes are substantially different, having only three gene values in common (2^nd, 4^th, and 6^th). Illustrated in Fig. 3D, the offspring chromosome 321C-328C are produced from the members of Fig. 3C, in the same manner as discussed with regard to Fig. 3B. The new offspring ads 321-328 corresponding to the offspring chromosomes 321C-328C are presented to potential viewers, and evaluated for effectiveness in the same manner as the prior ads 301-308, 311-318. The normalized scores 321E-328E corresponding to each offspring are presented in the dashed boxes of Fig. 3D.

From the collection of the parent member ads 315, 304, ..., 318, 311 of Fig. 2C and newly generated offspring ads 321-328 of Fig. 3D, the best performing ads are selected to be parents to the next generation of offspring. Fig. 3E illustrates the selection of the eight best performing ads for subsequent pairwise mating. As can be seen, the average effectiveness of this third generation of parent members is substantially higher than each of the two previous generations. Also note that, as might be expected, the occurrence of ads having a small Size characteristic is rare in this third generation. As can be seen in Fig. 3E, the gene 368 that distinguishes between the smaller (8-10pt) and larger (12-14pt) size text has substantially converged to a value of 1, corresponding to the larger (12-14pt) Size characteristic. Five of the eight high performing chromosomes have a Size characteristic (10) of 14pt; two have a Size characteristic (11) of 12pt; and one has a Size characteristic (01) of lOpt. Note also that the encoding of the Size characteristic uses a Grey-code encoding, wherein adjacent values differ by only one bit. Such an encoding helps to insure that the offspring are near in value to the parent for those characteristics that have an ordered sense, such as size and intensity characteristics. Fig. 4 illustrates an example flow diagram for the evolutionary generation of advertisements to optimize their attention-getting effectiveness. At 410, the characteristics that will be evaluated and varied are identified, as well as the manner in which the characteristics will be encoded as genes of a chromosome. At 420, the initial population of ads is defined; and, at 430, the characteristics of these ads are encoded into chromosomes for future offspring generation. Alternatively, the desired characteristics could be encoded first, then the ads having these characteristics drawn or created automatically.

At 440, the effectiveness of using each ad is determined. As stated above, a common effectiveness measure for the effectiveness of an ad is based on the number of times the ad is selected for further information. Alternative measures would be common to one of ordinary skill in the art in light of this disclosure. For example, the time that a user remains on a page that contains the ad may be used as an indication that the ad has attracted the user's attention, and used in lieu of or in addition to the ad selection measure. In like manner, the number of times a viewer purchased an item via a link through the ad, or the dollar amount of such a purchase, could be used independent of, or in conjunction with, the number of times the ad is selected. In a preferred embodiment, a purchase is given a larger effectiveness value than a mere selection, and the ad's value is based on a sum of the effectiveness values. For a targeted intended audience, the value of the measure of effectiveness may include a weighting factor that depends upon a demographic classification of the viewer that selects the ad. The weighting factor itself may be controlled by a gene or set of genes, as well as other factors that control how demographics influence how ads are generated.

The loop 450-472 performs the evolutionary process. At 450, offspring ad characteristics are generated from the characteristics of the current members of the population. In a preferred embodiment, the CHC algorithm is used. Parent ads having maximum diversity of characteristics are selected to generate a pair of offspring ads, as discussed with regard to Fig. 3C.

Offspring ads having the offspring ad characteristics are generated at 455. This ad generation may be an automated process, a manual process, or a combination of both. For example, a change of font style may be effected automatically, but a change of text size or message content may require more than a direct substitution, due to other constraining factors or artistic considerations. Each of the offspπng ads is evaluated, at 460, using the same measure of effectiveness that was used for the oπginal members at 440. For example, the offspnng ads may be displayed on vaπous web pages for a week, and the number of selections counted for each ad. In an alternative embodiment, to reduce sampling vanations, space for one advertisement is allocated on a selected web page Each time the selected web page is accessed by a viewer, a different offspnng ad is placed in the allocated space. Each offspnng ad's effectiveness is measured by the number of times the ad is selected after a predetermined number of viewer accesses to the web page containing the ad. The predetermined number of accesses is determined based on the degree of evaluation accuracy desired. Taking additional samples reduces the noise associated with the measure and increases the reliability of the measure for companson purposes. Conversely, taking additional samples requires additional evaluation time per generation of ads. In a preferred embodiment, common engmeenng tradeoff analysis techniques and statistical sampling techniques are applied to determine the appropnate evaluation sample size and/or evaluation duration. It is important to note that the evaluation techniques employed must be such that the results of an evaluation of one generation of ads is comparable to the evaluation results of each of the pπor generations of ads.

The selection of the next generation of parent ads is effected at 470. Any offspnng ad that has a better measure of effectiveness than one of any of the parent ads replaces that parent ad, and becomes a parent ad in the next generation, as the program loops back to 450 to generate new offspnng ads. If, at 472, the process has converged, or the process is terminated by, for example, a time-out signal, or a user interrupt, the evolutionary process ceases. Optionally, at 476, the entire process may be repeated, to search along a new evolutionary path In a preferred embodiment, mutations are introduced to all remaining member ads except the best performing member, and the entire process is repeated, via 440. When the search is terminated, at 476, the best performing ad of the remaining member ads is selected as the best solution found, at 490. In a preferred embodiment, the selected best performing ad, or set of better performing ads, is used for subsequent "production" advertising, without the burden of data collection and effectiveness evaluations. Recognizing the transient nature of viewer preferences, and the adverse effects of repetitive exposure, the evolutionary ad generation process of Fig 4 is peπodically repeated, to assure that the identified better performing ads are still performing better, and to potentially improve each ad's effectiveness by introducing changes to alleviate the effects of boredom. j _{

Fig. 5 illustrates an example block diagram of a system for providing evolving advertisements/An advertisement 501 is characterized 510 to form a chromosome 501C. The advertisement 501 is also evaluated 550 to provide a measure of effectiveness 551 associated with the advertisement 501, and correspondingly, the advertisement chromosome 501C. A number of advertisements 501 are similarly processed. The evolutionary algorithm device 540 collects the measure of effectiveness 551 associated with each chromosome 501 C, and produces a next generation chromosome 51 IC, based on the measure of effectiveness of each of the advertisements, as discussed above. The advertisement creator 580, which may be human, machine, or combination of both, creates a next generation ad 511 based on the characteristics of the next generation chromosome 51 IC. The next generation ad 511 replaces the original ad 501, which is the characterize 510 and evaluated 550, as above.

The evaluator 550 includes a presenter 552 that presents the ad 501 to one or more viewers, and a means for determining a user reaction to the ad 501. In a preferred embodiment, the number of times a user selects the ad is counted 554. Optionally, other counters 556 and measuring devices may be coupled to the evaluator 550 to enhance the quality or significance of the measure of effectiveness, as discussed above. The counts from the counters 554, 556 are processed by the measure of effectiveness generator 558 to produce the measure of effectiveness 551.

The foregoing merely illustrates the principles of the invention. It will thus be appreciated that those skilled in the art will be able to devise various arrangements which, although not explicitly described or shown herein, embody the principles of the invention and are thus within its spirit and scope. For example, there exist evolutionary processes that use a cloning process to provide a distribution of members in a population in proportion to an effectiveness measure. That is, for example, if the number of possible ads is fixed, due to the cost of producing each ad, and the number of uses for the ads is large, the better performing ads should be placed in use more often than the poorer performing ads. In this example, the rate at which an ad is replicated at each generation of uses for the ads is determined by the ad's performance during the prior generation. In like manner, non fixed length evolutionary algorithms may be used as well, and other evolutionary algorithm applications for advertising propagation and evaluation will be evident to one of ordinary skill in the art in light of this disclosure, and within the scope of the invention as claimed below.

A 'computer program' is to be understood as any software product stored on a computer-readable medium, downloadable via a network such as the Internet, or marketable in any other manner.

Claims

CLAIMS:

1. A method for developing a preferred advertisement population, comprising the steps of: characterizing each member ad of an ad population as a chromosome having a set of genes, presenting each member ad to one or more viewers, determining an effectiveness measure for each member ad, based on an associated reaction to each member ad from the one or more viewers, - generating a plurality of offspring ads from the ad population, each offspring ad of the plurality of offspring ads being characterized as an offspring chromosome having a set of genes that are based on the set of genes of one or more member ads of the ad population, presenting each offspring ad to one or more viewers, determining an offspring effectiveness measure for each offspring ad, based on an associated reaction to each offspring ad from the one or more viewers, and - forming the preferred advertisement population based on the effectiveness measure of each member ad and the offspring effectiveness measure of each offspring ad.

2. The method of claim 1, wherein the step of generating the plurality of offspring ads is also based on the effectiveness measure of each member ad of the population.

3. The method of claim 1, further including the step of: replacing the ad population with the preferred ad population, and repeating the steps of claim 1.

4. The method of claim 1, wherein: the step of presenting each member ad includes the step of displaying the member ad on a web page as a selectable entity, and the step of determining the effectiveness measure of each member ad includes the step of incrementing a count associated with the member ad whenever the selectable entity is selected.

5. The method of claim 4, wherein: the step of determining the effectiveness measure of each member ad also includes the step of incrementing a second count associated with the member ad whenever the viewer initiates an associated purchase after selecting the selectable entity.

6. The method of claim 1 , wherein the step of generating the plurality of offspring ads includes the step of determining a set of characteristics corresponding to the chromosome of each offspring ad, and developing each offspring ad based on the corresponding set of characteristics.

7. A computer program for developing a preferred population of advertisements, the computer program having an executable form that when executed on a computing device causes the computing device to: present each member ad of an ad population to one or more viewers, determine an effectiveness measure for each member ad, based on an associated reaction to each member ad from the one or more viewers, generate a plurality of offspring ad characteristics from characteristics of the ad population, present each of a plurality of offspring ads corresponding to the plurality of offspring ad characteristics to one or more viewers, - determine an offspring effectiveness measure for each offspring ad, based on an associated reaction to each offspring ad from the one or more viewers, and forming the preferred advertisement population based on the effectiveness measure of each member ad and each offspring ad.

8. The computer program of claim 7, further causing the computing device to: generate at least one of the plurality of offspring ads.

9. The computer program of claim 7, wherein the computing device generates the plurality of offspring ad characteristics based on the effectiveness measure of each member ad of the population.

10. The computer program of claim 7, further causing the computing device to: replace the ad population with the preferred ad population, and repeat the steps of claim 7.

11. The computer program of claim 7, wherein the computing device: - presents each member ad on a web page as a selectable entity, and determines the effectiveness measure of each member ad by incrementing a count associated with the member ad whenever the selectable entity is selected.

12. The computer program of claim 7, wherein the computing device further determines the effectiveness measure of each member ad by incrementing a second count associated with the member ad whenever the viewer initiates an associated purchase after selecting the selectable entity.

13. An evolving ad system comprising: - an evolutionary algorithm device that provides offspring ads that are based on characteristics of parent ads, based on an effectiveness measure associated with each parent ad, and an evaluation device that determines the effectiveness measure associated with each parent ad based on a reaction from one or more viewers to each parent ad.

14. The evolving ad system of claim 13, further including: a means for displaying each parent ad to the one or more viewers.

15. The evolving ad system of claim 13, further including: - an ad creator that creates the offspring adds from offspring characteristics that are based on the characteristics of the parent ads.