Advertisement display depth optimization to maximize click activity page yield

• US 7,831,456 B2
• Filed: 11/21/2007
• Issued: 11/09/2010
• Est. Priority Date: 11/21/2007
• Status: Expired due to Fees

• Current Assignee: R2 Solutions LLC (Acacia Research Corporation)
• Original Assignee: Yahoo! Inc. (Apollo Global Management, Inc.)
• Inventors: Demir, Ramazan, Tomak, Kerem, Jin, Jiaqi, Chaudhury, Supratim Roy
• Primary Examiner(s): DIXON, THOMAS A
• Application Number:
  US11/944,089
• Publication Number:
  US 20090132340A1
• Time in Patent Office: 1,084 Days
• Field of Search: 705/7
• US Class Current: 705/14.45
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0201 Market modelling; Market analysis; Collecting market data
• CPC Class Codes: G06Q 30/0246 Traffic

Abstract

A method for maximizing page yield in advertisement display to Web users includes tracking click activity associated with advertisements displayed to users through a hierarchal set of search results pages associated with a keyword; tracking bidding activity by advertisers related to the keyword and corresponding advertisements of the advertisements; applying at least one component model to the click or bidding activity as correlated with a particular search results page of the hierarchal set of search results pages via which the advertisements receiving the click activity are displayed, wherein the at least one component model generates a set of output values representative of the tracked click activity; and providing the output values to the at least one component model to iteratively optimize a number of advertisements displayed on the particular search results page via which at least some of the advertisements are displayed in response to a query for the keyword.

Claims

1. A method for maximizing page yield in the display of advertisements to searching Web users, the method executed by a server having a processor and memory, the method comprising:

tracking click activity, by the server, associated with a plurality of advertisements displayed to a plurality of users through a hierarchal set of search results pages associated with a keyword that are delivered by a search engine;

tracking bidding activity, by the server, of a plurality of advertisers related to the keyword and corresponding advertisements of the plurality of advertisements;

applying, by the server, at least one component model to the click or bidding activity as correlated with a particular search results page of the hierarchal set of search results pages via which the advertisements receiving the click activity are displayed, wherein the at least one component model generates a set of output values representative of the tracked click activity; and

providing, by the server, the output values to the at least one component model to iteratively optimize a number of advertisements displayed on the particular search results page via which at least some of the advertisements are displayed in response to a query for the keyword.

2. The method of claim 1, wherein the plurality of advertisements comprise sponsored advertisements.

3. The method of claim 1, wherein the number of advertisements displayed on the particular search results page represents a display depth, wherein display depths for each search results page of the set of search results pages may be optimized through the at least one component model.

4. The method of claim 3, wherein a level of an advertisement within the hierarchal set of search results pages comprises a rank, and wherein the at least one component model comprises a depth elasticity of click rank distribution that defines the probability of a click at a given rank-depth through the hierarchal set of search results pages.

5. The method of claim 4, wherein a click rank-depth is defined by C_r,d and its elasticity is given by

6. The method of claim 4, wherein applying the depth elasticity of the click rank distribution model comprises indexing the plurality of users to differentiate clicking activity between the plurality of users at different rank-depths of the hierarchal set of search results pages.

7. The method of claim 6, further comprising:

indexing the plurality of users according to at least one of the following dimensions;

intent of query;

those that like advertisements versus those that do not;

or those that tend to explore advertisements versus those that do not.

8. The method of claim 7, further comprising:

storing the indexed values associated with the plurality of users;

estimating an index probability function on a user by sampling from the stored index values in response to a query on the keyword; and

serving advertisements at a calculated rank-depth based on the estimated index probability function.

9. The method of claim 3, wherein the at least one component model comprises an advertiser class model, the method further comprising:

segmenting a plurality of advertisers that own the plurality of respective advertisements into at least one of a plurality of advertiser classes that depend at least on bid price elasticity.

10. The method of claim 9, wherein the plurality of advertiser classes comprise at least one of market maker, market follower, and market neutral.

11. The method of claim 9, further comprising:

measuring membership in an advertiser class by bid-click volume elasticity curves that measure willingness of advertisers to pay up to a click share or a volume change.

12. The method of claim 11, wherein measuring membership in an advertiser class is based on a mixture of features selected from the group consisting of an average bid, an average quality score, an average spend amount, and an average display rank.

13. The method of claim 3, wherein the output comprises a look-up table including the keyword and display depth for a specific date and time.

14. A method for maximizing page yield by optimizing display depth of advertisements displayed to searching Web users, the method executed by a server having a processor and memory, the method comprising:

(a) tracking click activity, by the server, on a plurality of advertisements by a plurality of users via a set of hierarchal search results pages delivered in response to a search query for a keyword by a search engine;

(b) tracking bidding activity, by the server, of a plurality of advertisers related to the keyword and corresponding advertisements of the plurality of advertisements; and

(c) calculating, by the processor, an optimum number of advertisements (d*) to display on a particular search results page of the set of hierarchal search results pages in response to the search query such that maximizes the expected page yield according to at least one business objective, wherein the expected page yield of the particular search results page is calculated by summing, at each of a plurality of display depths within d*, a probability of a click at the display depth times a bid price estimate at the display depth, wherein the probability of click and the bid price estimate are obtained from sampled related curves formed from the tracked click and bidding activities based on at least one query;

(d) executing, by the processor, a Monte Carlo randomization procedure to include;

(i) randomly sampling probability and bid/price curves related at least one different query;

(ii) recalculating d* as per step (c); and

(iii) repeating steps (i) and (ii) to calculate potential multiple optimum numbers (d*) of advertisements; and

(e) serving, by the server, to the particular search results page at least some of the plurality of advertisements according to at least one calculated value of d* in response to the search query for the keyword.

15. The method of claim 14, wherein the at least one business objective comprises maximized revenue or maximized user clicks, and wherein the plurality of advertisements comprise sponsored advertisements.

16. The method of claim 14, wherein a level of an advertisement within the hierarchal set of search results pages comprises a rank, and wherein multiple values of d* are calculated for each search results page of the set of hierarchal search results pages.

17. The method of claim 16, further comprising:

creating a histogram of the multiple d* for the keyword, including a frequency of a rank-depth corresponding to each d*.

18. The method of claim 16, the method further comprising:

computing class cardinality of the plurality of advertisers based on price bid elasticity of each advertiser for the keyword; and

if a value for d* reaches a rank-depth corresponding to a class K of advertiser, decrementing d* corresponding to the rank-depth by 1 until the advertiser class of the rank-depth corresponds to a class K−

1;

wherein the advertiser classes comprise at least, in descending order, market maker, market neutral, and market follower.

19. The method of claim 16, wherein the probability of a click at the given depth comprises a depth elasticity of click rank distribution at that rank-depth on the particular search results page, and wherein the bid price estimate is based on a price bid elasticity at that rank-depth.

20. A method for maximizing page yield by optimizing display depth of advertisements displayed to searching Web users, the method executed by a server having a processor and memory, the method comprising:

(a) tracking click activity, by the server, on a plurality of sponsored advertisements by a plurality of users via a set of hierarchal search results pages delivered in response to a search query for a keyword by a search engine;

(b) tracking bidding activity, by the server, of a plurality of advertisers related to the keyword and corresponding advertisements of the plurality of advertisements;

(c) calculating, by the processor, an optimum number of advertisements (k*) to display on a particular search results page of the set of hierarchal search results pages in response to the search query such that a probability that a total share of click volume being greater than that of any of the plurality of advertisers is less than or equal to a predetermined business tuning parameter (β

), wherein the probability values are obtained from a probability of click distribution based on the tracked click activity from at least one query;

(d) executing, by the processor, a Monte Carlo randomization procedure to include;

(i) randomly sampling a different probability of click distribution than that of step (c) based on at least one different query;

(ii) recalculating k* as per step (c); and

(iii) repeating steps (i) and (ii) to calculate potential multiple optimum numbers (k*) of advertisements; and

(e) serving, by the server, to the particular search results page at least some of the plurality of advertisements according to at least one calculated value of k* in response to the search query for the keyword.

21. The method of claim 20, wherein each value of k* represents a display depth on the particular search results page, wherein calculating k* in step (c) further comprises:

(f) computing class cardinality of the plurality of advertisers based on price bid elasticity of each advertiser for the keyword, wherein the class cardinality includes classes comprising at least market maker, market follower, and market neutral;

(g) decrementing k* by 1 if the class of the advertiser at the display depth corresponding to k* is market neutral; and

(h) repeating step (g) until the advertiser at the given depth is determined to be a market maker or a market follower.

22. The method of claim 21, wherein a level of an advertisement within the hierarchal set of search results pages comprises a rank, and wherein multiple values of k* are calculated for each search result page of the set of hierarchal search results pages.

23. The method of claim 22, wherein the tuning parameter (β

) reflects a percentage of clicks to be allocated to higher ranks of the hierarchal set of search results pages.

24. The method of claim 22, further comprising:

creating a histogram of the multiple k* for the keyword, including a frequency of a rank-depth corresponding to each k*.

25. A method for maximizing page yield by optimizing display depth of advertisements displayed to searching Web users, the method executed by a server having a processor and memory, the method comprising:

(a) tracking click activity, by the server, on a plurality of sponsored advertisements by a plurality of users via a set of hierarchal search results pages delivered in response to a search query for a keyword by a search engine;

(b) tracking bidding activity, by the server, of a plurality of advertisers related to the keyword and corresponding advertisements of the plurality of advertisements;

(c) calculating, by the processor, an optimum number of advertisements (k*) to display on a particular search results page of the set of hierarchal search results pages in response to the search query such that a probability that a total share of click volume being greater than that of any of the plurality of advertisers is less than or equal to a predetermined business tuning parameter (β

), wherein k* represents a display depth on the particular search results page;

(d) computing, by a class cardinality module coupled with the processor, class cardinality of the plurality of advertisers based on price bid elasticity of each advertiser for the keyword, wherein the class cardinality includes classes comprising at least market maker, market follower, and market neutral;

(e) decrementing, by the processor, k* by 1 if the class of the advertiser at the display depth corresponding to k* is market neutral;

(f) repeating, by the processor, step (e) until the advertiser at the given depth is determined to be a market maker or a market follower; and

(g) serving, by the server, to the particular search results page at least some of the plurality of advertisements according to at least one calculated value of k* in response to the search query for the keyword.