CN103268344A - Method for matching advertisements and pages without position bias influence - Google Patents
Method for matching advertisements and pages without position bias influence Download PDFInfo
- Publication number
- CN103268344A CN103268344A CN2013101977315A CN201310197731A CN103268344A CN 103268344 A CN103268344 A CN 103268344A CN 2013101977315 A CN2013101977315 A CN 2013101977315A CN 201310197731 A CN201310197731 A CN 201310197731A CN 103268344 A CN103268344 A CN 103268344A
- Authority
- CN
- China
- Prior art keywords
- advertisement
- page
- advertisements
- clicking rate
- under
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for matching advertisements and pages without a position bias influence. The method comprises the following steps of a, confirming showing positions, showing times and click times of all the advertisements in each search page; b using a statistical method for being capable of eliminating influences of page positions where the advertisements are located on advertisement clicks, and calculating an inherent click rate of each published advertisement under each page; c, calculating a similarity relationship about the advertisements between pages according to the inherent click rates of the pages and the advertisements; d, conducting suitable advertisements matching on a target page to be recommended according to other similar pages. The method can adjust click rates of specific advertisements under specific pages according to the showing positions of the advertisements and has the advantages of grasping the degree of correlation of the pages and the advertisements in a closer mode, more accurately matching the advertisements for the pages, and being more efficient, convenient and fast.
Description
Technical field
The present invention relates to be used to the field of search engine page match advertisements, specifically a kind of advertisement coupling choosing method based on the collaborative filtering neighbourhood model.
Background technology
When user's browsing page, the searching page of search engine particularly, the sidebar of the page can be showed some advertisements from top to bottom, these advertisements are exactly by the advertisement matching system, the matching degree of calculating and this page, the advertisement that matching degree is high are illustrated in the page sidebar by last position.
In various advertisement matching systems, whether the degree of correlation between specific webpage and the particular advertisement is mated an important indicator of this advertisement for this page.This index can be represented by the clicking rate of particular advertisement under the specific webpage.
Traditional clicking rate computing method find displaying number of times, the clicked number of times of advertisement under the page that will calculate that will calculate by daily record, just are divided by simply.Ignored the influence of advertisement display location to the ad click number of times.In the advertisement matching system in early days, decide the position of displaying according to gray bidding, the advertisement meeting of overbid is illustrated in by in the last location advertising, and does not consider whether this advertisement is relevant with the page.According to studies show that, it is easier to be clicked by last advertisement to be illustrated in the position, although the degree of correlation of this advertisement and its page of publication does not have the degree of correlation height of position advertisement on the lower and this page, this is that user's the custom of browsing causes.So, even incoherent advertisement also has a large amount of clicking rates, namely be regarded as relevantly with this page, this is illogical.Therefore, use traditional clicking rate computing method simply, the clicking rate of trying to achieve has position prejudice, and such clicking rate can not be expressed the degree of correlation of particular advertisement and specific webpage well.
Summary of the invention
The objective of the invention is the technological deficiency of seeing at ignore bit's offset in the prior art and a kind of advertisement matching process of getting rid of position prejudice that provides, this method can be according to the click situation of all advertisements under the ad-hoc location, adjust the clicking rate of indivedual advertisements, and according to the advertisement publishing situation of the similar page, adjust the advertisement coupling of target pages.
The concrete technical scheme that realizes the object of the invention is:
The method that a kind of advertisement of not having an influence of position prejudice and the page mate, this method comprises the steps:
A) determine all advertisements in each searched page display location, show number of times and number of clicks; Specifically comprise:
I) determines that page sum M, advertisement sum N, the page can show the total number of positions P of advertisement;
II) calculates under each page each advertisement in displaying number of times, the number of clicks of each position; Calculate displaying sum, the total number of clicks of all advertisements of each position under each page; Calculate each advertisement in displaying sum, the total number of clicks of each position; Calculate displaying sum, the total number of clicks of displaying sum, total number of clicks and each advertisement of all advertisements under each position; According to the data that obtain, the number of clicks of correspondence divided by showing number of times, is obtained each advertisement under each page in the clicking rate of the clicking rate of all advertisements of each position under the clicking rate of each position, each page, each advertisement all advertisements under the clicking rate of each position, each position and the clicking rate of each advertisement;
B) utilize statistical method, get rid of advertisement place page location to the influence that ad click causes, calculate the intrinsic clicking rate of each advertisement of publishing under each page; Specifically comprise:
I) according to total probability formula and Bayesian formula, calculate the clicked conditional probability of certain advertisement under certain page, i.e. the without prejudice clicking rate of certain advertisement under certain page got rid of location advertising to the influence of ad click rate;
C) according to the intrinsic clicking rate of the page and advertisement, the similarity about advertisement between the calculating page concerns; Specifically comprise:
I) page table is shown as a N dimensional vector, the without prejudice clicking rate of respective advertisement under corresponding this page of the element on each dimension; Add up to M page vector;
II) calculates M page vector of total similarity between any two by adjusting the cosine similarity;
D) for a target pages to be recommended, according to other similar pages, be the suitable advertisement of its coupling; Specifically comprise:
I) for the target pages q of a P to be matched advertisement, according to the similarity in the step c), finds K the page the most similar to page q, be called similar neighbours;
II) from similar neighbours, finds the advertising aggregator A that in similar neighbours, shows but under target pages q, do not show;
III) for each advertisement that belongs in the set A, be weighted summation according to the without prejudice clicking rate of corresponding advertisement among the similar neighbours and the similarity of similar neighbours and page q, calculate the without prejudice clicking rate that this advertisement is predicted under target pages q;
IV) the without prejudice clicking rate of all advertisements in the set A according to prediction sorted from big to small, select P maximum advertisement to recommend target pages q and show as advertisement.
Compare with background technology, the present invention has following advantage:
The present invention is when calculating the degree of correlation of the page and advertisement, consider that basic data has influenced the authenticity of data owing to the effect that is subjected to position prejudice, got rid of the influence of position prejudice to data by probability statistics and calculating, make the clicking rate of advertisement can truly reflect the degree of correlation of this advertisement and the page, more reasonably reduced the information that contains in the data.
The present invention with reference to the thought of collaborative recommendation, utilizes the similar page to carry out the selection of advertisement and the pre-estimation of degree of correlation when being page match advertisements, is page match advertisements more accurately.
The present invention does not need extra data support, does not need to carry out the analysis of text, the computational short cut of vector whole calculation process, improved the efficient of system.
Description of drawings
Fig. 1 is schematic flow sheet of the present invention;
Fig. 2 is the page synoptic diagram in the embodiment of the invention.
Embodiment
Consult Fig. 1, the present invention is applied in the advertisement matching system, determine that at first location advertising is to the influence of the click generation of advertisement, calculate and get rid of the without prejudice clicking rate of this influence back particular advertisement under specific webpage, situation according to page advertisement-printing finds similar neighbours' page then, in conjunction with the similar page, be the suitable advertisement of target pages coupling at last, its concrete steps are as follows:
The first step: from search engine logs, obtain each advertisement at each locational click, exhibition information, the total number of positions P of each page, calculate the clicking rate that has position prejudice, comprise the page
Following advertisement
In the position
Clicking rate
, the page
Upper/lower positions
The clicking rate of all advertisements
, advertisement
In the position
Clicking rate
, the position
The clicking rate of following all advertisements
And advertisement
Clicking rate
Second step: calculate at the page by following formula
Following advertisement
Clicked conditional probability, i.e. advertisement
At the page
Under the without prejudice clicking rate,
Make up page vector with the without prejudice clicking rate.
The 3rd step: utilize second to go on foot the page vector that obtains, a. utilizes the vector calculation adjustment cosine similarity between the page in twos, and b. stores these vectors, makes up without prejudice clicking rate database;
The 4th step: according to the relational database between the similarity structure page;
The 5th step: wait for that the user submits a query page to
The 6th step: find out and the page in the relational database that from the 4th step, obtains
The K of a similarity maximum page
, y=1 ~ K, with and and the page
Similarity
The 7th step: from the without prejudice database, find out in K the similar page and published, and at the page
In the advertising aggregator A that do not publish, for all advertisements in the set A
, calculate at the page by following formula
Following advertisement
Clicked conditional probability, i.e. advertisement
At the page
Under the without prejudice clicking rate:
Select P the advertisement coupling page of without prejudice clicking rate maximum then
The 8th step: judged whether to finish the match advertisements work of all pages, "Yes" changes end over to, and "No" changed for the 5th step over to.
Understand the present invention better by following examples.
Embodiment
Supposed to put down in writing in the search engine logs click record of three pages and advertisement-printing thereof, three pages are respectively: the search word that has occurred three times is the page of " weather "; The search word that has occurred twice is the page of " raining "; The page of search word once for " spring outing " appearred.The clicked situation of every kind of advertisement wherein represents that with the advertisement of underscore this advertisement is clicked in this page as shown in Figure 2 under each page, and it is clicked not represent that with the advertisement of glissade this advertisement does not have in this page.
To " rain " to the page below and carry out the advertisement coupling.
The first step: from search engine logs, obtain each advertisement in each locational click of each page, exhibition information, total number of positions, calculate the clicking rate that has position prejudice, comprise each page (" weather ", " rain ", " spring outing ") each advertisement (" weather research institute " down, " weather forecast order ", " umbrella ", " hygrometer distribution ", " free air temperature gauge distribution ", " barbecue grill taxi ", " tent taxi ") in each position (" position one ", " position two ", " position three ") clicking rate, each page (" weather ", " rain ", " spring outing ") following each position (" position one ", " position two ", " position three ") the clicking rate of all advertisements, each advertisement (" weather research institute ", " weather forecast order ", " umbrella ", " hygrometer distribution ", " free air temperature gauge distribution ", " barbecue grill taxi ", " tent taxi ") in each position (" position one ", " position two ", " position three ") clicking rate, each position (" position one ", " position two ", " position three ") clicking rate and each advertisement (" weather research institute " of down all advertisements, " weather forecast order ", " umbrella ", " hygrometer distribution ", " free air temperature gauge distribution ", " barbecue grill taxi ", " tent taxi ") clicking rate;
Second step: calculate without prejudice clicking rate (following only show non-vanishing clicking rate) by formula:
Pr (weather research institute | weather)=3/8; Pr (weather forecast is subscribed to | weather)=5/8;
Pr (umbrella is promoted | weather)=1/2; Pr (hygrometer is promoted | weather)=1/2;
Pr (weather forecast is subscribed to | rain)=1/4; Pr (hygrometer is promoted | rain))=1/3;
Pr (free air temperature gauge is promoted | rain)=1;
Pr (weather research institute | spring outing)=1/4; Pr (barbecue grill is hired out | weather)=1; Pr (tent is hired out | spring outing)=1/2
Obtain three page vectors thus: weather: (3/8,5/8,1/2,1/2,0,0,0); Rain: (0,1/4,0,1/3,1,0,0); Spring outing: (1/4,0,0,0,0,1,1/2).Similarity according to three pages of formula calculating obtains sim(weather, rains)=0.2787, sim(rains, spring outing)=0, sim(weather, spring outing)=0.0809;
The 3rd step: utilize second to go on foot the page vector that obtains, a. utilizes the vector calculation adjustment cosine similarity between the page in twos, and b. stores these vectors, makes up without prejudice clicking rate database;
The 4th step is according to the relational database between the similarity structure page;
The 5th step: wait for that the submit queries page " rains ";
The 6th step: from the 4th step, find out in the relational database that obtains and " rain " page " weather " of similarity maximum of the page, with and the similarity sim(weather that " rains " with the page, rainy);
The 7th step: from the without prejudice database, find out in the similar page " weather " and published, and the advertising aggregator A:(advertisement of in the page " rains ", not publishing " weather research institute ", advertisement " umbrella "), for all advertisements in the set A " weather research institute " and advertisement " umbrella ", calculate advertisement " weather research institute " and the clicked conditional probability of advertisement " umbrella " under the page " rains " by formula, i.e. advertisement " weather research institute " and the without prejudice clicking rate of advertisement " umbrella " under the page " rains ":
Pr(weather research institute | rain)=3/8, Pr (umbrella | rain)=1/2;
Select advertisement " umbrella " the coupling page of without prejudice clicking rate maximum " to rain " then;
The 8th step: judged whether to finish the match advertisements work of all pages, "Yes" changes end over to, and "No" changed for the 5th step over to.
Claims (5)
1. one kind is not had the advertisement of position prejudice influence and the method for page coupling, it is characterized in that this method comprises the steps:
A) determine all advertisements in each searched page display location, show number of times and number of clicks;
B) utilize statistical method, get rid of advertisement place page location to the influence that ad click causes, calculate the intrinsic clicking rate of each advertisement of publishing under each page;
C) according to the intrinsic clicking rate of the page and advertisement, the similarity about advertisement between the calculating page concerns;
D) for a target pages to be recommended, according to other similar pages, be the suitable advertisement of its coupling.
2. method according to claim 1 is characterized in that, described step a) comprises:
I) determines that page sum M, advertisement sum N, the page can show the total number of positions P of advertisement;
II) calculates under each page each advertisement in displaying number of times, the number of clicks of each position; Calculate displaying sum, the total number of clicks of all advertisements of each position under each page; Calculate each advertisement in displaying sum, the total number of clicks of each position; Calculate displaying sum, the total number of clicks of displaying sum, total number of clicks and each advertisement of all advertisements under each position; According to the data that obtain, the number of clicks of correspondence divided by showing number of times, is obtained each advertisement under each page in the clicking rate of the clicking rate of all advertisements of each position under the clicking rate of each position, each page, each advertisement all advertisements under the clicking rate of each position, each position and the clicking rate of each advertisement.
3. method according to claim 1 is characterized in that, described step b) comprises:
I) according to total probability formula and Bayesian formula, calculate the clicked conditional probability of certain advertisement under certain page, i.e. the without prejudice clicking rate of certain advertisement under certain page got rid of location advertising to the influence of ad click rate.
4. method according to claim 1 is characterized in that, described step c) comprises:
I) page table is shown as a N dimensional vector, the without prejudice clicking rate of respective advertisement under corresponding this page of the element on each dimension; Add up to M page vector;
II) calculates M page vector of total similarity between any two by adjusting the cosine similarity.
5. method according to claim 1 is characterized in that, described step d) comprises:
I) for the target pages q of a P to be matched advertisement, according to the similarity in the step c), finds K the page the most similar to page q, be called similar neighbours;
II) from similar neighbours, finds the advertising aggregator A that in similar neighbours, shows but under target pages q, do not show;
III) for each advertisement that belongs in the set A, be weighted summation according to the without prejudice clicking rate of corresponding advertisement among the similar neighbours and the similarity of similar neighbours and page q, calculate the without prejudice clicking rate that this advertisement is predicted under target pages q;
IV) the without prejudice clicking rate of all advertisements in the set A according to prediction sorted from big to small, select P maximum advertisement to recommend target pages q and show as advertisement.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310197731.5A CN103268344B (en) | 2013-05-24 | 2013-05-24 | A kind of method of mating with the page without the advertisement of position prejudice impact |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310197731.5A CN103268344B (en) | 2013-05-24 | 2013-05-24 | A kind of method of mating with the page without the advertisement of position prejudice impact |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103268344A true CN103268344A (en) | 2013-08-28 |
| CN103268344B CN103268344B (en) | 2016-04-06 |
Family
ID=49011973
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310197731.5A Active CN103268344B (en) | 2013-05-24 | 2013-05-24 | A kind of method of mating with the page without the advertisement of position prejudice impact |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103268344B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104992348A (en) * | 2015-06-24 | 2015-10-21 | 深圳市腾讯计算机系统有限公司 | Method and device for displaying information |
| CN105825400A (en) * | 2016-03-16 | 2016-08-03 | 合网络技术（北京）有限公司 | Advertisement material quality measuring method and system |
| CN106537436A (en) * | 2015-04-23 | 2017-03-22 | Sk 普兰尼特有限公司 | Retargeted advertised product recommendation user device and service providing device, advertised product recommendation system comprising same, method for controlling same and recording medium having computer program recorded therein |
| CN106875208A (en) * | 2016-12-27 | 2017-06-20 | 网易传媒科技（北京）有限公司 | A kind of method and apparatus for determining notable advertisement show position |
| CN107516247A (en) * | 2017-08-28 | 2017-12-26 | 天脉聚源（北京）科技有限公司 | A kind of method and device for predicting advertisement played data |
| WO2020134009A1 (en) * | 2018-12-29 | 2020-07-02 | 北京字节跳动网络技术有限公司 | Click rate prediction method and apparatus, and display position selection method and apparatus |
| CN113038242A (en) * | 2021-05-24 | 2021-06-25 | 武汉斗鱼鱼乐网络科技有限公司 | Method, device and equipment for determining display position of live broadcast card and storage medium |
| CN117950787A (en) * | 2024-03-22 | 2024-04-30 | 成都赛力斯科技有限公司 | Advertisement display method and device, electronic equipment and storage medium |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109753601B (en) * | 2018-11-28 | 2021-10-22 | 北京奇艺世纪科技有限公司 | Method and device for determining click rate of recommended information and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2005101263A1 (en) * | 2004-04-17 | 2005-10-27 | Nhn Corporation | System and method for selecting search listing in an internet search engine and ordering the search listings |
| CN101079039A (en) * | 2006-06-30 | 2007-11-28 | 腾讯科技（深圳）有限公司 | Search advertisement sequencing system and method |
| CN102592235A (en) * | 2011-12-28 | 2012-07-18 | 北京品友互动信息技术有限公司 | Internet advertisement serving system |
-
2013
- 2013-05-24 CN CN201310197731.5A patent/CN103268344B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2005101263A1 (en) * | 2004-04-17 | 2005-10-27 | Nhn Corporation | System and method for selecting search listing in an internet search engine and ordering the search listings |
| CN101079039A (en) * | 2006-06-30 | 2007-11-28 | 腾讯科技（深圳）有限公司 | Search advertisement sequencing system and method |
| CN102592235A (en) * | 2011-12-28 | 2012-07-18 | 北京品友互动信息技术有限公司 | Internet advertisement serving system |
Non-Patent Citations (1)
| Title |
|---|
| 卫强: "网络展示广告位置对点击率CTR影响的实证研究", 《信息系统学报》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106537436A (en) * | 2015-04-23 | 2017-03-22 | Sk 普兰尼特有限公司 | Retargeted advertised product recommendation user device and service providing device, advertised product recommendation system comprising same, method for controlling same and recording medium having computer program recorded therein |
| CN104992348A (en) * | 2015-06-24 | 2015-10-21 | 深圳市腾讯计算机系统有限公司 | Method and device for displaying information |
| CN104992348B (en) * | 2015-06-24 | 2018-07-10 | 深圳市腾讯计算机系统有限公司 | A kind of method and apparatus of information displaying |
| CN105825400A (en) * | 2016-03-16 | 2016-08-03 | 合网络技术（北京）有限公司 | Advertisement material quality measuring method and system |
| CN106875208A (en) * | 2016-12-27 | 2017-06-20 | 网易传媒科技（北京）有限公司 | A kind of method and apparatus for determining notable advertisement show position |
| CN107516247A (en) * | 2017-08-28 | 2017-12-26 | 天脉聚源（北京）科技有限公司 | A kind of method and device for predicting advertisement played data |
| WO2020134009A1 (en) * | 2018-12-29 | 2020-07-02 | 北京字节跳动网络技术有限公司 | Click rate prediction method and apparatus, and display position selection method and apparatus |
| CN113038242A (en) * | 2021-05-24 | 2021-06-25 | 武汉斗鱼鱼乐网络科技有限公司 | Method, device and equipment for determining display position of live broadcast card and storage medium |
| CN117950787A (en) * | 2024-03-22 | 2024-04-30 | 成都赛力斯科技有限公司 | Advertisement display method and device, electronic equipment and storage medium |
| CN117950787B (en) * | 2024-03-22 | 2024-05-31 | 成都赛力斯科技有限公司 | Advertisement display method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103268344B (en) | 2016-04-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103268344B (en) | A kind of method of mating with the page without the advertisement of position prejudice impact | |
| CN104281956A (en) | Dynamic recommendation method capable of adapting to user interest changes based on time information | |
| CN103324645A (en) | Method and device for recommending webpage | |
| CN108897750B (en) | Personalized place recommendation method and device integrating multiple contextual information | |
| CN102270241A (en) | Image retrieving method based on sparse nonnegative matrix factorization | |
| CN106528812B (en) | A kind of cloud recommended method based on USDR model | |
| CN103235822A (en) | Database generating and querying method | |
| CN104063555B (en) | The user model modeling method intelligently distributed towards remote sensing information | |
| CN111445280A (en) | Model generation method, restaurant ranking method, system, device and medium | |
| Jiang et al. | Evolution of Crop Planting Structure in Traditional Agricultural Areas and Its Influence Factors: A Case Study in Alar Reclamation | |
| Li et al. | Spatial interpolation method of precipitation based on topographic weighting. | |
| CN107463615A (en) | Method is recommended based on the real-time place to go of context and user interest in open network | |
| Li | Study on measurement of seasonality and its regional differences of self-driving tours market in China. | |
| Arya et al. | Community water harvesting structures' status, management and socio-economic impacts: case studies in Shivaliks foothills, Northern Himalayas, India. | |
| CN103106268A (en) | User preference search method | |
| Li BinBin et al. | Response of vegetation pattern characteristics to sediment yield in Dali River Basin. | |
| Li XueFeng et al. | Quantitative method for freezing-thawing environment of concrete in Qinghai-Tibet Plateau. | |
| Duan et al. | Spatial and temporal characteristics of precipitation in Chongqing from 1961 to 2014. | |
| Gao XueLi et al. | Evolution and driving forces of cultivated land quantity in Xiamen City using niche theory. | |
| Chang CheHao et al. | An evaluation framework for identifying the optimal raingauge network based on spatiotemporal variation in quantitative precipitation estimation. | |
| Ma ZiQiang et al. | The first comparisons of IMERG and the downscaled results based on IMERG in hydrological utility over the Ganjiang River basin. | |
| Maimaitijiang Maitiniyazi et al. | Spatial-temporal change of Urumqi urban land use and land cover based on grid cell approach. | |
| Jing Xueer et al. | Analysis and modelling of stormwater volume control performance of rainwater harvesting systems in four climatic zones of China. | |
| Ma HuiQin et al. | Wheat powdery mildew forecasting in filling stage based on remote sensing and meteorological data. | |
| Wu ChunHung | Landslide susceptibility mapping by using landslide ratio-based logistic regression: a case study in the southern Taiwan. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |