CN103034718A - Target data sequencing method and target data sequencing device - Google Patents

Abstract

The invention discloses a target data sequencing method and a target data sequencing device. The method comprises the following steps of acquiring seven attribution values of target data; calculating a sequence value of the target data according to the seven attribution values; and judging whether the sequence value conforms to a preset condition or not, and submitting the target data to a sequencing system to be sequenced if the sequence value conforms to the preset condition, wherein the seven attribution values are sequentially key word click rate of the target data, a matching mode indicator value of a key word in a searching engine, whether a wildcard character is contained in the target data or not, classification number of a website uniform resource locator (URL) of the target data, matching degree of a title of the target data and the key word, matching degree of description of the target data and the key word, and a scale indicator value of a unit in which the key word is located. By utilizing the method and the device, the efficiency for submitting the target data to the sequencing system can be increased, so that the sequencing efficiency of the sequencing system for the target data can be increased.

Description

A kind of target data sort method and device

Technical field

The present invention relates to the data sorting field, more particularly, relate to a kind of target data sort method and device.

Background technology

In real life, often need to sort to some data.Common such as the paid search advertisement is arranged.Its main operation form is: at first, advertiser has the keyword of commercial value by determining one or more, then by the bid ranking platform of search engine, obtain the preferential power of rank in relevant search result, attract relevant consumer to click the target approach website with this.At home, topmost paid advertisement release platform is Baidu (www.baidu.com).Take Baidu as example, its bid ranking platform is by the collective advertising merchant best bid of the keyword in the advertisement and the quality degree of the keyword in the advertisement to be decided.When bid was identical, the order ads that keyword quality degree is high was forward, will preferentially be thrown in.

For the advertisement that makes oneself can preferentially be thrown in, advertiser all can predict the keyword quality degree of advertisement of oneself in the prior art, determines that according to predicting the outcome whether an advertisement being committed to search engine sorts.The method of existing evaluate advertisements keyword quality degree height mainly is Artificial Diagnosis.Namely rely on experienced business personnel, perhaps the consultant judges quality, then reaches the standard grade and does the AB test.Specifically, a typical optimizing process can be summarized as follows: at first, the consultant can be that many advertising creatives (advertising creative specifically comprises title, describes 1, describes 2) write in keyword according to experience and the related service knowledge of oneself; Then, do preliminary screening according to subjective experience, pick out 2-3 and seem preferably intention; Next, these several preferably intention all are submitted to the search engine backstage carry out measure of merit, these intention of test period will be demonstrated to the searchers in turn.

The consultant to the click situation of each intention, analyzes the reason of different intention difference on effect by the test phase user, attempts finding systematic rule, further improves intention according to rule again, and the intention after the improvement enters second and takes turns on the line and test.By that analogy, reciprocation cycle is ceaselessly optimized intention.But existing have following shortcoming based on artificial method:

The first, this method mainly relies on people's subjective judgement, is easy to occur for same keyword, same intention, and different consultants' suggestion is not consistent.And the keyword that large enterprise will promote may reach 100,000 or 1,000,000 magnitudes, when scale surpasses to a certain degree appraisal just surmount manpower can and scope.Usually the consultant can be primarily focused on a small quantity in the optimization of the most popular keyword, unablely takes most relatively keywords of long-tails into account.Will reduce like this accuracy that advertisement keyword quality degree is estimated.

The second, the final judge of this method mainly relies on AB test actual on the line, but this just need to obtain real client's click data as judging basis, means the advertising expenditure that will drop into suitable number; And to obtain test result on the reliable and stable line, mean each take turns test usually the data that accumulate about 1 week of needs just can judge accurately, taken the plenty of time.

Based on above-mentioned quality degree Forecasting Methodology, so that the efficient of user when determining to submit advertising creative to search engine to is lower, thereby so that the ordering efficient of search engine is low.、

Above-mentioned phenomenon appears on the data of other type too, therefore, is badly in need of at present a kind of data being predicted to determine whether it is committed to the method for ordering system ordering.

Summary of the invention

The embodiment of the invention provides a kind of target data sort method and device, submits to target data to the efficient of ordering system to improve, thereby improves ordering system to the ordering efficient of target data.

For this reason, the embodiment of the invention provides following technical scheme:

A kind of target data sort method, the method comprises:

Obtain seven property values of described target data;

According to described seven property values, calculate the ranking value of described target data;

It is pre-conditioned to judge whether described ranking value meets, if meet, then submits to described target data to ordering system to sort;

Wherein, described seven property values that obtain described target data comprise:

Obtain the keyword clicking rate of described target data;

Obtain the match pattern expressivity value that described keyword arranges in search engine; Wherein, described match pattern is presented as broad match, phrase match, exact matching;

Judge the expressivity value that whether comprises asterisk wildcard in the described target data and generate judged result;

Obtain the classifying and numbering of the web page address URL of described target data;

Obtain the matching degree of title and the described keyword of described target data;

Obtain the description 2 of described target data and the matching degree of described keyword;

Obtain the scale expressivity value of unit, described keyword place.

Preferably, described according to described property value, the ranking value that calculates described target data comprises:

With described seven property values input sequencing regression model, find the solution described sequencing regression model, calculate described target data and get different ordering grade k ₁, k ₂, k ₃The time probable value; Wherein, described sequencing regression model is specially: $p(Y=k|X)=\left\{\begin{array}{c}\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),k=k_1\\ \mathrm{\Φ}(c_2-X^{\′}\mathrm{\β})-\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),\\ 1-\mathrm{\Φ}(c_2-X^{\′}\mathrm{\β}),k=k_3\end{array}\right.$ K=k ₂Y, k represent the ordering grade of described target data, and p (Y=k|X) expression Y gets the probability of k value; Φ (x) is the distribution density function of standardized normal distribution, c ₁, c ₂Be the threshold value of described model, β=(β ₀, β ₁..., β ₇) be the regression coefficient of described model; X=(x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇), described x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇Represent respectively the second description part of the title of described keyword clicking rate, described match pattern, described judged result, described classifying and numbering, described target data and the matching degree of described keyword, described target data and matching degree and the described expressivity value of described keyword;

Determine that maximum ordering grade corresponding to probable value is the ranking value of described target data.

Preferably, when being broad match, described match pattern expressivity value is 0, when when the phrase match, described match pattern expressivity value is 1, when being exact matching, described match pattern expressivity value is 2.

Preferably, when described target data contained asterisk wildcard, the expressivity value of described judged result was 1, and when described target data did not contain asterisk wildcard, the expressivity value of described judged result was 0.

Preferably, describedly obtain the title of described target data and the matching degree of described keyword comprises:

Extract the title of described target data;

Calculate the character number that occurs in keyword that contains in the described title and the ratio of the total number of characters of described title, described ratio is the matching degree of title and the described keyword of described target data.

And/or;

The described description of obtaining described target data 2 comprises with the matching degree of described keyword:

Extract the description 2 of described target data;

Calculate the ratio of total number of characters of the character number that occurs in keyword that contains in the described description 2 and described description 2, described ratio is the description 2 of described target data and the matching degree of described keyword.

Preferably, the described expressivity value of obtaining the scale of unit, described keyword place comprises:

Obtain the similar keyword number of unit, described keyword place;

Judge whether described number surpasses certain threshold value;

If surpass, determine that then described scale expressivity value is 0, if do not surpass, determine that then described scale expressivity value is 1.

Preferably, described method also comprises and makes up in advance the sequencing regression model, finds the solution the threshold value c of described model ₁, c ₂With the regression coefficient β of described model=(β ₀, β ₁..., β ₇): specifically comprise:

According to Maximum Likelihood Estimation, construct the likelihood function of described sequencing regression model $L\left(\mathrm{\β}\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\log \left\{P\left(Y_i\right|X_i)\right\}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{k=1}{\overset{3}{\mathrm{\Σ}}}I(Y_i=k)\log \left\{p_k({X_i}^{\′},\mathrm{\β},c)\right\},$ C=c ₁, c ₂Wherein, (X _i, Y _i) i sample data of expression;

Obtain historical property value and N corresponding history ordering grade of described other data of N other data; Described N is positive integer;

Take i described historical ordering grade as Y _i, take i historical property value as X _i, find the solution and make described likelihood function reach peaked (β, c).

Preferably, described target data is advertising creative.

The present invention also provides a kind of target data collator, and described device comprises:

The property value acquiring unit is for seven property values that obtain described target data;

The ranking value computing unit is used for according to described seven property values, calculates the ranking value of described target data;

Judging unit, it is pre-conditioned to be used for judging whether described ranking value meets;

Commit unit is used for meeting in described ranking value and submits to described target data to ordering system to sort when pre-conditioned;

Wherein, described property value acquiring unit specifically comprises:

Keyword clicking rate unit is for the keyword clicking rate of obtaining described target data;

The match pattern unit is used for obtaining the match pattern expressivity value that described keyword arranges at search engine; Described match pattern is presented as broad match, phrase match, exact matching;

The asterisk wildcard unit is used for judging whether described target data comprises asterisk wildcard and generate the expressivity value of judged result;

The web page address unit is for the classifying and numbering of the web page address URL that obtains described target data;

The title matching unit is for the matching degree of the title and the described keyword that obtain described target data;

Describe 2 matching units, be used for obtaining the description 2 of described target data and the matching degree of described keyword;

The scale unit is for the scale expressivity value of obtaining unit, described keyword place.

Preferably, described ranking value computing unit specifically comprises:

Input block is used for described seven property values input sequencing regression model; Described sequencing regression model is specially:

$p(Y=k|X)=\left\{\begin{array}{c}\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),k=k_1\\ \mathrm{\Φ}(c_2-X^{\′}\mathrm{\β})-\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),\\ 1-\mathrm{\Φ}(c_2-X^{\′}\mathrm{\β}),k=k_3\end{array}\right.$ K=k ₂Wherein, Y, k represent the ordering grade of described target data, and p (Y=k|X) expression Y gets the probability of k value; Φ (x) is the distribution density function of standardized normal distribution, c ₁, c ₂Be the threshold value of described model, β=(β ₀, β ₁..., β ₇) be the regression coefficient of described model; X=(x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇), described x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇Represent respectively the second description part of the title of described keyword clicking rate, described match pattern, described judged result, described classifying and numbering, described target data and the matching degree of described keyword, described target data and matching degree and the described expressivity value of described keyword;

The ranking value computation subunit is used for finding the solution described sequencing regression model, calculates described target data and gets different ordering grade k ₁, k ₂, k ₃The time probable value;

Ordering classification unit is used for determining that maximum ordering grade corresponding to probable value is the ranking value of described target data.

Preferably, when being broad match, described match pattern expressivity value is 0, when when the phrase match, described match pattern expressivity value is 1, when being exact matching, described match pattern expressivity value is 2.

Preferably, described asterisk wildcard unit, concrete being used for when described target data contains asterisk wildcard, the expressivity value that generates judged result is 1, when described target data did not contain asterisk wildcard, the expressivity value that generates judged result was 0.

Preferably, described title matching unit comprises:

The title extraction unit is for the title that extracts described target data;

Title matching degree computing unit is used for calculating the character number that occurs that described title contains and the ratio of the total number of characters of described title in keyword, described ratio is the matching degree of title and the described keyword of described target data.

And/or;

Described description 2 matching units comprise:

Describe 2 extraction units, be used for the description 2 of extracting described target data;

Describe 2 matching units, be used for calculating the ratio of the character number that occurs that described description 2 contains and total number of characters of described description 2 in keyword, described ratio is the description 2 of described target data and the matching degree of described keyword.

Preferably, described scale unit comprises:

Similar keyword unit is for the similar keyword number that obtains unit, described keyword place;

Judging unit is used for judging whether described number surpasses certain threshold value; If surpass, determine that then described expressivity value is 0, if do not surpass, determine that then described expressivity value is 1.

Preferably, described device also comprises sequencing regression model construction unit, is used for making up the sequencing regression model in advance, finds the solution the threshold value c of described model ₁, c ₂With the regression coefficient β of described model=(β ₀, β ₁..., β ₇): specifically comprise:

The likelihood function unit is used for according to Maximum Likelihood Estimation, constructs the likelihood function of described sequencing regression model $L\left(\mathrm{\β}\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\log \left\{P\left(Y_i\right|X_i)\right\}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{k=1}{\overset{3}{\mathrm{\Σ}}}I(Y_i=k)\log \left\{p_k({X_i}^{\′},\mathrm{\β},c)\right\},$ C=c ₁, c ₂Wherein, (X _i, Y _i) i sample data of expression;

The historical data unit is used for obtaining the historical property value of N other data and the history ordering grade of N correspondence of described other data; Described N is positive integer;

Likelihood function is found the solution the unit, is used for take i described historical ordering grade as Y _i, take i historical property value as X _i, find the solution and make described likelihood function reach peaked (β, c).

Preferably, described target data is advertising creative.

Compare with prior art, the invention has the beneficial effects as follows:

The present invention by obtaining target data seven property values and according to seven property values, calculate the ranking value of described target data, then it is pre-conditioned to judge whether ranking value meets, when meeting, submit to target data to ordering system to sort, compare existing artificial target of prediction data to determine whether submit to target data to the method for goal systems, the present invention has improved the efficient of submission target data to ordering system, thereby has improved the ordering efficient of ordering system to target data.

Description of drawings

Fig. 1 is that advertising creative of the present invention forms synoptic diagram;

Fig. 2 is user account structural drawing among the present invention;

Fig. 3 is the embodiment of the invention 1 method flow diagram;

Fig. 4 is the embodiment of the invention 2 method flow diagrams;

Fig. 5 is the embodiment of the invention 3 structure drawing of device.

Embodiment

In order to make those skilled in the art person understand better the scheme of the embodiment of the invention, below in conjunction with drawings and embodiments the embodiment of the invention is described in further detail.

Understand for convenient, following content describes the present invention take the paid search advertisement as example.

As stated in the Background Art, the paid search advertisement is one of most important advertisement putting mode on the present internet.Based on existing search engine, when only having keyword corresponding to the search word that comprises in the advertisement with user input, this advertisement just can be demonstrated, and the keyword among the present invention refers to before advertisement putting, the literal that the embodiment advertisement promotion that advertiser buys from search engine provider is intended to.Be keyword " plane ticket ", " 1.5 folding " that advertisement " take journey plane ticket preferential 1.5 folding " is bought such as certain advertiser.

Advertisement comprise network address URL that advertiser promotes and one section brief, have an agitative advertising slogan.Advertising slogan wherein is called as advertising creative, and is said such as background technology, advertising creative by title, describe 1 and describe 2 and form.Need to prove that above-mentioned said advertisement is demonstrated the actual advertising creative that refers to and is demonstrated.Describe 1 and describe 2 literal that refer to describe advertisement, distinguished when writing by advertiser.Accompanying drawing 1 shows the title of an advertising creative and describes 1 and describe 2.

When the user submits a search word to, often have a plurality of advertisers and bought corresponding keyword.At this moment, search engine will be represented a plurality of advertising creatives that comprise this keyword.The degree that is concerned of advertisement in the position influence that advertising creative represents.Existing search engine is generally at upper left side and the upper right side showing advertisements intention of the page.According to statistics, the degree that is concerned of the advertising creative of position, page upper left side is higher than the degree that is concerned of position, upper right side.According to order from top to bottom, the degree that is concerned of advertising creative descends successively.For obtaining the high degree that is concerned, advertiser will compete mutually the order that represents of advertisement, in order to oneself advertising creative is presented in the higher position of the degree of being concerned.

At present, the advertisement bidding rank rule of search engine comprises best bid and two factors of keyword quality degree usually.

Best bid refers to that advertiser is the each clicked ceiling price of paying of certain advertisement.When a plurality of advertisements were competed mutually, if best bid is identical, the advertising creative that keyword quality degree is high can obtain higher rank, thereby can be rendered on the relatively high position of the degree of being concerned.

Take Baidu as example, according to the record among the marketing http://yingxiao.baidu.com/support/fc/detail_629.html castk=LTE%3D of document Baidu of official of Baidu, the factors such as the clicking rate of Baidu's main taking into account critical word when the quality degree of keyword is estimated, the correlativity of advertising creative keyword and account general performance.

Need to prove that the clicking rate of keyword has referred to get rid of the clicking rate that advertising creative represents the impacts such as position in the document.The clicking rate of this moment can reflect that keyword itself attracts user's degree.

The correlativity of advertising creative keyword mainly refers to the degree of correlation of degree of correlation, keyword and the target web of keyword and advertising creative.

The general performance of account refers to the history popularization performance of other keywords in the account.Search engine is set up promoted account with managing advertisement for each advertiser that buys the paid search advertisement.Fig. 2 shows the structure of account: account comprises popularization plan, promotes unit and keyword Three Estate.Each popularization plan comprises a plurality of popularizations unit, and each is promoted the unit and comprises a plurality of keywords.Each is promoted the unit and includes several advertising creatives.When the search word of user's submission was triggered to certain keyword, a certain advertisement of this unit, keyword place will be demonstrated.

Be the rank that the advertisement that is committed to search engine can be obtained, advertiser needs make prediction to the keyword quality degree of advertising creative in advance, in order to determine that whether submitting advertisements to search engine throws in.

In conjunction with above-mentioned factor of evaluation to keyword quality degree, the embodiment of the invention 1 provides a kind of method that advertising creative is sorted, and referring to Fig. 3, the method specifically comprises:

S11, obtain the keyword clicking rate of described advertising creative.

Such as, the click volume of certain keyword is 5, and the amount of representing is 10, and then the clicking rate of this keyword is 50%.

In actual applications, the clicking rate of keyword can obtain in several ways.Such as can be directly from relevant proce's-verbal, obtaining, perhaps can by click volume and the amount of representing of the keyword that from the proce's-verbal, gets access to, calculate.

S12, obtain the match pattern that described keyword arranges in search engine; Described match pattern is presented as broad match, phrase match, exact matching.

When user search, system can select corresponding keyword automatically, and the advertising creative that this keyword is corresponding is presented in face of the user.Different matching ways, the user search word is different from the corresponding relation between the keyword, and the advertising creative that represents based on same search word is just different.

Exact matching---only have the matching way that just can be demonstrated with the literal identical keyword of search word.

Phrase match---with the literal identical keyword of search word (being exact matching), comprise this search word and this search word insert the keyword of putting upside down form and with search word be the matching way that synon keyword can be demonstrated.

Broad match---with the literal identical keyword of search word (being exact matching), comprise this search word and this search word and insert the keyword of putting upside down form, be synon keyword and be the matching way that the keyword of the related variants form of search word can be demonstrated with search word.

Take search word as " Expert English language training by qualified teachers " as example, above-mentioned three kinds of match patterns are elaborated.

Exact matching---when only having keyword to be " Expert English language training by qualified teachers ", the advertising creative that this keyword is corresponding just can be demonstrated out.

When phrase match---keyword is " Expert English language training by qualified teachers ", " English training in summer time " (comprising fully), " training English " (putting upside down form), " English training " (synonym).Corresponding advertising creative can be demonstrated out.

When broad match---keyword is " Expert English language training by qualified teachers ", " English training in summer time " (comprising fully), " training English " (putting upside down form), " English training " (synonym), " education on foreign language " (variant form).Corresponding advertising creative can be demonstrated out.

Can find out that by above-mentioned example same keyword uses three kinds of match patterns, the search word scope that can cause triggering it progressively enlarges.Wherein, wide in range match pattern such as broad match can allow corresponding advertising creative that the chances that are demonstrated are arranged more.But this advertising creative that can allow a part not meet search intention is presented in face of the user, so that the user clicks the possibility less of advertisement, and corresponding lower clicking rate.And rigorous match pattern such as exact matching can make the advertising creative consistent with search intention be presented in face of the user, thus advertisement clicked may also can become large, corresponding higher clicking rate.

Calculate for follow-up, in embodiments of the present invention, the corresponding numerical value of broad match pattern, phrase match pattern, exact matching pattern can be set be respectively 0,1,2.

S13, judge whether comprise asterisk wildcard in the described advertising creative, and generate judged result.

Asterisk wildcard is a kind of special sentence, normally used as " { } " or " * " or "? " Deng expression.

When it was used for advertisement, its sign was generally: { acquiescence keyword }.Insert the advertising creative of asterisk wildcard when being demonstrated, will substitute above-mentioned asterisk wildcard sign with the keyword that triggers.Such as:

When an advertising creative is:

Title: Baidu's { fresh flower }, for look is added in life

Baidu's { fresh flower } is at hand, and the fragrant boundless universe allows happy upgrading, and Baidu's { fresh flower } freely send flower in the five rings, consulting 400-000-0000.

When user search " fresh flower of giving a present ", triggered keyword " fresh flower ", when this advertisement is demonstrated, its effect as shown in Figure 2, asterisk wildcard is wherein substituted by keyword " fresh flower ", and general rise of prices of the stocks and other securities shows automatically.

As seen, in advertising creative, use asterisk wildcard can guarantee that the part of advertising creative is consistent with the keyword that the user buys, thereby make keyword and advertising creative have certain correlativity.Simultaneously, trigger keyword in user search, when advertising creative was demonstrated, part consistent with term or that be close in meaning showed automatic general rise of prices of the stocks and other securities.And general rise of prices of the stocks and other securities can attract user's attention, thereby promotes clicking rate.Be whether to comprise the correlativity that asterisk wildcard affects keyword and advertising creative in the advertising creative, affect simultaneously the clicking rate of keyword.

Calculate for follow-up, in embodiments of the present invention, can arrange when containing asterisk wildcard in the advertising creative, value is 1, does not contain sometimes, and value is 0.

S14, obtain the classifying and numbering of the URL of described advertisement.

Search word has embodied user's search intention, and different webpages is different from the degree of correlation of user search intent.When being " Jingdone district " when search word, showing that the homepage in store, Jingdone district more meets search intention, and when search word is " Jingdone district purchases by group ", show that the webpage that Jingdone district purchases by group more can meet search intention than other webpage of demonstration such as homepage.As seen when a keyword and a webpage height correlation, the easier click of facilitating the user, thus improve clicking rate.It is the clicking rate that the web page address URL at advertising creative place affects keyword.

For this reason, can the quality degree of advertising creative be predicted in conjunction with the URL of advertisement.Concrete, the URL that can be advertisement classifies and numbers, and when being homepage such as the URL when this advertisement, is numbered 1, and the next stage webpage of homepage is numbered 2.About the concrete application of advertisement URL in model, will be elaborated follow-up.

S15, obtain the matching degree of title and the described keyword of described advertising creative.

The title of advertising creative is embodying the main contents of advertisement, and therefore, the matching degree between title and the keyword has reflected the degree of correlation of advertising creative and keyword.

The ratio of the keyword character number that contains in the title is larger, and when general rise of prices of the stocks and other securities showed, the proportion of general rise of prices of the stocks and other securities part was just larger in the title, and this advertising creative also just more can attract user's notice.Therefore, in specific embodiments of the invention, the title of advertising creative and the matching degree of described keyword can represent with the ratio of the number of characters of the key word that contains in the title.Specifically can calculate in the following way:

Extract the title of described advertising creative;

Calculate the number of characters that occurs in keyword that contains in the described title and the ratio of the total number of characters of described title, described ratio is the matching degree of title and the described keyword of described advertising creative.

Such as, when the title of an advertising creative is " Baidu's fresh flower is for look is added in life ", when keyword was " fresh flower ", the total number of characters of title was 9, and the number of characters of the keyword that occurs in the title is 2, and then the matching degree of the title of this advertising creative and keyword is 2/9.

S16, the description 2 of obtaining described advertising creative and the matching degree of described keyword.

The description of advertising creative is the direct embodiment of ad content.Therefore, the matching degree of the description 2 of advertising creative and keyword is embodying the degree of correlation of advertising creative and keyword.

The ratio of describing the keyword character number that contains in 2 is larger, and when general rise of prices of the stocks and other securities showed, the proportion of describing general rise of prices of the stocks and other securities part in 2 is just larger, and was corresponding, and this advertising creative also just more can attract user's notice.Therefore, in specific embodiments of the invention, the description 2 of advertising creative can represent with the ratio of the number of characters of describing the key word that contains in 2 with the matching degree of keyword.Specifically can calculate in the following way:

Extract the description 2 in the advertising creative.

The number of characters that occurs in keyword that contains in the calculating description 2 and the ratio of total number of characters of description 2, described ratio are the description 2 of advertising creative and the matching degree of described keyword.

Such as, when description 2 was " Baidu's fresh flower freely send flower in the five rings, consulting 400-000-0000 ", its total number of characters was 25, and the number of characters of the keyword of appearance is 2, and then the description 2 of advertising creative is 2/25 with the matching degree of keyword.

S17, obtain the expressivity value that the scale of unit is promoted at described keyword place.

Take Baidu as example, it promotes the subsection that the unit is a series of keyword/advertising creatives of management, and each promotes the shared a plurality of advertising creatives of a plurality of keywords under the unit, forms the multi-to-multi of keyword and advertising creative.

Under these circumstances, only have the similarity of each keyword in the unit high as far as possible, each keyword in the guarantee unit and advertising creative height correlation.If the unit scale is excessive, mean that the keyword grouping in the unit is careful not, similarity is not high each other, thereby affects the matching degree of keyword and advertising creative.As seen, the scale of keyword place popularization unit affects the matching degree of keyword and advertising creative.

In an embodiment of the present invention, the expressivity value of the scale of unit is promoted at the keyword place.Can represent in the following way:

When the keyword number that contains the unit was promoted at the keyword place greater than predetermined threshold value, value was 0, and during less than predetermined threshold value, value is 1.

Wherein, according to the record of official of Baidu document and the checking of experimental data, above-mentioned predetermined threshold value preferred 50.

S18, according to the expressivity value of the matching degree of the description 2 of the title of described keyword clicking rate, described match pattern, described judged result, described classifying and numbering, described advertising creative and the matching degree of described keyword, described advertising creative and described keyword and unit, described keyword place totally seven property values, calculate the quality degree of advertising creative, and with this ranking value as advertising creative.

Among the present invention, can calculate in several ways the ranking value of advertising creative.In a preferred embodiment, can make up the sequencing regression model according to above-mentioned seven property values, obtain the ranking value of advertising creative by finding the solution the sequencing regression model.Concrete, it is as follows to make up model:

$p(Y=k|X)=\left\{\begin{array}{c}\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),k=k_1\\ \mathrm{\Φ}(c_2-X^{\′}\mathrm{\β})-\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),\\ 1-\mathrm{\Φ}(c_2-X^{\′}\mathrm{\β}),k=k_3\end{array}\right.$ K=k ₂Wherein, Y, k represent the ordering grade of described advertising creative, take Baidu as example, advertising creative can be divided into Three Estate, are respectively 1,2,3.P (Y=k|X) expression Y gets the probability of k value, and Φ (x) is the distribution density function of standardized normal distribution, c ₁, c ₂Be the threshold value of described model, β=(β ₀, β ₁..., β ₇) be the regression coefficient of described model; X=(x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇), x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇Represent successively above-mentioned seven property values.

The solving result of above-mentioned model has represented that the quality degree of advertising creative is respectively 1,2,3 o'clock probability.Among the present invention, ordering grade corresponding to the value of maximum probability is

(formula 2) is the ranking value of target data.

Need to prove, at the beginning of model creation, the threshold value c of model ₁, c ₂And the regression coefficient β of model=(β ₀, β ₁..., β ₇) all be unknown.Among the present invention, can find the solution above-mentioned model threshold and model regression coefficient by bringing known historical sample data into.Concrete, can according to Maximum Likelihood Estimation, construct the likelihood function of described sequencing regression model $L\left(\mathrm{\β}\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\log \left\{P\left(Y_i\right|X_i)\right\}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{k=1}{\overset{3}{\mathrm{\Σ}}}I(Y_i=k)\log \left\{p_k({X_i}^{\′},\mathrm{\β},c)\right\},$ C=(c ₁, c ₂), wherein; (X _i, Y _i) i sample data of expression.Obtain historical property value and N corresponding history ordering grade of described other data of N other data; Described N is positive integer; Take i described historical ordering grade as Y _i, take i historical property value as X _i, find the solution and make described likelihood function reach peaked β, c ₁, c ₂

After finding the solution above-mentioned value, utilize formula $p(Y=k|X)=\left\{\begin{array}{c}\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),k=k_1\\ \mathrm{\Φ}(c_2-X^{\′}\mathrm{\β})-\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),\\ 1-\mathrm{\Φ}(c_2-X^{\′}\mathrm{\β}),k=k_3\end{array}\right.$ K=k ₂And known X value can be tried to achieve corresponding Y value.

Need to prove that corresponding different advertisement URL finds the solution the β, the c that obtain ₁, c ₂Value is different, i.e. each series advertisements URL, a corresponding sequencing regression model.Therefore, in the present invention, can other property values of this advertising creative be brought in the corresponding sequencing regression model according to the classifying and numbering of an advertising creative URL and find the solution.

S19, to judge whether described ranking value meets pre-conditioned, if meet, then submits to described advertising creative to search engine to sort.

Such as, can set pre-conditioned for ranking value greater than 2, so by the Y value that obtains is judged, can determine whether an advertising creative to be committed in the search engine ordering and throw in.

Further, when the ranking value of an advertising creative is low, can by with same popularization unit in 7 property values of the high advertising creative of ranking value compare, find that the lower advertisement of ranking value needs improved aspect, and with this it be optimized.

Above-mentioned all is to describe the present invention as an example of advertising creative example.In actual applications, also have other data similarly to sort.For this, the embodiment of the invention 2 provides a kind of target data sort method, and referring to Fig. 4, the method comprises:

S21, obtain seven property values of described target data.

Wherein, described seven property values that obtain described target data comprise:

S211, obtain the keyword clicking rate of described target data.

S212, obtain the match pattern expressivity value that described keyword arranges in search engine; Wherein, described match pattern is presented as broad match, phrase match, exact matching.

S213, judge the expressivity value that whether comprises asterisk wildcard in the described target data and generate judged result.

S214, obtain the classifying and numbering of the web page address URL of described target data.

S215, obtain the matching degree of title and the described keyword of described target data.

S216, the description 2 of obtaining described target data and the matching degree of described keyword.

S217, obtain the scale expressivity value of unit, described keyword place.

S22, according to described seven property values, calculate the ranking value of described target data.

S23, to judge whether described ranking value meets pre-conditioned, if meet, then submits to described target data to ordering system to sort.

Concrete, with described seven property values input sequencing regression model; Wherein, described sequencing regression model is specially:

$p(Y=k|X)=\left\{\begin{array}{c}\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),k=k_1\\ \mathrm{\Φ}(c_2-X^{\′}\mathrm{\β})-\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),\\ 1-\mathrm{\Φ}(c_2-X^{\′}\mathrm{\β}),k=k_3\end{array}\right.$ K=k ₂Wherein, Y, k represent the ordering grade of described target data, and p (Y=k|X) expression Y gets the probability of k value; Φ (x) is the distribution density function of standardized normal distribution, c ₁, c ₂Be the threshold value of described model, β=(β ₀, β ₁..., β ₇) be the regression coefficient of described model; X=(x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇), described x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇Represent successively above-mentioned seven property values.

According to described sequencing regression model, calculate described target data and get different ordering grade k ₁, k ₂, k ₃The time probable value.

Determine that maximum ordering grade corresponding to probable value is the ordering grade of described target data.

Judge whether described ordering grade conforms to a predetermined condition, if meet, submit to described target data to ordering system to sort.

Embodiment is similar with advertising creative, and described method also comprises the threshold value c that finds the solution described model ₁, c ₂With described regression coefficient β=(β ₀, β ₁..., β ₇): specifically comprise:

According to Maximum Likelihood Estimation, construct the likelihood function of described sequencing regression model $L\left(\mathrm{\β}\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\log \left\{P\left(Y_i\right|X_i)\right\}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{k=1}{\overset{3}{\mathrm{\Σ}}}I(Y_i=k)\log \left\{p_k({X_i}^{\′},\mathrm{\β},c)\right\},$ C=(c ₁, c ₂), wherein; (X _i, Y _i) i sample data of expression.

Obtain historical property value and N corresponding history ordering grade of described other data of N other data; Described N is positive integer.

Take i described historical ordering grade as Y _i, take i historical property value as X _i, find the solution and make described likelihood function reach peaked (β, c).

Corresponding said method embodiment, the embodiment of the invention 3 also provides a kind of target data collator, and referring to Fig. 5, this device comprises:

Property value acquiring unit 11 is for seven property values that obtain described target data.

Ranking value computing unit 12 is used for according to described seven property values, calculates the ranking value of described target data.

Judging unit 13, it is pre-conditioned to be used for judging whether described ranking value meets.

Commit unit 14 is used for meeting in described ranking value and submits to described target data to ordering system to sort when pre-conditioned.

Wherein, described property value acquiring unit 11 specifically comprises:

Keyword clicking rate unit 111 is for the keyword clicking rate of obtaining described target data.

Match pattern unit 112 is used for obtaining the match pattern expressivity value that described keyword arranges at search engine; Described match pattern is presented as broad match, phrase match, exact matching.

Asterisk wildcard unit 113 is used for judging whether described target data comprises asterisk wildcard and generate the expressivity value of judged result.

Web page address unit 114 is for the classifying and numbering of the web page address URL that obtains described target data.

Title matching unit 115 is for the matching degree of the title and the described keyword that obtain described target data.

The title of advertising creative and the matching degree of described keyword can represent with the ratio of the number of characters of the key word that contains in the title.Corresponding, described title matching unit comprises:

The title extraction unit is for the title that extracts described target data;

Title matching degree computing unit is used for calculating the character number that occurs that described title contains and the ratio of the total number of characters of described title in keyword, described ratio is the matching degree of title and the described keyword of described target data.

Describe 2 matching units 116, be used for obtaining the description 2 of described target data and the matching degree of described keyword.The description 2 of advertising creative can represent with the ratio of the number of characters of describing the key word that contains in 2 with the matching degree of keyword.Corresponding, described description 2 matching units comprise:

Describe 2 extraction units, be used for the description 2 of extracting described target data.

Describe 2 matching units, be used for calculating the ratio of the character number that occurs that described description 2 contains and total number of characters of described description 2 in keyword, described ratio is the description 2 of described target data and the matching degree of described keyword.

Scale unit 117 is for the scale expressivity value of obtaining unit, described keyword place.

Concrete, the scale expressivity value of unit, described keyword place can obtain by the number of judgement unit, place keyword and the relation of predetermined threshold value, corresponding, described scale unit comprises:

Similar keyword unit is for the similar keyword number that obtains unit, described keyword place.

Judging unit is used for judging whether described number surpasses predetermined threshold value; If surpass, determine that then described expressivity value is 0, if do not surpass, determine that then described expressivity value is 1.

Among the present invention, the ranking value computing unit can calculate the ranking value of advertising creative in several ways.In a preferred embodiment, can make up the sequencing regression model according to above-mentioned seven property values, obtain the ranking value of advertising creative by finding the solution the sequencing regression model.Corresponding, the ranking value computing unit specifically comprises:

Input block is used for described seven property values input sequencing regression model; Described sequencing regression model is specially: $p(Y=k|X)=\left\{\begin{array}{c}\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),k=k_1\\ \mathrm{\Φ}(c_2-X^{\′}\mathrm{\β})-\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),\\ 1-\mathrm{\Φ}(c_2-X^{\′}\mathrm{\β}),k=k_3\end{array}\right.$ K=k ₂Wherein, Y, k represent the ordering grade of described target data, and p (Y=k|X) expression Y gets the probability of k value; Φ (x) is the distribution density function of standardized normal distribution, c ₁, c ₂Be the threshold value of described model, β=(β ₀, β ₁..., β ₇) be the regression coefficient of described model; X=(x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇), described x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇Represent successively respectively described seven property values.

The ranking value computation subunit is used for finding the solution described sequencing regression model, calculates described target data and gets different ordering grade k ₁, k ₂, k ₃The time probable value.

Ordering classification unit is used for determining that maximum ordering grade corresponding to probable value is the ordering grade of described target data.Computing formula is specially: $\widehat{Y_i^{*}=\underset{1\≤k\≤3}{\arg \max }}P(Y_i^{*}=k|X_i^{*}).$

Need to prove, at the beginning of model creation, the threshold value c of model ₁, c ₂And the regression coefficient β of model=(β ₀, β ₁, _..., β ₇) all be unknown.Therefore, among the present invention, described device also comprises sequencing regression model construction unit, is used for making up the sequencing regression model in advance, finds the solution the threshold value c of described model ₁, c ₂With described regression coefficient β=(β ₀, β ₁..., β ₇): this unit specifically comprises:

The likelihood function unit is used for according to Maximum Likelihood Estimation, constructs the likelihood function of described sequencing regression model $L\left(\mathrm{\β}\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\log \left\{P\left(Y_i\right|X_i)\right\}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{k=1}{\overset{3}{\mathrm{\Σ}}}I(Y_i=k)\log \left\{p_k({X_i}^{\′},\mathrm{\β},c)\right\},$ C=(c ₁, c ₂), wherein; (X _i, Y _i) i sample data of expression.

The historical data unit is used for obtaining the historical property value of N other data and the history ordering grade of N correspondence of described other data; Described N is positive integer.

Likelihood function is found the solution the unit, is used for take i described historical ordering grade as Yi, take i historical property value as Xi, finds the solution and makes described likelihood function reach peaked β, c ₁, c ₂

After finding the solution above-mentioned value, utilize formula $p(Y=k|X)=\left\{\begin{array}{c}\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),k=k_1\\ \mathrm{\Φ}(c_2-X^{\′}\mathrm{\β})-\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),\\ 1-\mathrm{\Φ}(c_2-X^{\′}\mathrm{\β}),k=k_3\end{array}\right.$ K=k ₂And known X value can be tried to achieve corresponding Y value.

The same with embodiment of the method, said apparatus is specifically as follows a kind of advertising creative collator.

Need to prove that device embodiment of the present invention is corresponding with the inventive method embodiment, relevant portion reference method embodiment gets final product, and no longer is described in detail herein.

More than disclosed only be preferred implementation of the present invention; but the present invention is not limited thereto; any those skilled in the art can think do not have a creationary variation, and not breaking away from some improvements and modifications of doing under the principle of the invention prerequisite, all should drop in protection scope of the present invention.

Claims (16)

1. a target data sort method is characterized in that, described method comprises:

Obtain seven property values of described target data;

According to described seven property values, calculate the ranking value of described target data;

It is pre-conditioned to judge whether described ranking value meets, if meet, then submits to described target data to ordering system to sort;

Wherein, described seven property values that obtain described target data comprise:

Obtain the keyword clicking rate of described target data;

Obtain the match pattern expressivity value that described keyword arranges in search engine; Wherein, described match pattern is presented as broad match, phrase match, exact matching;

Judge the expressivity value that whether comprises asterisk wildcard in the described target data and generate judged result;

Obtain the classifying and numbering of the web page address URL of described target data;

Obtain the matching degree of title and the described keyword of described target data;

Obtain the description 2 of described target data and the matching degree of described keyword;

Obtain the scale expressivity value of unit, described keyword place.

2. method according to claim 1 is characterized in that, described according to described property value, the ranking value that calculates described target data comprises:

With described seven property values input sequencing regression model, find the solution described sequencing regression model, calculate described target data and get different ordering grade k ₁, k ₂, k ₃The time probable value; Wherein, described sequencing regression model is specially:

$p(Y=k|X)=\left\{\begin{array}{c}\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),k=k_1\\ \mathrm{\Φ}(c_2-X^{\′}\mathrm{\β})-\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),\\ 1-\mathrm{\Φ}(c_2-X^{\′}\mathrm{\β}),k=k_3\end{array}\right.$ K=k ₂Y, k represent the ordering grade of described target data, and p (Y=k|X) expression Y gets the probability of k value; Φ (x) is the distribution density function of standardized normal distribution, c ₁, c ₂Be the threshold value of described model, β=(β ₀, β ₁..., β ₇) be the regression coefficient of described model; X=(x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇), described x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇Represent respectively the second description part of the title of described keyword clicking rate, described match pattern, described judged result, described classifying and numbering, described target data and the matching degree of described keyword, described target data and matching degree and the described expressivity value of described keyword;

Determine that maximum ordering grade corresponding to probable value is the ranking value of described target data.

3. method according to claim 2 is characterized in that, when being broad match, described match pattern expressivity value is 0, when when the phrase match, described match pattern expressivity value is 1, when being exact matching, described match pattern expressivity value is 2.

4. method according to claim 2 is characterized in that, when described target data contained asterisk wildcard, the expressivity value of described judged result was 1, and when described target data did not contain asterisk wildcard, the expressivity value of described judged result was 0.

5. method according to claim 2 is characterized in that, describedly obtains the title of described target data and the matching degree of described keyword comprises:

Extract the title of described target data;

Calculate the character number that occurs in keyword that contains in the described title and the ratio of the total number of characters of described title, described ratio is the matching degree of title and the described keyword of described target data.

And/or;

The described description of obtaining described target data 2 comprises with the matching degree of described keyword:

Extract the description 2 of described target data;

Calculate the ratio of total number of characters of the character number that occurs in keyword that contains in the described description 2 and described description 2, described ratio is the description 2 of described target data and the matching degree of described keyword.

6. method according to claim 2 is characterized in that, the described expressivity value of obtaining the scale of unit, described keyword place comprises:

Obtain the similar keyword number of unit, described keyword place;

Judge whether described number surpasses certain threshold value;

If surpass, determine that then described scale expressivity value is 0, if do not surpass, determine that then described scale expressivity value is 1.

7. method according to claim 2 is characterized in that, described method also comprises and makes up in advance the sequencing regression model, finds the solution the threshold value c of described model ₁, c ₂With the regression coefficient β of described model=(β ₀, β ₁..., β ₇): specifically comprise:

According to Maximum Likelihood Estimation, construct the likelihood function of described sequencing regression model $L\left(\mathrm{\β}\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\log \left\{P\left(Y_i\right|X_i)\right\}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{k=1}{\overset{3}{\mathrm{\Σ}}}I(Y_i=k)\log \left\{p_k({X_i}^{\′},\mathrm{\β},c)\right\},$ C=c ₁, c ₂Wherein, (X _i, Y _i) i sample data of expression;

Obtain historical property value and N corresponding history ordering grade of described other data of N other data; Described N is positive integer;

Take i described historical ordering grade as Y _i, take i historical property value as X _i, find the solution and make described likelihood function reach peaked (β, c).

8. each described method is characterized in that according to claim 1-7, and described target data is advertising creative.

9. a target data collator is characterized in that, described device comprises:

The property value acquiring unit is for seven property values that obtain described target data;

The ranking value computing unit is used for according to described seven property values, calculates the ranking value of described target data;

Judging unit, it is pre-conditioned to be used for judging whether described ranking value meets;

Commit unit is used for meeting in described ranking value and submits to described target data to ordering system to sort when pre-conditioned;

Wherein, described property value acquiring unit specifically comprises:

Keyword clicking rate unit is for the keyword clicking rate of obtaining described target data;

The match pattern unit is used for obtaining the match pattern expressivity value that described keyword arranges at search engine; Described match pattern is presented as broad match, phrase match, exact matching;

The asterisk wildcard unit is used for judging whether described target data comprises asterisk wildcard and generate the expressivity value of judged result;

The web page address unit is for the classifying and numbering of the web page address URL that obtains described target data;

The title matching unit is for the matching degree of the title and the described keyword that obtain described target data;

Describe 2 matching units, be used for obtaining the description 2 of described target data and the matching degree of described keyword;

The scale unit is for the scale expressivity value of obtaining unit, described keyword place.

10. device according to claim 9 is characterized in that, described ranking value computing unit specifically comprises:

Input block is used for described seven property values input sequencing regression model; Described sequencing regression model is specially:

$p(Y=k|X)=\left\{\begin{array}{c}\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),k=k_1\\ \mathrm{\Φ}(c_2-X^{\′}\mathrm{\β})-\mathrm{\Φ}(c_1-X^{\′}\mathrm{\β}),\\ 1-\mathrm{\Φ}(c_2-X^{\′}\mathrm{\β}),k=k_3\end{array}\right.$ K=k ₂Wherein, Y, k represent the ordering grade of described target data, and p (Y=k|X) expression Y gets the probability of k value; Φ (x) is the distribution density function of standardized normal distribution, c ₁, c ₂Be the threshold value of described model, β=(β ₀, β ₁..., β ₇) be the regression coefficient of described model; X=(x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇), described x ₁, x ₂, x ₃, x ₄, x ₅, x ₆, x ₇Represent respectively the second description part of the title of described keyword clicking rate, described match pattern, described judged result, described classifying and numbering, described target data and the matching degree of described keyword, described target data and matching degree and the described expressivity value of described keyword;

The ranking value computation subunit is used for finding the solution described sequencing regression model, calculates described target data and gets different ordering grade k ₁, k ₂, k ₃The time probable value;

Ordering classification unit is used for determining that maximum ordering grade corresponding to probable value is the ranking value of described target data.

11. device according to claim 10 is characterized in that, when being broad match, described match pattern expressivity value is 0, when when the phrase match, described match pattern expressivity value is 1, when being exact matching, described match pattern expressivity value is 2.

12. device according to claim 10 is characterized in that, described asterisk wildcard unit, concrete being used for when described target data contains asterisk wildcard, the expressivity value that generates judged result is 1, and when described target data did not contain asterisk wildcard, the expressivity value that generates judged result was 0.

13. device according to claim 10 is characterized in that, described title matching unit comprises:

The title extraction unit is for the title that extracts described target data;

Title matching degree computing unit is used for calculating the character number that occurs that described title contains and the ratio of the total number of characters of described title in keyword, described ratio is the matching degree of title and the described keyword of described target data.

And/or;

Described description 2 matching units comprise:

Describe 2 extraction units, be used for the description 2 of extracting described target data;

Describe 2 matching units, be used for calculating the ratio of the character number that occurs that described description 2 contains and total number of characters of described description 2 in keyword, described ratio is the description 2 of described target data and the matching degree of described keyword.

14. device according to claim 10 is characterized in that, described scale unit comprises:

Similar keyword unit is for the similar keyword number that obtains unit, described keyword place;

Judging unit is used for judging whether described number surpasses certain threshold value; If surpass, determine that then described expressivity value is 0, if do not surpass, determine that then described expressivity value is 1.

15. device according to claim 10 is characterized in that, described device also comprises sequencing regression model construction unit, is used for making up the sequencing regression model in advance, finds the solution the threshold value c of described model ₁, c ₂With the regression coefficient β of described model=(β ₀, β ₁..., β ₇): specifically comprise:

The likelihood function unit is used for according to Maximum Likelihood Estimation, constructs the likelihood function of described sequencing regression model $L\left(\mathrm{\β}\right)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\log \left\{P\left(Y_i\right|X_i)\right\}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{k=1}{\overset{3}{\mathrm{\Σ}}}I(Y_i=k)\log \left\{p_k({X_i}^{\′},\mathrm{\β},c)\right\},$ C=c ₁, c ₂Wherein, (X _i, Y _i) i sample data of expression;

The historical data unit is used for obtaining the historical property value of N other data and the history ordering grade of N correspondence of described other data; Described N is positive integer;

Likelihood function is found the solution the unit, is used for take i described historical ordering grade as Y _i, take i historical property value as X _i, find the solution and make described likelihood function reach peaked (β, c).

16. each described device is characterized in that according to claim 9-15, described target data is advertising creative.

Images