CN103092877A - Method and device for recommending keyword - Google Patents
Method and device for recommending keyword Download PDFInfo
- Publication number
- CN103092877A CN103092877A CN2011103467591A CN201110346759A CN103092877A CN 103092877 A CN103092877 A CN 103092877A CN 2011103467591 A CN2011103467591 A CN 2011103467591A CN 201110346759 A CN201110346759 A CN 201110346759A CN 103092877 A CN103092877 A CN 103092877A
- Authority
- CN
- China
- Prior art keywords
- keyword
- client
- word
- click volume
- attributive character
- 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
Images
Abstract
The invention provides a method and device for recommending a keyword. The method comprises the steps of: obtaining a search term of a customer; classifying the customer by utilizing a classifier trained through consuming willingness samples of customers in advance, and determining the consuming willingness of the customer; and selecting the front M keywords from a candidate keyword library according to a comprehensive ranking index to serve as the keywords recommended to the customer, wherein the comprehensive ranking index of the keywords is calculated by utilizing a matching degree of attributive characteristics of keywords and consuming willingness of the customer and semantic relevancy between the keywords and the search item, and the M is a preset positive integer. The method and device for recommending the keyword can save recommendation resource and improve recommendation success rate.
Description
[technical field]
The present invention relates to field of computer technology, particularly a kind of keyword recommendation method and device.
[background technology]
Search is promoted as a kind of successful internet advertising format, and commercial value is outstanding, and searched engine supplier adopts widely.In order to advertise on network or marketing, client (enterprise of the client who relates in the present invention for advertising or market by network) buys keyword to the search engine supplier, when the user common user of network (user who relates in the present invention for) uses this keyword to start on search engine to search for, when representing the large search result, can represent to the user client's who buys this keyword advertisement.Usually above the large search result or the right side, sequence can change according to other clients' that buy this keyword purchase payment status in the position of this advertisement.The form of advertisement typically refers to the link to this client web site.
Before the client buys keyword, usually need login keyword commending system, after the input inquiry word keyword commending system can select from the candidate keywords storehouse and this query word between the keyword that meets certain requirements of the degree of correlation, with the keyword selected according to and query word between the degree of correlation keyword that rear selection comes top n that sorts recommend the client as recommended keywords, therefrom select to buy for the client.
Yet, existing keyword recommendation method has only been considered the semantic relevancy between keyword and query word, do not consider client's consumption wish, for example, although keyword and the semantic relevancy between query word of some recommendation are higher, but the attribute of these keywords does not satisfy client's consumption wish, these keywords just can not bought by the client, wasted the recommendation resource, the client also needs just can find by inquiry repeatedly and is fit to the own keyword of consuming wish, also waste the recommendation resource, reduced the recommendation success ratio.
[summary of the invention]
The invention provides a kind of keyword recommendation method and device, recommend resource so that save, improve and recommend success ratio.
Concrete technical scheme is as follows:
A kind of keyword recommendation method, the method comprises:
A, obtain client's query word;
B, utilize in advance and to consume by the client sorter that the wish sample training goes out described client is classified, determine described client's consumption wish;
C, select the overall ranking index to come the keyword of front M as the keyword to described lead referral from the candidate keywords storehouse, wherein the overall ranking index of keyword is to utilize matching degree and the semantic relevancy between keyword and described query word between the attributive character of keyword and described client's consumption wish to calculate, and described M is default positive integer.
According to one preferred embodiment of the present invention, described step C specifically comprises:
Search from the candidate keywords storehouse and described query word between semantic relevancy satisfy the keyword that the default degree of correlation requires, select semantic relevancy to come the keyword of top n, described N is default positive integer;
Select the overall ranking index to come the individual keyword of front M as the keyword to described lead referral from N the keyword of selecting, described M is less than or equal to described N, and wherein the overall ranking index of keyword is to utilize matching degree and the semantic relevancy between keyword and described query word between the attributive character of keyword and described client's consumption wish to calculate.
According to one preferred embodiment of the present invention, described client's consumption wish comprises: the trade information of buying the bid information of keyword, the regional information of buying keyword or purchase keyword;
The attributive character of described keyword comprises: trade information under the price of keyword, the regional information of keyword or keyword.
According to one preferred embodiment of the present invention, the computing method of the matching degree between the attributive character of keyword and described client's consumption wish are:
Calculate respectively each attributive character of keyword to the satisfaction degree of described client's consumption wish, then be weighted the matching degree between the consumption wish that summation obtains the attributive character of keyword and described client; Perhaps,
Cosine similarity between the proper vector that calculating is made of each attributive character of keyword and the proper vector that is made of client's consumption wish obtains the matching degree between the attributive character of keyword and described client's consumption wish.
According to one preferred embodiment of the present invention, the overall ranking index W eight of keyword calculates according to following formula:
Weight=α * Mat+ β * Cor, described Mat are the matching degree between the attributive character of keyword and described client's consumption wish, and described Cor is the semantic relevancy between keyword and described query word, and α and β are default weighting coefficient.
According to one preferred embodiment of the present invention, the method also comprises:
M the keyword that D, calculation procedure C select estimate click volume, to described lead referral keyword the time, according to estimating click volume, a described M keyword is sorted.
According to one preferred embodiment of the present invention, the account form of estimating click volume of keyword comprises:
Mode one, keyword is carried out word segmentation processing obtain each word, determine based on the search daily record click volume that each word is corresponding respectively, again the click volume of each word is weighted the click volume of estimating that summation obtains keyword, the click volume that wherein word is corresponding is definite or definite by the webpage click amount that comprises this word in the title of adding up in the search daily record by the volumes of searches of this word of adding up in the search daily record, and the weighting parameters that each word is corresponding is determined by the competency of each word; Perhaps,
Mode two, determine the click volume of estimating of keyword by the semantic relevancy between keyword and described client's intention, estimating between click volume of semantic relevancy between keyword and described client's intention and keyword is linear, and linear dimensions obtains as sample training by having the click volume of promoting keyword and existing keyword.
A kind of keyword recommendation apparatus, this device comprises:
The query word acquiring unit is for the query word that obtains the client;
The wish determining unit be used for to be utilized in advance and to be consumed by the client sorter that the wish sample training goes out described client is classified, and determines described client's consumption wish;
The keyword recommendation unit, be used for coming the individual keyword of front M as the keyword to described lead referral from candidate keywords storehouse selection overall ranking index, wherein the overall ranking index of keyword is to utilize matching degree and the semantic relevancy between keyword and described query word between the attributive character of keyword and described client's consumption wish to calculate, and described M is default positive integer.
According to one preferred embodiment of the present invention, described keyword recommendation unit specifically comprises:
The first chooser unit, be used for from the candidate keywords storehouse search and described query word between semantic relevancy satisfy the keyword that the default degree of correlation requires, select semantic relevancy to come the keyword of top n, described N is default positive integer;
The second chooser unit, be used for selecting the overall ranking index to come the individual keyword of front M as the keyword to described lead referral from N keyword of described the first chooser unit selection, described M is less than or equal to described N, and wherein the overall ranking index of keyword is to utilize matching degree and the semantic relevancy between keyword and described query word between the attributive character of keyword and described client's consumption wish to calculate.
According to one preferred embodiment of the present invention, described client's consumption wish comprises: the trade information of buying the bid information of keyword, the regional information of buying keyword or purchase keyword;
The attributive character of described keyword comprises: trade information under the price of keyword, the regional information of keyword or keyword.
According to one preferred embodiment of the present invention, during the matching degree of described keyword recommendation unit between the attributive character of calculating keyword and described client's consumption wish, calculate respectively each attributive character of keyword to the satisfaction degree of described client's consumption wish, then be weighted the matching degree between the consumption wish that summation obtains the attributive character of keyword and described client; Perhaps,
Cosine similarity between the proper vector that calculating is made of each attributive character of keyword and the proper vector that is made of client's consumption wish obtains the matching degree between the attributive character of keyword and described client's consumption wish.
According to one preferred embodiment of the present invention, described keyword recommendation unit is calculated the overall ranking index W eight of keyword according to Weight=α * Mat+ β * Cor, described Mat is the matching degree between the attributive character of keyword and described client's consumption wish, described Cor is the semantic relevancy between keyword and described query word, and α and β are default weighting coefficient.
According to one preferred embodiment of the present invention, this device also comprises:
The keyword sequencing unit is used for calculating the click volume of estimating of M keyword that described keyword recommendation unit selects,, according to estimating click volume, a described M keyword is sorted during to described lead referral keyword in described keyword recommendation unit.
According to one preferred embodiment of the present invention, described keyword sequencing unit calculates the click volume of estimating of keyword in the following ways:
Mode one, keyword is carried out word segmentation processing obtain each word, determine based on the search daily record click volume that each word is corresponding respectively, again the click volume of each word is weighted the click volume of estimating that summation obtains keyword, the click volume that wherein word is corresponding is definite or definite by the webpage click amount that comprises this word in the title of adding up in the search daily record by the volumes of searches of this word of adding up in the search daily record, and the weighting parameters that each word is corresponding is determined by the competency of each word; Perhaps,
Mode two, determine the click volume of estimating of keyword by the semantic relevancy between keyword and described client's intention, estimating between click volume of semantic relevancy between keyword and described client's intention and keyword is linear, and linear dimensions obtains as sample training by having the click volume of promoting keyword and existing keyword.
As can be seen from the above technical solutions, after the present invention obtains client's query word, can utilize in advance and to consume by the client sorter that the wish sample training goes out the client is classified, determine client's consumption wish, and further introduce matching degree between the attributive character of keyword and client's consumption wish when recommended keywords, thereby make the consumption wish that as far as possible satisfies the client to the keyword of lead referral, reduce the inquiry times that the client finds the moment oneself consumption wish keyword, improve the recommendation success ratio, also saved the recommendation resource.
[description of drawings]
The method flow diagram that Fig. 1 provides for the embodiment of the present invention one;
The structure drawing of device that Fig. 2 provides for the embodiment of the present invention two.
[embodiment]
In order to make the purpose, technical solutions and advantages of the present invention clearer, describe the present invention below in conjunction with the drawings and specific embodiments.
Embodiment one,
The method flow diagram that Fig. 1 provides for the embodiment of the present invention one, as shown in Figure 1, the method can comprise the following steps:
Step 101: the query word that obtains the client.
This query word is generally the query word of inputting when the client logs in the keyword commending system.
Step 102: search from the candidate keywords storehouse and this query word between semantic relevancy satisfy the keyword that the default degree of correlation requires, select semantic relevancy to come the keyword of top n, N is default positive integer.
In this step, in the calculated candidate keywords database between keyword and query word the mode of semantic relevancy can adopt existing semantic relevancy computing method.For example, can select following method: after keyword and query word are carried out word segmentation processing respectively, respectively based on obtaining the weights of each word after reverse document frequency (IDF) the calculating word segmentation processing that obtains each word after word segmentation processing, obtain the proper vector of keyword and the proper vector of query word, the cosine similarity between calculating two proper vectors obtains the semantic similarity between keyword and query word; Perhaps, according to the literal registration calculating keyword between keyword and query word and the degree of correlation between query word, etc.The present invention is not limited semantic relevancy computing method herein.
Step 103: utilize in advance and to consume by the client sorter that the wish sample training goes out this client is classified, determine this client's consumption wish.
In embodiments of the present invention, utilize in advance the client to consume the wish sample training and go out sorter, training classifier is process under line.The client consumes the wish sample and can select from client's consumption data of known consumption wish, and client's consumption wish includes but not limited to: buy the trade information of the bid information of keyword, the regional information of buying keyword, purchase keyword etc.The client's that final sorter classification obtains consumption wish can embody with the form of different range, namely buys bid scope under the bid of keyword, territorial scope, industry scope etc., and these scopes can be the discrete value forms, can be also the successive value forms.
Can adopt but be not limited to following sorter: boost sorter, support vector machine (SUV) sorter etc.
Client's consumption data input sorter in the past of input inquiry word is classified, just can determine this client's consumption wish.For example, determine this client and buy bid scope under the bid of keyword between 50,000 to 100,000, territorial scope in Hebei province, the industry scope is medical equipment industry etc.
There is no fixing sequencing between step 103 and step 102, can successively carry out with random order, can carry out simultaneously yet.
Step 104: the overall ranking index that utilizes matching degree between the attributive character of keyword and this client's consumption wish and the semantic relevancy between keyword and query word to calculate an above-mentioned N keyword, select the overall ranking index to come the individual keyword of front M as the keyword to this lead referral, M is the positive integer that is less than or equal to N.
In the candidate keywords storehouse, each keyword has certain attributive character, and these attributive character include but not limited to equally: trade information etc. under the price of keyword, the regional information of keyword, keyword.Determined whether the client can select this keyword to buy on the absolutely large degree of matching degree between the attributive character of these keywords and client's consumption wish, therefore, when calculating the overall ranking index of an above-mentioned N keyword, the matching degree between the attributive character of introducing keyword and client's consumption wish.
During matching degree between the attributive character of calculating keyword and client's consumption wish, can calculate respectively each attributive character and the client be consumed the satisfaction degree of wish, then be weighted summation.For example, the price of calculating respectively keyword is bought satisfaction degree, the keyword of the bid scope under the bid of keyword to this client regional information is bought the satisfaction degree, keyword of the regional information of keyword to this client under, trade information is bought the satisfaction degree of the trade information of keyword to this client, then these satisfaction degrees are weighted summation, obtain the matching degree between the attributive character of keyword and client's consumption wish.
Also can be with the attributive character constitutive characteristic of keyword vector, client's consumption wish constitutive characteristic vector calculates cosine similarity between two proper vectors as the matching degree between the attributive character of keyword and client's consumption wish.
When calculating the overall ranking index of keyword, also the matching degree between keyword and this client's consumption wish and the semantic relevancy between keyword and query word can be weighted summation, namely adopt following formula:
Weight＝α*Mat+β*Cor。(1)
Wherein, Weight is the overall ranking index of keyword, and Mat is the matching degree between the attributive character of this keyword and client's consumption wish, and Cor is the semantic relevancy between keyword and query word, and α and β are default weighting coefficient.
Then N the keyword of step 102 being selected sorts according to the overall ranking index, selects to come the individual keyword of front M as the keyword to lead referral.
Need to prove, why first execution in step 102 is selected N keyword according to semantic relevancy, select again M keyword as the keyword of recommending the client from this N keyword according to the overall ranking index, because in the process of calculating the overall ranking index, calculating keyword and client, to consume the calculated amount that between wish, matching degree expends larger, and the calculated amount of the computing semantic degree of correlation is less, can reduce the overall calculation amount like this.If do not consider the problem of calculated amount, also execution in step 102 not, but the direct overall ranking index of each keyword in the calculated candidate keywords database select the overall ranking index to come the keyword of front M as the keyword of recommending the client.
If think further to improve to recommend efficient at this, can continue to carry out following steps.
Step 105: calculate the click volume of estimating of an above-mentioned M keyword, to the lead referral keyword time, according to estimating click volume, an above-mentioned M keyword is sorted.
Calculate keyword estimate click volume the time, can be in the following way:
Mode one, can at first carry out participle to keyword and obtain each word, determine based on the search daily record click volume that each word is corresponding respectively, then the click volume of each word is weighted the click volume of estimating that summation finally obtains keyword.
Wherein, when determining the click volume of each word that participle obtains based on the search daily record, the volumes of searches that can add up word from the search daily record, volumes of searches can reflect user's click demand usually, therefore, reflects the click volume of word by the volumes of searches of word; Also can comprise the click volume of the webpage of this word in the statistics title from the search daily record, with the click volume of this click volume reflection word.
For example, certain keyword carries out obtaining term1, term2 and term3 after word segmentation processing, respectively according to top described utilization search daily record, determine the click volume of term1, term2 and term3, then the click volume with term1, term2 and term3 is weighted summation, obtains the click volume of estimating of this keyword.
Be weighted when summation, the weights of each word can determine by the competency of each word, and the competency of word can be determined by the part of speech of word, IDF etc.For example, corresponding high value of weights of noun, verb can be set, the corresponding lower value of the weights of adjective, adverbial word.
Mode two, because the final user is subjected to customer ideas corresponding to this recommendation results to affect larger on the click of certain Extended Results, that is to say, customer ideas affects the user to the click of recommendation results, correspondingly, semantic relevancy between keyword and customer ideas has just reflected the click volume of estimating of this keyword, therefore, can determine by the semantic relevancy between keyword and customer ideas the click volume of estimating of keyword.Semantic relevancy and estimate between click volume and can present linear relationship, for example:
Keyword estimate semantic relevancy * A+B between click volume=keyword and customer ideas.
Wherein, A and B are linear dimensions, the click volume that has the existing popularization keyword of promoting keyword and counting on can be gone out the value of A and B as sample training.
Client's intention can obtain from this client's popularization unit, promotes the unit and generally includes product, keyword, intention or webpage, and intention can be client's slogan, advertising words etc.
At concrete example of this measure, for example a client inputs a query word " fresh flower ", selects from the candidate keywords storehouse according to semantic relevancy to come the keyword of front 50 with the semantic relevancy of query word, can obtain following keyword:
suscribe to fresh flower on the net, buy fresh flower, the express delivery fresh flower, the fresh flower gaily decorated basket, send fresh flower on the net, the fresh flower distribution network, shopping on net birthday fresh flower, the commercial affairs fresh flower, the wedding celebration fresh flower, fresh flower is ordered net, the online purchase fresh flower, birthday fresh flower express delivery, the birthday fresh flower is subscribed, the fresh flower express delivery, fresh flower express delivery service, fresh flower express company, the wedding fresh flower, online fresh flower is ordered, whole world fresh flower, the packing fresh flower, China Post's fresh flower express delivery, fresh flower is sold, the celebration fresh flower, aircraft band fresh flower, the gift fresh flower, suscribe to the fresh flower cake on the net, buy fresh flower on the net, the price of fresh flower, China's fresh flower special delivery net, the birthday greeting fresh flower, fresh flower is sent in the strange land, flower is ordered in online fresh flower shop, the fresh flower bouquet, the fresh flower cake, postal fresh flower express delivery, international fresh flower express delivery, fresh flower qq, fresh flower shop, Haidian, the fresh flower delivery service, whole nation fresh flower, Haidian fresh flower, the marriage fresh flower, the fresh rose flower picture, fresh rose flower is wholesale, fresh flower packing picture, ONLINE birthday fresh flower, flower present express delivery net, the fresh flower express delivery, China's flower present, fresh flower shop, Fengtai.
Utilize in advance consume sorter that the wish sample training goes out this client classified by the client after, obtain the client and consume wish and be: buy the corresponding high price scope of bid scope under the bid of keyword, territorial scope is Beijing.
Utilize respectively matching degree between the attributive character of above-mentioned keyword and this client's consumption wish to calculate the overall ranking index of each keyword.
give an example, the attributive character of supposing keyword " ordered flowers " is: high price, regional information is the whole nation, the keyword prices feature is 1 to the satisfaction degree that the client buys the keyword bid so, the territorial scope satisfaction degree that regional information is bought keyword to the client is that 0.5 (suppose to overlap is 0.5, zero lap is 0, in full accord is 1), suppose that corresponding weighting coefficient is 0.5, this keyword and the client matching degree of consuming between wish can be 1*0.5+0.5*0.5=0.75 so, if between this keyword and query word, semantic relevancy is 0.8, if in formula (1), weighting coefficient α and β are respectively 0.4 and 0.6, the overall ranking index of this keyword " ordered flowers " is: 0.75*0.4+0.6*0.8=0.78.
After calculating respectively the overall ranking index of above-mentioned 50 keywords, select the overall ranking index to come the keyword of front 7 as the keyword to this lead referral: to order fresh flower, fresh flower distribution network, birthday fresh flower and subscribe, buy fresh flower, fresh flower express company, online purchase fresh flower, strange land on the net and send fresh flower.
Again these 7 keywords to lead referral are sorted: the click volume of estimating of calculating respectively each keyword, estimate the mode two of click volume as example take calculating, if this client's intention is " where buying fresh flower saves more money ", calculate respectively the semantic relevancy between 7 keywords and this intention, estimate click volume and this semantic relevancy is linear, final ranking results can for: buy fresh flower, online purchase fresh flower, ordered flowers, birthday fresh flower reservation, fresh flower distribution network, fresh flower express company on the net, fresh flower is sent in the strange land.According to this ranking results, the client recommended in these 7 keywords.
Be more than the description that method provided by the present invention is carried out, be described in detail below by two pairs of devices provided by the present invention of embodiment.
Embodiment two,
The structure drawing of device that Fig. 2 provides for the embodiment of the present invention two, as shown in Figure 2, this keyword recommendation apparatus can comprise: query word acquiring unit 200, wish determining unit 210 and keyword recommendation unit 220.
Query word acquiring unit 200 obtains client's query word.
This query word is generally the query word of inputting when the client logs in the keyword commending system.
Wish determining unit 210 is utilized in advance and to be consumed by the client sorter that the wish sample training goes out the client is classified, and determines client's consumption wish.
The client consumes the wish sample and can select from client's consumption data of known consumption wish, and client's consumption wish can include but not limited to: the trade information of buying the bid information of keyword, the regional information of buying keyword or purchase keyword.
The client's that final sorter classification obtains consumption wish can embody with the form of different range, namely buys bid scope under the bid of keyword, territorial scope, industry scope etc., and these scopes can be the discrete value forms, can be also the successive value forms.
Can adopt but be not limited to following sorter: boost sorter, support vector machine (SUV) sorter etc.
Particularly, keyword recommendation unit 220 can comprise: the first chooser unit 221 and the second chooser unit 222.
The first chooser unit 221 search from the candidate keywords storehouse and query word between semantic relevancy satisfy the keyword that the default degree of correlation requires, select semantic relevancy to come the keyword of top n, N is default positive integer.
In the calculated candidate keywords database between keyword and query word the mode of semantic relevancy can adopt existing semantic relevancy computing method.For example, can select following method: after keyword and query word are carried out word segmentation processing respectively, respectively based on obtaining the weights of each word after the IDF calculating word segmentation processing that obtains each word after word segmentation processing, obtain the proper vector of keyword and the proper vector of query word, the cosine similarity between calculating two proper vectors obtains the semantic similarity between keyword and query word; Perhaps, according to the literal registration calculating keyword between keyword and query word and the degree of correlation between query word, etc.The present invention is not limited semantic relevancy account form herein.
The second chooser unit 222 selects the overall ranking index to come the keyword conduct of front M to the keyword of lead referral from N the keyword that the first chooser unit 221 is selected, M is less than or equal to N, and wherein the overall ranking index of keyword is to utilize matching degree and the semantic relevancy between keyword and query word between the attributive character of keyword and client's consumption wish to calculate.
The attributive character of keyword can include but not limited to: trade information under the price of keyword, the regional information of keyword or keyword.
Particularly, during the matching degree of the second chooser unit 222 in keyword recommendation unit 220 between the attributive character of calculating keyword and client's consumption wish, can adopt following dual mode:
First kind of way: calculate respectively each attributive character of keyword to the satisfaction degree of client's consumption wish, then be weighted the matching degree between the consumption wish that summation obtains the attributive character of keyword and client.
The second way: calculate proper vector that each attributive character by keyword consists of and the proper vector that consisted of by client's consumption wish between the cosine similarity, obtain the matching degree between the attributive character of keyword and client's consumption wish.
The second chooser unit 222 in keyword recommendation unit 220 calculates the overall ranking index W eight of keyword according to Weight=α * Mat+ β * Cor, Mat is the matching degree between the attributive character of keyword and client's consumption wish, Cor is the semantic relevancy between keyword and query word, and α and β are default weighting coefficient.
If thinking further to improve recommends efficient, can carry out estimating of click volume to the keyword of recommending the client, and sort based on estimating click volume, this moment, this device can also comprise: keyword sequencing unit 230.
Particularly, keyword sequencing unit 230 can calculate the click volume of estimating of keyword in the following ways:
Mode one, keyword is carried out word segmentation processing obtain each word, determine based on the search daily record click volume that each word is corresponding respectively, again the click volume of each word is weighted the click volume of estimating that summation obtains keyword, the click volume that wherein word is corresponding is definite or definite by the webpage click amount that comprises this word in the title of adding up in the search daily record by the volumes of searches of this word of adding up in the search daily record, and the weighting parameters that each word is corresponding is determined by the competency of each word.
Mode two, determine the click volume of estimating of keyword by the semantic relevancy between keyword and client's intention, estimating between click volume of semantic relevancy between keyword and client's intention and keyword is linear, and linear dimensions obtains as sample training by having the click volume of promoting keyword and existing keyword.
Above-mentioned client's intention can obtain from this client's popularization unit, promotes the unit and generally includes product, keyword, intention or webpage, and intention can be client's slogan, advertising words etc.
The above is only preferred embodiment of the present invention, and is in order to limit the present invention, within the spirit and principles in the present invention not all, any modification of making, is equal to replacement, improvement etc., within all should being included in the scope of protection of the invention.
Claims (14)
1. a keyword recommendation method, is characterized in that, the method comprises:
A, obtain client's query word;
B, utilize in advance and to consume by the client sorter that the wish sample training goes out described client is classified, determine described client's consumption wish;
C, select the overall ranking index to come the keyword of front M as the keyword to described lead referral from the candidate keywords storehouse, wherein the overall ranking index of keyword is to utilize matching degree and the semantic relevancy between keyword and described query word between the attributive character of keyword and described client's consumption wish to calculate, and described M is default positive integer.
2. method according to claim 1, is characterized in that, described step C specifically comprises:
Search from the candidate keywords storehouse and described query word between semantic relevancy satisfy the keyword that the default degree of correlation requires, select semantic relevancy to come the keyword of top n, described N is default positive integer;
Select the overall ranking index to come the individual keyword of front M as the keyword to described lead referral from N the keyword of selecting, described M is less than or equal to described N, and wherein the overall ranking index of keyword is to utilize matching degree and the semantic relevancy between keyword and described query word between the attributive character of keyword and described client's consumption wish to calculate.
3. method according to claim 1, is characterized in that, described client's consumption wish comprises: the trade information of buying the bid information of keyword, the regional information of buying keyword or purchase keyword;
The attributive character of described keyword comprises: trade information under the price of keyword, the regional information of keyword or keyword.
4. according to claim 1,2 or 3 described methods, is characterized in that, the computing method of the matching degree between the attributive character of keyword and described client's consumption wish are:
Calculate respectively each attributive character of keyword to the satisfaction degree of described client's consumption wish, then be weighted the matching degree between the consumption wish that summation obtains the attributive character of keyword and described client; Perhaps,
Cosine similarity between the proper vector that calculating is made of each attributive character of keyword and the proper vector that is made of client's consumption wish obtains the matching degree between the attributive character of keyword and described client's consumption wish.
5. according to claim 1,2 or 3 described methods, is characterized in that, the overall ranking index W eight of keyword calculates according to following formula:
Weight=α * Mat+ β * Cor, described Mat are the matching degree between the attributive character of keyword and described client's consumption wish, and described Cor is the semantic relevancy between keyword and described query word, and α and β are default weighting coefficient.
6. according to claim 1,2 or 3 described methods, is characterized in that, the method also comprises:
M the keyword that D, calculation procedure C select estimate click volume, to described lead referral keyword the time, according to estimating click volume, a described M keyword is sorted.
7. method according to claim 6, is characterized in that, the account form of estimating click volume of keyword comprises:
Mode one, keyword is carried out word segmentation processing obtain each word, determine based on the search daily record click volume that each word is corresponding respectively, again the click volume of each word is weighted the click volume of estimating that summation obtains keyword, the click volume that wherein word is corresponding is definite or definite by the webpage click amount that comprises this word in the title of adding up in the search daily record by the volumes of searches of this word of adding up in the search daily record, and the weighting parameters that each word is corresponding is determined by the competency of each word; Perhaps,
Mode two, determine the click volume of estimating of keyword by the semantic relevancy between keyword and described client's intention, estimating between click volume of semantic relevancy between keyword and described client's intention and keyword is linear, and linear dimensions obtains as sample training by having the click volume of promoting keyword and existing keyword.
8. a keyword recommendation apparatus, is characterized in that, this device comprises:
The query word acquiring unit is for the query word that obtains the client;
The wish determining unit be used for to be utilized in advance and to be consumed by the client sorter that the wish sample training goes out described client is classified, and determines described client's consumption wish;
The keyword recommendation unit, be used for coming the individual keyword of front M as the keyword to described lead referral from candidate keywords storehouse selection overall ranking index, wherein the overall ranking index of keyword is to utilize matching degree and the semantic relevancy between keyword and described query word between the attributive character of keyword and described client's consumption wish to calculate, and described M is default positive integer.
9. device according to claim 8, is characterized in that, described keyword recommendation unit specifically comprises:
The first chooser unit, be used for from the candidate keywords storehouse search and described query word between semantic relevancy satisfy the keyword that the default degree of correlation requires, select semantic relevancy to come the keyword of top n, described N is default positive integer;
The second chooser unit, be used for selecting the overall ranking index to come the individual keyword of front M as the keyword to described lead referral from N keyword of described the first chooser unit selection, described M is less than or equal to described N, and wherein the overall ranking index of keyword is to utilize matching degree and the semantic relevancy between keyword and described query word between the attributive character of keyword and described client's consumption wish to calculate.
10. device according to claim 8, is characterized in that, described client's consumption wish comprises: the trade information of buying the bid information of keyword, the regional information of buying keyword or purchase keyword;
The attributive character of described keyword comprises: trade information under the price of keyword, the regional information of keyword or keyword.
11. device according to claim 8, it is characterized in that, during the matching degree of described keyword recommendation unit between the attributive character of calculating keyword and described client's consumption wish, calculate respectively each attributive character of keyword to the satisfaction degree of described client's consumption wish, then be weighted the matching degree between the consumption wish that summation obtains the attributive character of keyword and described client; Perhaps,
Cosine similarity between the proper vector that calculating is made of each attributive character of keyword and the proper vector that is made of client's consumption wish obtains the matching degree between the attributive character of keyword and described client's consumption wish.
12. 9 or 10 described devices according to claim 8,, it is characterized in that, described keyword recommendation unit is calculated the overall ranking index W eight of keyword according to Weight=α * Mat+ β * Cor, described Mat is the matching degree between the attributive character of keyword and described client's consumption wish, described Cor is the semantic relevancy between keyword and described query word, and α and β are default weighting coefficient.
13. according to claim 8,9 or 10 described devices is characterized in that, this device also comprises:
The keyword sequencing unit is used for calculating the click volume of estimating of M keyword that described keyword recommendation unit selects,, according to estimating click volume, a described M keyword is sorted during to described lead referral keyword in described keyword recommendation unit.
14. device according to claim 13 is characterized in that, described keyword sequencing unit calculates the click volume of estimating of keyword in the following ways:
Mode one, keyword is carried out word segmentation processing obtain each word, determine based on the search daily record click volume that each word is corresponding respectively, again the click volume of each word is weighted the click volume of estimating that summation obtains keyword, the click volume that wherein word is corresponding is definite or definite by the webpage click amount that comprises this word in the title of adding up in the search daily record by the volumes of searches of this word of adding up in the search daily record, and the weighting parameters that each word is corresponding is determined by the competency of each word; Perhaps,
Mode two, determine the click volume of estimating of keyword by the semantic relevancy between keyword and described client's intention, estimating between click volume of semantic relevancy between keyword and described client's intention and keyword is linear, and linear dimensions obtains as sample training by having the click volume of promoting keyword and existing keyword.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110346759.1A CN103092877B (en) | 2011-11-04 | 2011-11-04 | A kind of keyword recommendation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110346759.1A CN103092877B (en) | 2011-11-04 | 2011-11-04 | A kind of keyword recommendation method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103092877A true CN103092877A (en) | 2013-05-08 |
| CN103092877B CN103092877B (en) | 2016-09-07 |
Family
ID=48205456
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110346759.1A Active CN103092877B (en) | 2011-11-04 | 2011-11-04 | A kind of keyword recommendation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103092877B (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103440084A (en) * | 2013-09-04 | 2013-12-11 | 五八同城信息技术有限公司 | User option optimization method and device |
| CN103577595A (en) * | 2013-11-15 | 2014-02-12 | 北京奇虎科技有限公司 | Keyword pushing method and device based on current browse webpage |
| CN103577392A (en) * | 2013-11-15 | 2014-02-12 | 北京奇虎科技有限公司 | Keyword pushing method and device based on current browse webpage |
| CN103744937A (en) * | 2013-12-31 | 2014-04-23 | 百度在线网络技术（北京）有限公司 | Method and device used for providing advices to inquire sequences |
| CN104424291A (en) * | 2013-09-02 | 2015-03-18 | 阿里巴巴集团控股有限公司 | Method and device for sorting search results |
| CN104424342A (en) * | 2013-09-11 | 2015-03-18 | 携程计算机技术（上海）有限公司 | Method for keyword matching, and device, server and system of method |
| CN104601635A (en) * | 2013-10-30 | 2015-05-06 | 北京千橡网景科技发展有限公司 | Method and equipment for intelligently adjusting friend recommendation frequency |
| CN104657882A (en) * | 2015-02-06 | 2015-05-27 | 百度在线网络技术（北京）有限公司 | Method and device for obtaining popularization effect data |
| CN105045901A (en) * | 2015-08-05 | 2015-11-11 | 百度在线网络技术（北京）有限公司 | Search keyword push method and device |
| CN105930488A (en) * | 2016-05-03 | 2016-09-07 | 乐视控股（北京）有限公司 | Information search processing method and apparatus |
| CN106445965A (en) * | 2015-08-11 | 2017-02-22 | 阿里巴巴集团控股有限公司 | Information popularization processing method and apparatus |
| CN106649738A (en) * | 2016-12-23 | 2017-05-10 | 北京奇虎科技有限公司 | Method and device for aggregating personage information message in search engine result page |
| CN106651415A (en) * | 2015-10-30 | 2017-05-10 | 北京国双科技有限公司 | Keyword quality score detecting method and device |
| WO2017121314A1 (en) * | 2016-01-12 | 2017-07-20 | 腾讯科技（深圳）有限公司 | Information recommendation method and apparatus |
| CN107169045A (en) * | 2017-04-19 | 2017-09-15 | 中国人民解放军国防科学技术大学 | A kind of query word method for automatically completing and device based on temporal signatures |
| CN107463600A (en) * | 2017-06-12 | 2017-12-12 | 百度在线网络技术（北京）有限公司 | Advertisement putting keyword recommendation method and device, advertisement placement method and device |
| CN107704500A (en) * | 2017-08-28 | 2018-02-16 | 昆明理工大学 | A kind of news category method based on semantic analysis Yu the multiple cosine law |
| CN108073633A (en) * | 2016-11-17 | 2018-05-25 | 百度在线网络技术（北京）有限公司 | The bid of keyword and matching way setting method and device |
| CN108073794A (en) * | 2016-11-18 | 2018-05-25 | 百度在线网络技术（北京）有限公司 | Method for authenticating and device |
| CN108228794A (en) * | 2017-12-29 | 2018-06-29 | 三角兽（北京）科技有限公司 | Apparatus for management of information, information processing unit and automatically reply/comment method |
| CN108427756A (en) * | 2018-03-16 | 2018-08-21 | 中国人民解放军国防科技大学 | Personalized query word completion recommendation method and device based on same-class user model |
| CN108664508A (en) * | 2017-03-31 | 2018-10-16 | 百度在线网络技术（北京）有限公司 | Information-pushing method and device |
| CN109872195A (en) * | 2019-03-08 | 2019-06-11 | 北京搜狐新媒体信息技术有限公司 | A kind of data processing method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5983214A (en) * | 1996-04-04 | 1999-11-09 | Lycos, Inc. | System and method employing individual user content-based data and user collaborative feedback data to evaluate the content of an information entity in a large information communication network |
| CN101479760A (en) * | 2006-06-28 | 2009-07-08 | 微软公司 | Online keyword buying, advertisement and marketing |
| CN101673286A (en) * | 2008-09-08 | 2010-03-17 | 索尼株式会社 | Apparatus, method and computer program for content recommendation and recording medium |
| CN101957819A (en) * | 2009-07-21 | 2011-01-26 | 北京大学 | Place name searching method and system based on context |
-
2011
- 2011-11-04 CN CN201110346759.1A patent/CN103092877B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5983214A (en) * | 1996-04-04 | 1999-11-09 | Lycos, Inc. | System and method employing individual user content-based data and user collaborative feedback data to evaluate the content of an information entity in a large information communication network |
| CN101479760A (en) * | 2006-06-28 | 2009-07-08 | 微软公司 | Online keyword buying, advertisement and marketing |
| CN101673286A (en) * | 2008-09-08 | 2010-03-17 | 索尼株式会社 | Apparatus, method and computer program for content recommendation and recording medium |
| CN101957819A (en) * | 2009-07-21 | 2011-01-26 | 北京大学 | Place name searching method and system based on context |
Cited By (35)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104424291A (en) * | 2013-09-02 | 2015-03-18 | 阿里巴巴集团控股有限公司 | Method and device for sorting search results |
| CN104424291B (en) * | 2013-09-02 | 2018-12-21 | 阿里巴巴集团控股有限公司 | The method and device that a kind of pair of search result is ranked up |
| CN103440084B (en) * | 2013-09-04 | 2017-04-05 | 五八同城信息技术有限公司 | User option optimization method and device |
| CN103440084A (en) * | 2013-09-04 | 2013-12-11 | 五八同城信息技术有限公司 | User option optimization method and device |
| CN104424342A (en) * | 2013-09-11 | 2015-03-18 | 携程计算机技术（上海）有限公司 | Method for keyword matching, and device, server and system of method |
| CN104601635A (en) * | 2013-10-30 | 2015-05-06 | 北京千橡网景科技发展有限公司 | Method and equipment for intelligently adjusting friend recommendation frequency |
| CN103577595A (en) * | 2013-11-15 | 2014-02-12 | 北京奇虎科技有限公司 | Keyword pushing method and device based on current browse webpage |
| CN103577392A (en) * | 2013-11-15 | 2014-02-12 | 北京奇虎科技有限公司 | Keyword pushing method and device based on current browse webpage |
| CN103744937B (en) * | 2013-12-31 | 2018-03-23 | 百度在线网络技术（北京）有限公司 | It is a kind of to be used to provide the method and apparatus for suggesting search sequence |
| CN103744937A (en) * | 2013-12-31 | 2014-04-23 | 百度在线网络技术（北京）有限公司 | Method and device used for providing advices to inquire sequences |
| CN104657882A (en) * | 2015-02-06 | 2015-05-27 | 百度在线网络技术（北京）有限公司 | Method and device for obtaining popularization effect data |
| CN104657882B (en) * | 2015-02-06 | 2018-07-13 | 百度在线网络技术（北京）有限公司 | The acquisition methods and device of promotion effect data |
| CN105045901A (en) * | 2015-08-05 | 2015-11-11 | 百度在线网络技术（北京）有限公司 | Search keyword push method and device |
| CN105045901B (en) * | 2015-08-05 | 2019-04-30 | 百度在线网络技术（北京）有限公司 | The method for pushing and device of search key |
| CN106445965A (en) * | 2015-08-11 | 2017-02-22 | 阿里巴巴集团控股有限公司 | Information popularization processing method and apparatus |
| CN106445965B (en) * | 2015-08-11 | 2019-12-24 | 阿里巴巴集团控股有限公司 | Information popularization processing method and device |
| CN106651415A (en) * | 2015-10-30 | 2017-05-10 | 北京国双科技有限公司 | Keyword quality score detecting method and device |
| CN106651415B (en) * | 2015-10-30 | 2021-06-08 | 北京国双科技有限公司 | Keyword quality measurement detection method and device |
| US11301525B2 (en) | 2016-01-12 | 2022-04-12 | Tencent Technology (Shenzhen) Company Limited | Method and apparatus for processing information |
| WO2017121314A1 (en) * | 2016-01-12 | 2017-07-20 | 腾讯科技（深圳）有限公司 | Information recommendation method and apparatus |
| CN105930488A (en) * | 2016-05-03 | 2016-09-07 | 乐视控股（北京）有限公司 | Information search processing method and apparatus |
| CN108073633A (en) * | 2016-11-17 | 2018-05-25 | 百度在线网络技术（北京）有限公司 | The bid of keyword and matching way setting method and device |
| CN108073794B (en) * | 2016-11-18 | 2022-03-15 | 上海优扬新媒信息技术有限公司 | Authentication method and device |
| CN108073794A (en) * | 2016-11-18 | 2018-05-25 | 百度在线网络技术（北京）有限公司 | Method for authenticating and device |
| CN106649738A (en) * | 2016-12-23 | 2017-05-10 | 北京奇虎科技有限公司 | Method and device for aggregating personage information message in search engine result page |
| CN108664508A (en) * | 2017-03-31 | 2018-10-16 | 百度在线网络技术（北京）有限公司 | Information-pushing method and device |
| CN107169045A (en) * | 2017-04-19 | 2017-09-15 | 中国人民解放军国防科学技术大学 | A kind of query word method for automatically completing and device based on temporal signatures |
| CN107463600B (en) * | 2017-06-12 | 2020-12-29 | 百度在线网络技术（北京）有限公司 | Advertisement delivery keyword recommendation method and device and advertisement delivery method and device |
| CN107463600A (en) * | 2017-06-12 | 2017-12-12 | 百度在线网络技术（北京）有限公司 | Advertisement putting keyword recommendation method and device, advertisement placement method and device |
| CN107704500B (en) * | 2017-08-28 | 2021-05-14 | 昆明理工大学 | News classification method based on semantic analysis and multiple cosine theorem |
| CN107704500A (en) * | 2017-08-28 | 2018-02-16 | 昆明理工大学 | A kind of news category method based on semantic analysis Yu the multiple cosine law |
| CN108228794A (en) * | 2017-12-29 | 2018-06-29 | 三角兽（北京）科技有限公司 | Apparatus for management of information, information processing unit and automatically reply/comment method |
| CN108427756A (en) * | 2018-03-16 | 2018-08-21 | 中国人民解放军国防科技大学 | Personalized query word completion recommendation method and device based on same-class user model |
| CN108427756B (en) * | 2018-03-16 | 2021-02-12 | 中国人民解放军国防科技大学 | Personalized query word completion recommendation method and device based on same-class user model |
| CN109872195A (en) * | 2019-03-08 | 2019-06-11 | 北京搜狐新媒体信息技术有限公司 | A kind of data processing method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103092877B (en) | 2016-09-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103092877A (en) | Method and device for recommending keyword | |
| CN107562818B (en) | Information recommendation system and method | |
| CN108335137A (en) | Sort method and device, electronic equipment, computer-readable medium | |
| TWI512653B (en) | Information providing method and apparatus, method and apparatus for determining the degree of comprehensive relevance | |
| TW201501059A (en) | Method and system for recommending information | |
| CN103870973A (en) | Information push and search method and apparatus based on electronic information keyword extraction | |
| JP2014508333A (en) | Method and system for displaying cross-website information | |
| CN102663627A (en) | Personalized recommendation method | |
| EP1649395A1 (en) | Methods and systems for understanding a meaning of a knowledge item using information associated with the knowledge item | |
| CN107679103B (en) | Attribute analysis method and system for entity | |
| CN102902691A (en) | Recommending method and recommending system | |
| CN103136683A (en) | Method and device for calculating product reference price and method and system for searching products | |
| CN103294677A (en) | Searching method of electronic commerce search engine and electronic commerce search engine | |
| CN110111167A (en) | A kind of method and apparatus of determining recommended | |
| CN104408648A (en) | Method and device for choosing items | |
| CN110570233A (en) | User buyback time prediction method and device for e-commerce platform | |
| CN103984705A (en) | Search result displaying method, device and system | |
| CN110647679A (en) | Search result comprehensive ordering method under o2o scene | |
| CN103136250B (en) | Change dump recognition methods, device and information search method, system | |
| CN115496566A (en) | Regional specialty recommendation method and system based on big data | |
| Zhao et al. | Online comments of multi-category commodities based on emotional tendency analysis | |
| CN108537599B (en) | Query feedback method and device based on keyword aggregation and storage medium | |
| TW201801006A (en) | Personalized online marketing recommendation method capable of predicting the trend of preference according to factors such as browsing date, browsing time, and website visited, to thereby provide marketing materials | |
| CN115131108A (en) | E-commerce commodity screening system | |
| KR101985603B1 (en) | Recommendation method based on tripartite graph |
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 |