CN103425677B - Keyword classification model determines method, keyword classification method and device - Google Patents
Keyword classification model determines method, keyword classification method and device Download PDFInfo
- Publication number
- CN103425677B CN103425677B CN201210156903.XA CN201210156903A CN103425677B CN 103425677 B CN103425677 B CN 103425677B CN 201210156903 A CN201210156903 A CN 201210156903A CN 103425677 B CN103425677 B CN 103425677B
- Authority
- CN
- China
- Prior art keywords
- key word
- value
- classification
- word
- characteristic index
- 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.)
- Active
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This application discloses a kind of keyword classification model and determine method, keyword classification method and device, including: the basic data of website based on storage, the characteristic index value of each key word that the recommendation information of the business object determined in keyword set on this website is issued;And the implication characterized according to each key word, determining the sample value of each key word, the sample value of key word characterizes this key word generic;And characteristic index value of based on each key word and sample value, use and set sorting algorithm, determine keyword classification model.Accordingly, also include: the basic data of website based on storage, determine the characteristic index value of designated key word;And characteristic index value of based on this designated key word, use this keyword classification model, this designated key word is classified.The scheme using the embodiment of the present application to provide, improves for issuing the accuracy that the key word that used of recommendation information carries out classifying on website.
Description
Technical field
The application relates to Internet technical field, particularly relates to a kind of keyword classification model and determines that method, key word divide
Class method and device.
Background technology
In existing Internet technology, website typically can be issued the information of some business objects, for logging in this website
User browse, and further for the post-treatment operations of specific transactions object.Such as, with e-commerce website it is
Example, business object can be specifically the product that seller user issues, and the information of business object can be specifically the description letter of product
Breath, the attribute information of product, and the purchase information etc. of product, the user that browses logging in e-commerce website can be by browsing
The various information of release product, understand the details of this product, it is possible to further perform collection, buy or recommend it
He users etc. process operation;As a example by community website, business object can be specifically the model that community users is issued, business object
Information can be specifically the description information of model, the content information etc. of model, the browsing user and can lead to of website, login community
Cross the various information of the model browsing issue, understand the details of this model, it is possible to further perform collection, money order receipt to be signed and returned to the sender or
Recommend other users etc. and process operation.
At present, website provides the user of business object, browse its business issued to attract more to browse user
The information of object, it will usually recommendation information is set for business object, better simply can be by straight for the description information of business object
Connect as recommendation information, and by issuing the recommendation information of this business object on website, browse for browsing user, in order to attract
Browse user and further browse the out of Memory of this business object.Such as, when business object is product, it is recommended that information i.e. phase
When in advertising message.The issue application of the recommendation information of business object, applies more and more extensive in current internet site,
Especially use in e-commerce website is the most universal.
But, owing to the resource-constrained that the recommendation information of business object is issued can be carried out on website, and user on website
The quantity of the business object provided is relatively big, and the amount of the recommendation information that request is issued is the biggest, so cannot realize for each use
The recommendation information of all business objects at family is all issued.
In order to solve this problem, it is achieved selectively recommendation information is issued, sending out for recommendation information in prior art
Cloth is provided with recommendation issue condition, only sets issue condition when the recommendation information of a business object of a user meets
Time, just the recommendation information of this business object of this user can be issued.Such as, recommendation is realized by following treatment mechanism
The issue of breath:
The user providing business object will need the recommendation information issued to join in the recommendation unit of self, and is each
Recommendation information binding key word.When browsing user and using a key word to scan on website, will be according to presetting
Selection strategy, the recommendation information bound with this key word provided from each user, select this recommendation that will issue
Breath, and the selected recommendation information bound with this key word is issued, browse for browsing user.
In actual applications, user is in order to the recommendation information of the business object self provided to greatest extent is at net
Standing and above issue, the recommendation information of the business object often provided with it by substantial amounts of key word is bound.But, some are crucial
The scope ratio of the business object that word is contained is wide, and when using this class keywords to scan for, its search intention is inconspicuous, such as
" customizing ", when using this class keywords to scan for, the business object that the recommendation information that represented is corresponding, is the most not this
The business object that the user that scans for is interested, i.e. realizes this recommendation information relatively low from the conversion ratio being presented to click on, and
Relatively low from the conversion ratio clicking feedback, thus cause the recommendation effect using this key word issue recommendation information poor.This kind of
Key word is currently referred to as wide in range word.
On the other hand, the key word quantity that the recommendation information of the business object owing to providing with user carries out binding is relatively big,
Cause when the recommendation information of the business object that user provides is issued, need data volume to be processed relatively big, thus need
Process resource more, and treatment effeciency is relatively low;Further, when using wide in range word to issue recommendation information, owing to cannot obtain relatively
Good recommendation effect, also causes the utilization rate processing resource relatively low.
Another further aspect, in current practice, Website server can be from substantial amounts of key word, according to certain recommendation plan
Slightly, recommend some key words to the user providing business object, for carrying out with the recommendation information of the business object of user's offer
Binding.But, owing to recommendable key word quantity is relatively big, cause Website server selecting recommendation from substantial amounts of key word
During to the key word of user, the process resource of needs is more, and treatment effeciency is relatively low.
If it is possible to some key words belonging to wide in range word are accurately distinguished, will reduce Website server to
The quantity of optional key word during user's recommended keywords, thus reduce the consumption processing resource, and improve treatment effeciency;And
The issue using this class keywords to carry out recommendation information can also be avoided so that when carrying out recommendation information issue on website, subtract
Process the consumption of resource less, and improve resource utilization.
Summary of the invention
In view of this, the embodiment of the present application provide a kind of keyword classification model determine method, keyword classification method and
Device, carries out classifying inaccurate for issuing the key word that used of recommendation information on website in prior art for solving
Problem, and process the bigger problem of consumption of resource.
The embodiment of the present application is achieved through the following technical solutions:
The embodiment of the present application provides a kind of keyword classification model and determines method, including:
The basic data of website based on storage, pushing away of the business object determined in keyword set on described website
Recommending the characteristic index value of each key word that information is issued, the described characteristic index value of key word characterizes and uses the issue of this key word to push away
Recommend the recommendation effect after information;
The implication characterized according to described each key word, determines the sample value of described each key word, the described sample of key word
Value characterizes this key word generic;
Described characteristic index value based on described each key word and described sample value, use and set sorting algorithm, determine pass
Keyword disaggregated model.
The embodiment of the present application additionally provides a kind of keyword classification method based on above-mentioned keyword classification model, including:
The basic data of website based on storage, determines the characteristic index value of designated key word;
Characteristic index value based on described designated key word, uses described keyword classification model, to described designated key
Word is classified.
The embodiment of the present application additionally provides a kind of keyword recommendation method based on above-mentioned keyword classification method, including:
Use described keyword classification method, each key word in designated key set of words is carried out classification process;
In each key word from described designated key set of words, determine and belong to the key word specifying classification;
According to setting Generalization bounds, belong to the key word specifying classification from described, select key word to recommend in website
The user of upper offer business object.
The embodiment of the present application additionally provides a kind of keyword classification model and determines device, including:
Fisrt feature determines unit, for the basic data of website based on storage, determines in keyword set for institute
State the characteristic index value of each key word that the recommendation information of business object on website is issued, the described characteristic index value of key word
Characterize the recommendation effect after using this key word to issue recommendation information;
Sample value determines unit, for the implication characterized according to described each key word, determines the sample of described each key word
Value, the described sample value of key word characterizes this key word generic;
Model determines unit, and for described characteristic index value based on described each key word and described sample value, employing sets
Determine sorting algorithm, determine keyword classification model.
The embodiment of the present application additionally provides a kind of keyword classification device based on above-mentioned keyword classification model, including:
Second feature determines unit, for the basic data of website based on storage, determines that the feature of designated key word refers to
Scale value;
Taxon, for characteristic index value based on described designated key word, uses described keyword classification model, right
Described designated key word is classified.
The embodiment of the present application additionally provides a kind of key word recommendation apparatus based on above-mentioned keyword classification method, including:
Classification processing unit, is used for using described keyword classification method, to each key word in designated key set of words
Carry out classification process;
Key word determines unit, in each key word from described designated key set of words, determines and belongs to appointment
The key word of classification;
Recommendation unit, for according to setting Generalization bounds, belongs to the key word specifying classification from described, selects key word
Recommend the user that business object is provided on website.
In the technique scheme that the embodiment of the present application provides, when determining keyword classification model, it is primarily based on storage
The basic data of website, each key that the recommendation information of the business object determined in keyword set on this website is issued
The characteristic index value of word, this feature desired value of key word characterizes the recommendation effect after using this key word to issue recommendation information
Really, and the implication characterized according to each key word, determining the sample value of each key word, the sample value of key word characterizes this key word
Generic, is then based on the characteristic index value of each key word and the sample value determined, uses and set sorting algorithm, determine key
Word disaggregated model.Accordingly, when key word being classified based on this keyword classification model, it is primarily based on the website of storage
Basic data, determine the characteristic index value of a designated key word, and characteristic index value based on this designated key word, use
This keyword classification model, classifies to this designated key word such that it is able to be divided into and known sample by this designated key word
In the classification that this value is corresponding.Due to the determination of keyword classification model in such scheme, being basic data based on website, institute is really
The characteristic index value of the fixed key word that can characterize the recommendation effect after issuing recommendation information is carried out, and i.e. with reference to key word
Actually used situation, so, use this keyword classification model can key word be classified more accurately, i.e. can be more
Determine wide in range word accurately, such that it is able to get rid of the wide in range word determined from the key word bound with recommendation information, make
When must carry out recommendation information issue on website, amount of calculation is less, and then decreases the consumption processing resource, improves resource profit
By rate.
Further, in the above-mentioned keyword classification method that the embodiment of the present application is provided, when being applied to key word recommendation, permissible
First by keyword classification method, each key word in designated key set of words is carried out classification process, and closes from this appointment
In each key word in keyword set, determining and belong to the key word specifying classification, the key word of this appointment classification is non-width
The key word of general word class, and according to setting Generalization bounds, be subordinated in the key word of appointment classification, select key word to recommend
Website provides the user of business object.Due to when to user's recommended keywords, by wide in range word from designated key word
Set is got rid of, decrease can the quantity of recommended keywords so that follow-up when carrying out recommendation process, amount of calculation is less, enters
And decrease the consumption processing resource, and improve treatment effeciency.
Other features and advantage will illustrate in the following description, and, partly become from description
Obtain it is clear that or understand by implementing the application.The purpose of the application and other advantages can be by the explanations write
Structure specifically noted in book, claims and accompanying drawing realizes and obtains.
Accompanying drawing explanation
Accompanying drawing is for providing further understanding of the present application, and constitutes a part for description, implements with the application
Example is used for explaining the application together, is not intended that the restriction to the application.In the accompanying drawings:
Fig. 1 determines the flow chart of method for the keyword classification model that the embodiment of the present application provides;
The flow chart of the keyword classification method based on keyword classification model that Fig. 2 provides for the embodiment of the present application;
Fig. 3 determines the detail flowchart of method for the keyword classification model that the embodiment of the present application provides;
The detailed process of the keyword classification method based on keyword classification model that Fig. 4 provides for the embodiment of the present application
Figure;
The flow chart of the keyword recommendation method based on keyword classification method that Fig. 5 provides for the embodiment of the present application;
Fig. 6 determines the structural representation of device for the keyword classification model that the embodiment of the present application provides;
The structural representation of the keyword classification device based on keyword classification model that Fig. 7 provides for the embodiment of the present application
Figure;
The structural representation of the key word recommendation apparatus based on keyword classification method that Fig. 8 provides for the embodiment of the present application
Figure.
Detailed description of the invention
Carry out at Accurate classification, and minimizing to provide the key word used for issue recommendation information on website
The implementation of the consumption of reason resource, the embodiment of the present application provides a kind of keyword classification model and determines that method, key word divide
Class method and device, this technical scheme can apply to carry out point for issuing the key word that used of recommendation information on website
The process of class, both can be implemented as a kind of method, it is also possible to be embodied as a kind of device.Below in conjunction with Figure of description to the application
Preferred embodiment illustrate, it will be appreciated that preferred embodiment described herein is merely to illustrate and explains the application, and
It is not used in restriction the application.And in the case of not conflicting, the embodiment in the application and the feature in embodiment can phases
Combination mutually.
The embodiment of the present application provides a kind of keyword classification model to determine method, as it is shown in figure 1, include:
Step 101, the basic data of website based on storage, determine the business pair on this website in keyword set
The characteristic index value of each key word that the recommendation information of elephant is issued, the characteristic index value of key word characterizes and uses this key word to issue
Recommendation effect after recommendation information.
Step 102, the implication characterized according to each key word, determine the sample value of each key word, the sample value table of key word
Levy this key word generic.
Step 103, characteristic index value based on each key word and sample value, use and set sorting algorithm, determine key word
Disaggregated model.
Strict sequencing is not had between above-mentioned steps 101 and step 102.
Accordingly, the embodiment of the present application also provides for a kind of keyword classification method based on above-mentioned keyword classification model,
As in figure 2 it is shown, include:
Step 201, the basic data of website based on storage, determine the characteristic index value of designated key word.
Step 202, characteristic index value based on this designated key word, use this keyword classification model, close this appointment
Keyword is classified.
It is also preferred that the left in the embodiment of the present application, the basic data of website based on storage, determined by the feature of key word refer to
Scale value, can include multiple characteristic index value, such as, at least include following characteristic index value it
Classification shatter value, use the degree of depth, the meansigma methods of utilization cost, the standard deviation of utilization cost, searching times, represent
Amount, hits, clicking rate, classification click degree, the implication characterized about each characteristic index value is described in detail follow-up.
The setting sorting algorithm used in above-mentioned steps 103, can be of the prior art various raw based on sample data
Become the sorting algorithm of forecast model, it is for instance possible to use SVM(Support Vector Machine, support vector machine) algorithm
In C-SVC algorithm, determine the keyword classification model of two classification.
Below in conjunction with the accompanying drawings, the method and device provided the application with specific embodiment is described in detail.
Fig. 3 determines the detail flowchart of method for the keyword classification model that the embodiment of the present application provides, specifically include as
Lower process step:
Step 301, obtaining the basic data of website of storage, the basic data of acquisition can be used for follow-up determining for this net
The characteristic index value of the key word that the recommendation information of the business object on standing is issued.
Need data volume to be processed to effectively control to determine in keyword classification model process, specifically can obtain and set
The basic data that the section of fixing time is corresponding, setting the time period can be arranged the most flexibly, such as, be set to recently
One month.
The basic data of website, including some intrinsic attribute datas of website, such as, the business object that website is issued
The categorical data etc. of affiliated classification, it is also possible to include some record data that user browses web sites, in the embodiment of the present application, specifically
Need to obtain those basic datas, can according to follow-up it needs to be determined that characteristic index value be determined, such as, specifically can obtain
Following basic data:
The search of query web browses record, obtains and browses user when using each key word to scan for, is represented
The recommendation information of business object, the recommendation information now represented can be typically the description information of business object;And represent
Whether recommendation information is further performed clicking operation, and represents other of business object corresponding to this recommendation information and believe in detail
Breath;
Obtain the list of the key word providing the user of business object to propose utilization cost on website, i.e. carry out recommendation information
Issue the list of the key word used, and obtain the business pair corresponding with the recommendation information of each key word binding used
As, and classification belonging to the business object of this correspondence;And the use generation that user proposes can also be obtained for each key word
Valency.
Step 302, above-mentioned basic data based on the website obtained, determine that in keyword set, the feature of each key word refers to
Scale value, wherein, the characteristic index value of key word characterizes the recommendation effect after using this key word to issue recommendation information.
In the embodiment of the present application, determined by the characteristic index value of key word, can at least include following characteristic index value
One of:
Classification shatter value, use the degree of depth, the meansigma methods of utilization cost, the standard deviation of utilization cost, searching times, represent
Amount, hits, clicking rate, classification click degree, wherein:
Classification shatter value, characterizes the quantity of classification belonging to the business object corresponding with the recommendation information of this key word binding;
Belonging to the quantity of classification the biggest, represent that the scope of the business object that this key word contains is the widest, then use this key word to carry out
Search intention represented during search is the most inconspicuous, i.e. uses the recommendation effect after this key word issue recommendation information poor, instead
It, the quantity of affiliated classification is the least, then recommendation effect is the best;
In the embodiment of the present application, in order to effectively control amount of calculation, can be determined for the classification of specified level, i.e. when
When the classification of this specified level belonging to two business objects is identical, i.e. represent that classification belonging to the two business object is identical
, such as, the business object binding recommendation information corresponding with a key word includes three business objects, and three business objects
Affiliated bibliography system is respectively " ABC ", " ABD " and " EF ", and when specified level is the second level, first business object is with second
Second level classification belonging to individual business object is B, then assert classification belonging to first business object and second business object
Identical, so, the quantity of classification belonging to these three business object is 2.
Use the degree of depth, the user for issuing service object on the web site uses the quantity of the user of this key word;Make
Quantity with the user of this key word is the biggest, represents that more user wishes by this key word pushing away the business object of self
Information of recommending is issued on website, show also the business object that the user couple browsed web sites is relevant to this key word to a certain extent
Interested, so, use the recommendation effect after this key word issue recommendation information the best, otherwise, the quantity of user is the least, then
Recommendation effect is the poorest.
The meansigma methods of utilization cost, for using the meansigma methods of utilization cost that each user of this key word proposes respectively;Should
Meansigma methods is the biggest, represents more user to wish by this key word and is sent out on website by the recommendation information of the business object of self
Cloth, the business object that show also the user couple browsed web sites to a certain extent relevant to this key word is interested, so, make
The recommendation effect after recommendation information is issued the best with this key word, otherwise, meansigma methods is the least, then recommendation effect is the poorest;
In order to effectively control amount of calculation, the meansigma methods of utilization cost, the use generation that specifically can propose respectively from each user
In valency, choose the utilization cost setting quantity, and determine the meansigma methods of the selected utilization cost setting quantity.
The standard deviation of described key word utilization cost, for using utilization cost that each user of this key word proposes respectively
Standard deviation;This standard deviation is the least, represents each user that this key word proposes utilization cost, for the degree of recognition of this key word
Differ the least, so, use the recommendation effect after this key word issue recommendation information the best, otherwise, standard deviation is the biggest, then recommend
Effect is the poorest;
In order to effectively control amount of calculation, the standard deviation of utilization cost, the use generation that specifically can propose respectively from each user
In valency, choose the utilization cost setting quantity, and determine the standard deviation of the selected utilization cost setting quantity.
Searching times, for using the number of times of this key word searching service object on the web site;This number of times is the biggest, represents clear
The business object that the user couple of website of looking at is relevant to this key word is interested, so, use this key word to issue recommendation information
After recommendation effect the best, otherwise, searching times is the least, then recommendation effect is the poorest.
The amount of representing, the quantity of the recommendation information represented during for using this key word searching service object on the web site;Should
Quantity is the biggest, represents that more user provides the quantity of the business object that this recommendation information is corresponding the biggest on website, certain journey
The demand that show also the user couple the browsed web sites business object relevant to this key word on degree is more, so, use this pass
The recommendation effect that keyword is issued after recommendation information is the best, otherwise, this quantity is the least, then recommendation effect is the poorest;
Hits, after using this key word searching service object on the web site, the recommendation information represented is clicked
Number of times;This number of times is the biggest, represents that the business object that the user couple browsed web sites is relevant to this key word is interested, also illustrates that
Using the Search Results that this key word obtains after scanning for, the scope of the business object contained is the least, and i.e. search intention is more
Substantially, so, use the recommendation effect after this key word issue recommendation information the best, otherwise, this number of times is the least, then recommendation effect
The poorest;
Clicking rate, after using this key word searching service object on the web site, the recommendation information represented is clicked
Number of times, with the ratio of the quantity of the recommendation information represented;This ratio is the biggest, represents the user couple and this key word browsed web sites
Relevant business object is interested, also illustrates that the Search Results using this key word to obtain after scanning for, the industry contained
The scope of business object is the least, i.e. search intention is the most obvious, so, the recommendation effect after using this key word to issue recommendation information is got over
Good, otherwise, this number of times is the least, then recommendation effect is the poorest;
Classification click degree, characterizes the recommendation information represented after using this key word searching service object on the web site
In, the quantity of classification belonging to the business object that clicked recommendation information is corresponding;The quantity of this affiliated classification is the biggest, represents this pass
In the business object that keyword is contained, the scope of the business object interested to user browsed web sites is the widest, then use this pass
Search intention represented when keyword scans for is the most inconspicuous, the recommendation effect after i.e. using this key word to issue recommendation information
Poor, otherwise, the quantity of this affiliated classification is the least, then recommendation effect is the best;
In the embodiment of the present application, in order to effectively control amount of calculation, the determination similar to above-mentioned classification shatter value can be used
Mode, the classification for specified level is determined, and is no longer described in detail at this;
In the embodiment of the present application, the determination of classification click degree, it is also possible to reference to the determination result of above-mentioned classification shatter value, i.e.
Based in the recommendation information represented after this key word on the web site searching service object, clicked recommendation information is corresponding
The quantity of classification belonging to business object, with above-mentioned classification shatter value, uses and sets mapping function, such as weighted sum function, really
Determine classification click degree, wherein, when there is not clicked recommendation information, the business object institute that clicked recommendation information is corresponding
The quantity belonging to classification can use setting value, and this setting value is equivalent to penalty factor, specifically can carry out spirit according to actual needs
Live and arrange, be no longer described in detail at this.
Step 303, the implication characterized according to each key word in above-mentioned keyword set, determine the sample value of each key word,
The sample value of key word characterizes this key word generic.
Specifically can determine the sample value of this key word by the semantic analysis to key word, such as, concrete employing is as follows
Mode determines the sample value of a key word:
First key word is carried out word segmentation processing, obtain each word segmentation result;
Then for each word segmentation result, determine and whether this word segmentation result comprises business object core word, business object
Core word can be configured according to actual needs, such as, characterize the noun etc. of business object title;
When, in each word segmentation result, when there is the word segmentation result comprising business object core word, determining the sample of this key word
Value on the occasion of, on the occasion of characterize sample be positive sample, i.e. characterize this key word and be not belonging to wide in range word, belong to non-wide in range word one class;
When, in each word segmentation result, when there is not the word segmentation result comprising business object core word, determining the sample of this key word
This value is negative value, and it is negative sample that negative value characterizes sample, i.e. characterizes this key word and belongs to wide in range word one class.
Strict sequencing is not had between this step and above-mentioned steps 301 and step 302.
After the characteristic index value of each key word in determining keyword set and sample value, i.e. can use setting point
Class algorithm, determines keyword classification model, in the embodiment of the present application, proposes to use the C-SVC algorithm in SVM algorithm, determines
The keyword classification model of two classification, idiographic flow such as following step:
Step 304, for the above-mentioned every kind of characteristic index value determined, this kind of characteristic index value of each key word is returned
One change processes, and obtains this kind of characteristic index value after the normalization of each key word, equation below specifically can be used to carry out normalizing
Change processes:
Wherein, FijFor i-th kind of characteristic index value of jth key word, Min (F in each key wordi) it is the of each key word
Minima in i kind characteristic index value, Max (Fi) be each key word i-th kind of characteristic index value in maximum,For each pass
I-th kind of characteristic index value after the normalization of jth key word in keyword.
Characteristic index value after step 305, normalization based on each key word and sample value, determine and meet in sorting algorithm
Object function and constraints, the optimal value of the parameter in kernel function in sorting algorithm.
Wherein, the object function of sorting algorithm, employing equation below:
The constraints of sorting algorithm, employing equation below:
yi(wTφ(xi)+b)≥1-εi;
εi≥0;
Wherein, xiFor the characteristic vector being made up of each characteristic index value of the i-th key word in each key word;yiIt is i-th
The sample value of individual key word, value is+1 or-1, and+1 represents positive sample, and-1 represents negative sample;W and b is surpassing in higher dimensional space
Plane parameter, is similar to slope and the intercept of linear function in two-dimensional space;εiFor the error ginseng corresponding with i-th key word
Number, can be configured according to actual needs;C > 0 is the compensating parameter of error term;L is the number of each key word in keyword set
Amount.
SVM algorithm is for one group of sample set (xi,yi), solve the optimization problem of above-mentioned object function and constraints.
Wherein, function phi is characteristic vector x of sampleiBeing mapped to higher dimensional space, SVM solution to model is actually in higher dimensional space one
The linear separability hyperplane of individual maximization frontier distance.C > 0 is the compensating parameter of error term, K (xi,xj)=φ (xi)Tφ(xj)
Being referred to as kernel function, conventional kernel function has the most several:
Linear kernel function: K (xi,xj)=xi Txj;
Polynomial kernel function: K (xi,xj)=(gamma*xi Txj+r)d;
RBF: K (xi,xj)=exp (-gamma | | xi-xj||2);
Sigmoid kernel function: K (xi,xj)=tanh (gamma*xi Txj+r);
Wherein, gamma, r, d are nuclear parameters.
In this step, the characteristic index value after i.e. based on each key word normalization and sample value, determine that meeting classification calculates
Object function in method and constraints, the optimal value of the parameter in kernel function in sorting algorithm.Such as, the embodiment of the present application
In, Selection of kernel function uses RBF, then uses k-fold cross-validation method and gridding method iteration, obtains the core of optimum
The numerical value of function parameter gamma.
The above-mentioned optimal value of the parameter of the kernel function that step 306, use obtain, refers to the feature after the normalization of each key word
Scale value is trained, and obtains keyword classification model.
Such as, based on above-mentioned object function and constraints, and selected kernel function, use the parameter obtained
The optimal value of the parameter of gamma, is trained the characteristic index value after the normalization of each key word, obtains keyword classification model
As follows:
(wTφ(x)+b);
Wherein, w and b is the hyperplane parameter utilizing the characteristic vector matching of sample to obtain, and φ is the kernel function factor, and x is
The characteristic vector of each characteristic index value composition of key word to be sorted.
By above-mentioned steps 301-step 306, after determining the keyword classification model for key word is classified,
Follow-up can use this keyword classification model that key word is classified, as shown in Figure 4, specifically include and process step as follows:
Step 401, obtaining the basic data of website of storage, the basic data of acquisition can be used for follow-up determining for this net
The characteristic index value of the key word that the recommendation information of the business object on standing is issued.
In this step, the concrete basic data obtained can refer to the description in above-mentioned steps 301, no longer carries out at this in detail
Describe.
Step 402, based on the above-mentioned basic data of website obtained, determine the spy of designated key word needing to carry out to classify
Levy desired value.
The characteristic index value specifically determined in this step can refer to the description in above-mentioned steps 302, no longer carries out at this in detail
Thin description.
Step 403, for the above-mentioned every kind of characteristic index value determined, and based on a determination that key word during keyword classification model
This kind of characteristic index value of each key word in set, is normalized this kind of characteristic index value of designated key word,
Obtain this kind of characteristic index value after the normalization of designated key word.
Characteristic index value after step 404, normalization based on designated key word, uses above-mentioned keyword classification model,
Designated key word is classified.
Such as, by the feature vector, X of designated key word, bring above-mentioned keyword classification model (w intoTφ (x)+b) in, if
(wTφ (X)+b) > 0, then X represents a positive sample, represents that this designated key word belongs to non-wide in range part of speech;If (wTφ(X)
+ b) < 0, then X represents a negative sample, represents that this designated key word belongs to wide in range part of speech.
For the knot using above-mentioned keyword classification model that some key words are classified in actual applications shown in table 1
Really:
Table 1:
Key word | Characteristic vector | End value | Result |
Colored | (11,8,1.375,1.125,1.411,104,0.0015,0.727,0.032) | Less than 0 | Wide in range word |
Fast-selling | (57,156,0.365,1.031,2.462,1726,0.00526,2.736,0.0079) | Less than 0 | Wide in range word |
Man's money scarf | (1,3,0.333,0.2667,0.153,87,0.061,0.160,1.732) | More than 0 | Non-wide in range word |
The such scheme using the embodiment of the present application to provide, when determining keyword classification model, with reference to based on website
Basic data, determined by can characterize the characteristic index value of the key word issuing the recommendation effect after recommendation information, i.e. join
Examine the actually used situation of key word, so, use this keyword classification model can key word be divided more accurately
Class, i.e. can determine wide in range word more accurately, is determined such that it is able to get rid of from the key word bound with recommendation information
Wide in range word so that carry out on website recommendation information issue time, decrease process resource consumption, improve the utilization of resources
Rate.
Based on above-mentioned keyword classification method, the embodiment of the present application also proposes this keyword classification method, be applied to
Website provide in the scheme of user's recommended keywords of business object, the concrete key word recommendation side proposed as shown in Figure 5
Method, including:
Step 501, use this keyword classification method, each key word in designated key set of words is carried out at classification
Reason.
In this step, it is intended that each key word in keyword set, it is in advance according to setting statistical, is added up
To can be as recommending user, for carrying out, with recommendation information, the key word bound, during for distinguishing follow-up actual recommendation
Can recommended keywords, each key word in this designated key set of words can be referred to as former can recommended keywords.
Wherein, set statistical such as, the user that browses of website to be existed to use various modes of the prior art
Browse web sites during the page, the key word used when scanning for, join in designated key set of words;Can also lead to
Cross the recommendation information to the business object on website or business object, carry out word segmentation processing, and from word segmentation result, filter out pass
Keyword, joins in the combination of designated key word.
In step 502, each key word from this designated key set of words, determine and belong to the key word specifying classification.
By the classification of key word is processed by above-mentioned steps 502, it may be determined that go out the classification of each key word, the application
In embodiment, based on above-mentioned keyword classification model and keyword classification method, it may be determined that the classification going out each key word is
Wide in range word class, or be non-wide in range word class, and using non-wide in range word class as above-mentioned appointment classification, determine and belong to non-width
The key word of general word class, substantially, this belongs to the key word of non-wide in range word class, i.e. as can recommended keywords, be used for pushing away
Recommend to user.
Step 503, according to set Generalization bounds, be subordinated to specify classification key word in, select key word recommend
The user of business object is provided on website.
Wherein, set Generalization bounds to be no longer described in detail at this to use various modes of the prior art.
Use the keyword recommendation method shown in above-mentioned Fig. 5, due to when to user's recommended keywords, by wide in range word
Get rid of from designated key set of words, decrease can the quantity of recommended keywords so that follow-up when carrying out recommendation process, meter
Calculation amount is less, and then decreases the consumption processing resource, and improves treatment effeciency.
Based on same inventive concept, determine method, phase according to the keyword classification model that the above embodiments of the present application provide
Ying Di, the embodiment of the present application additionally provides a kind of keyword classification model and determines device, and its structural representation as shown in Figure 6, has
Body includes:
Fisrt feature determines unit 601, for the basic data of website based on storage, determine in keyword set for
The characteristic index value of each key word that the recommendation information of the business object on described website is issued, the described characteristic index of key word
Value characterizes the recommendation effect after using this key word to issue recommendation information;
Sample value determines unit 602, for the implication characterized according to described each key word, determines the sample of described each key word
This value, the described sample value of key word characterizes this key word generic;
Model determines unit 603, for described characteristic index value based on described each key word and described sample value, uses
Set sorting algorithm, determine keyword classification model.
Further, sample value determines unit 602, specifically for key word is carried out word segmentation processing, obtains each participle knot
Really;And when, in described each word segmentation result, when there is the word segmentation result comprising business object core word, determining the sample of this key word
Value on the occasion of, on the occasion of characterize sample be positive sample;When, in described each word segmentation result, not existing and comprise dividing of business object core word
During word result, determining that the sample value of this key word is negative value, it is negative sample that negative value characterizes sample.
Further, model determines unit 603, specifically for for every kind of characteristic index value, to described each key word
Described characteristic index value is normalized, and obtains the characteristic index value after the normalization of described each key word;And based on institute
State the characteristic index value after the described normalization of each key word and described sample value, determine the object function meeting in sorting algorithm
With constraints, the optimal value of the parameter in kernel function in sorting algorithm;And use described optimal value of the parameter, to described respectively
Characteristic index value after the described normalization of key word is trained, and obtains keyword classification model.
Further, fisrt feature determines unit 601, specifically for the basic data of network based on storage, the most really
One of fixed following characteristic index value:
Classification shatter value, use the degree of depth, the meansigma methods of utilization cost, the standard deviation of utilization cost, searching times, represent
Amount, hits, clicking rate, classification click degree, wherein:
Described classification shatter value, characterizes the number of classification belonging to the business object corresponding with the recommendation information of this key word binding
Amount;
The described use degree of depth, for using the quantity of the user of this key word in the user of issuing service object on website;
The meansigma methods of described utilization cost, for using the average of utilization cost that each user of this key word proposes respectively
Value;
The standard deviation of described utilization cost, for using the standard of utilization cost that each user of this key word proposes respectively
Difference;
Described searching times, for using the number of times of this key word searching service object on the web;
The described amount of representing, the number of the recommendation information represented during for using this key word searching service object on the web
Amount;
Described hits, after using this key word searching service object on the web, the recommendation information represented
Clicked number of times;
Described clicking rate, after using this key word searching service object on the web, the recommendation information represented
Clicked number of times, with the ratio of the quantity of the recommendation information represented;
Described classification click degree, characterizes the recommendation represented after using this key word searching service object on the web
In information, the quantity of classification belonging to the business object that clicked recommendation information is corresponding.
Based on same inventive concept, the pass based on above-mentioned keyword classification model provided according to the above embodiments of the present application
Keyword sorting technique, correspondingly, the embodiment of the present application additionally provides a kind of key word based on above-mentioned keyword classification model and divides
Class device, its structural representation is as it is shown in fig. 7, specifically include:
Second feature determines unit 701, for the basic data of website based on storage, determines the feature of designated key word
Desired value;
Taxon 702, for characteristic index value based on described designated key word, uses described keyword classification mould
Type, classifies to described designated key word.
Based on same inventive concept, the key word based on keyword classification method provided according to the above embodiments of the present application
Recommendation method, correspondingly, the embodiment of the present application additionally provides a kind of key word based on above-mentioned keyword classification method and recommends dress
Putting, its structural representation as shown in Figure 8, specifically includes:
Classification processing unit 801, is used for using described keyword classification method, to each key in designated key set of words
Word carries out classification process;
Key word determines unit 802, in each key word from described designated key set of words, determines and belongs to finger
Determine the key word of classification;
Recommendation unit 803, for according to setting Generalization bounds, belongs to the key word specifying classification from described, selects to close
The user providing business object on website recommended in keyword.
The function of above-mentioned each module may correspond to the respective handling step in flow process shown in Fig. 1 to Fig. 5, the most superfluous at this
State.
In sum, the scheme that the embodiment of the present application provides, including: the basic data of website based on storage, determine pass
The characteristic index value of each key word that the recommendation information of the business object on this website is issued in keyword set, key word
Characteristic index value characterizes the recommendation effect after using this key word to issue recommendation information;And the implication characterized according to each key word,
Determining the sample value of each key word, the sample value of key word characterizes this key word generic;And it is based on each key word
Characteristic index value and sample value, use and set sorting algorithm, determine keyword classification model.Accordingly, also include: based on storage
The basic data of website, determine the characteristic index value of designated key word;And characteristic index value of based on this designated key word, adopt
With this keyword classification model, this designated key word is classified.The scheme using the embodiment of the present application to provide, improves pin
The accuracy that the key word being used issue recommendation information on website is classified.
The above-mentioned keyword classification model that embodiments herein is provided determines that device and keyword classification device can lead to
Cross computer program to realize.Those skilled in the art are it should be appreciated that above-mentioned Module Division mode is only numerous module draws
One in the mode of dividing, if being divided into other modules or not dividing module, as long as above-mentioned keyword classification model determines device
With keyword classification device, there is above-mentioned functions, all should be within the protection domain of the application.
The application is with reference to method, equipment (system) and the flow process of computer program according to the embodiment of the present application
Figure and/or block diagram describe.It should be understood that can the most first-class by computer program instructions flowchart and/or block diagram
Flow process in journey and/or square frame and flow chart and/or block diagram and/or the combination of square frame.These computer programs can be provided
Instruction arrives the processor of general purpose computer, special-purpose computer, Embedded Processor or other programmable data processing device to produce
A raw machine so that the instruction performed by the processor of computer or other programmable data processing device is produced for real
The device of the function specified in one flow process of flow chart or multiple flow process and/or one square frame of block diagram or multiple square frame now.
These computer program instructions may be alternatively stored in and computer or other programmable data processing device can be guided with spy
Determine in the computer-readable memory that mode works so that the instruction being stored in this computer-readable memory produces and includes referring to
Make the manufacture of device, this command device realize at one flow process of flow chart or multiple flow process and/or one square frame of block diagram or
The function specified in multiple square frames.
These computer program instructions also can be loaded in computer or other programmable data processing device so that at meter
Perform sequence of operations step on calculation machine or other programmable devices to produce computer implemented process, thus at computer or
The instruction performed on other programmable devices provides for realizing at one flow process of flow chart or multiple flow process and/or block diagram one
The step of the function specified in individual square frame or multiple square frame.
Obviously, those skilled in the art can carry out various change and the modification essence without deviating from the application to the application
God and scope.So, if these amendments of the application and modification belong to the scope of the application claim and equivalent technologies thereof
Within, then the application is also intended to comprise these change and modification.
Claims (12)
1. a keyword classification model determines method, it is characterised in that including:
The basic data of website based on storage, the recommendation of the business object determined in keyword set on described website
At least one the characteristic index value of each key word that breath is issued, the described characteristic index value of key word characterizes and uses this key word to send out
Recommendation effect after cloth recommendation information;
The implication characterized according to described each key word, determines the sample value of described each key word, the described sample value table of key word
Levy this key word generic;
Described characteristic index value based on described each key word and described sample value, use and set sorting algorithm, determine key word
Disaggregated model.
2. the method for claim 1, it is characterised in that the implication characterized according to key word, determines the sample of this key word
This value, specifically includes:
Key word is carried out word segmentation processing, obtains each word segmentation result;
When, in described each word segmentation result, when there is the word segmentation result comprising business object core word, determining the sample of this key word
Value on the occasion of, on the occasion of characterize sample be positive sample;
When, in described each word segmentation result, when there is not the word segmentation result comprising business object core word, determining the sample of this key word
This value is negative value, and it is negative sample that negative value characterizes sample.
3. the method for claim 1, it is characterised in that described characteristic index value based on described each key word and described
Sample value, uses and sets sorting algorithm, determine keyword classification model, specifically include:
For every kind of characteristic index value, the described characteristic index value of described each key word is normalized, obtains described
Characteristic index value after the normalization of each key word;
Characteristic index value after described normalization based on described each key word and described sample value, determine and meet in sorting algorithm
Object function and constraints, the optimal value of the parameter in kernel function in sorting algorithm;
Use described optimal value of the parameter, the characteristic index value after the described normalization of described each key word is trained, obtains
Keyword classification model.
4. the method as described in claim 1-3 is arbitrary, it is characterised in that the basic data of network based on storage, determines
The characteristic index value of key word, at least includes one of following characteristic index value:
Classification shatter value, the use degree of depth, the meansigma methods of utilization cost, the standard deviation of utilization cost, searching times, the amount of representing, point
Hit number, clicking rate, classification click degree, wherein:
Described classification shatter value, characterizes the quantity of classification belonging to the business object corresponding with the recommendation information of this key word binding;
The described use degree of depth, for using the quantity of the user of this key word in the user of issuing service object on website;
The meansigma methods of described utilization cost, for using the meansigma methods of utilization cost that each user of this key word proposes respectively;
The standard deviation of described utilization cost, for using the standard deviation of utilization cost that each user of this key word proposes respectively;
Described searching times, for using the number of times of this key word searching service object on the web;
The described amount of representing, the quantity of the recommendation information represented during for using this key word searching service object on the web;
Described hits, after using this key word searching service object on the web, the recommendation information represented is by point
The number of times hit;
Described clicking rate, after using this key word searching service object on the web, the recommendation information represented is by point
The number of times hit, with the ratio of the quantity of the recommendation information represented;
Described classification click degree, characterizes the recommendation information represented after using this key word searching service object on the web
In, the quantity of classification belonging to the business object that clicked recommendation information is corresponding.
5. a keyword classification method based on the described keyword classification model in claim 1, it is characterised in that bag
Include:
The basic data of website based on storage, determines the characteristic index value of designated key word;
Characteristic index value based on described designated key word, uses described keyword classification model, enters described designated key word
Row classification.
6. a keyword recommendation method based on the keyword classification method described in claim 5, it is characterised in that including:
Use described keyword classification method, each key word in designated key set of words is carried out classification process;
In each key word from described designated key set of words, determine and belong to the key word specifying classification;
According to setting Generalization bounds, belong to the key word specifying classification from described, select key word to recommend and carry on website
User for business object.
7. a keyword classification model determines device, it is characterised in that including:
Fisrt feature determines unit, for the basic data of website based on storage, determines in keyword set for described net
At least one the characteristic index value of each key word that the recommendation information of the business object on standing is issued, the described feature of key word refers to
Scale value characterizes the recommendation effect after using this key word to issue recommendation information;
Sample value determines unit, for the implication characterized according to described each key word, determines the sample value of described each key word, closes
The described sample value of keyword characterizes this key word generic;
Model determines unit, for described characteristic index value based on described each key word and described sample value, uses to set and divides
Class algorithm, determines keyword classification model.
8. device as claimed in claim 7, it is characterised in that described sample value determines unit, specifically for entering key word
Row word segmentation processing, obtains each word segmentation result;And when, in described each word segmentation result, there is the participle knot comprising business object core word
Time really, determine the sample value of this key word on the occasion of, be positive sample on the occasion of characterizing sample;When in described each word segmentation result, do not deposit
When comprising the word segmentation result of business object core word, determining that the sample value of this key word is negative value, it is negative that negative value characterizes sample
Sample.
9. device as claimed in claim 7, it is characterised in that described model determines unit, specifically for for every kind of feature
Desired value, is normalized the described characteristic index value of described each key word, obtains the normalization of described each key word
After characteristic index value;And the characteristic index value after described normalization of based on described each key word and described sample value, determine
Meet the object function in sorting algorithm and constraints, the optimal value of the parameter in kernel function in sorting algorithm;And make
Use described optimal value of the parameter, the characteristic index value after the described normalization of described each key word is trained, obtains key word
Disaggregated model.
10. the device as described in claim 7-9 is arbitrary, it is characterised in that described fisrt feature determines unit, specifically for base
In the basic data of network of storage, at least it is defined below one of characteristic index value:
Classification shatter value, the use degree of depth, the meansigma methods of utilization cost, the standard deviation of utilization cost, searching times, the amount of representing, point
Hit number, clicking rate, classification click degree, wherein:
Described classification shatter value, characterizes the quantity of classification belonging to the business object corresponding with the recommendation information of this key word binding;
The described use degree of depth, for using the quantity of the user of this key word in the user of issuing service object on website;
The meansigma methods of described utilization cost, for using the meansigma methods of utilization cost that each user of this key word proposes respectively;
The standard deviation of described utilization cost, for using the standard deviation of utilization cost that each user of this key word proposes respectively;
Described searching times, for using the number of times of this key word searching service object on the web;
The described amount of representing, the quantity of the recommendation information represented during for using this key word searching service object on the web;
Described hits, after using this key word searching service object on the web, the recommendation information represented is by point
The number of times hit;
Described clicking rate, after using this key word searching service object on the web, the recommendation information represented is by point
The number of times hit, with the ratio of the quantity of the recommendation information represented;
Described classification click degree, characterizes the recommendation information represented after using this key word searching service object on the web
In, the quantity of classification belonging to the business object that clicked recommendation information is corresponding.
11. 1 kinds of keyword classification devices based on the described keyword classification model in claim 7, it is characterised in that bag
Include:
Second feature determines unit, for the basic data of website based on storage, determines the characteristic index value of designated key word;
Taxon, for characteristic index value based on described designated key word, uses described keyword classification model, to described
Designated key word is classified.
12. 1 kinds of key word recommendation apparatus based on the keyword classification method described in claim 5, it is characterised in that including:
Classification processing unit, is used for using described keyword classification method, carries out each key word in designated key set of words
Classification processes;
Key word determines unit, in each key word from described designated key set of words, determines and belongs to appointment classification
Key word;
Recommendation unit, for according to setting Generalization bounds, belongs to the key word specifying classification from described, selects key word to recommend
Give the user that business object is provided on website.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210156903.XA CN103425677B (en) | 2012-05-18 | 2012-05-18 | Keyword classification model determines method, keyword classification method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210156903.XA CN103425677B (en) | 2012-05-18 | 2012-05-18 | Keyword classification model determines method, keyword classification method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103425677A CN103425677A (en) | 2013-12-04 |
CN103425677B true CN103425677B (en) | 2016-08-24 |
Family
ID=49650435
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210156903.XA Active CN103425677B (en) | 2012-05-18 | 2012-05-18 | Keyword classification model determines method, keyword classification method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103425677B (en) |
Families Citing this family (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103744940B (en) * | 2013-12-31 | 2017-05-31 | 北京百度网讯科技有限公司 | A kind of method and apparatus that recommended keywords are determined based on geographic position of information releaser |
CN104050240A (en) * | 2014-05-26 | 2014-09-17 | 北京奇虎科技有限公司 | Method and device for determining categorical attribute of search query word |
CN105138523A (en) * | 2014-05-30 | 2015-12-09 | 富士通株式会社 | Method and device for determining semantic keywords in text |
CN104077366B (en) * | 2014-06-13 | 2018-03-23 | 北京百度网讯科技有限公司 | A kind of method and apparatus for being used to determine characteristic information in the network device |
CN105630830A (en) * | 2014-11-05 | 2016-06-01 | 腾讯科技(深圳)有限公司 | Method and device for establishing information relationship list |
CN105786810B (en) * | 2014-12-16 | 2019-07-12 | 阿里巴巴集团控股有限公司 | The method for building up and device of classification mapping relations |
CN106708901B (en) * | 2015-11-17 | 2021-06-15 | 北京国双科技有限公司 | Clustering method and device for search words in website |
CN105868243A (en) * | 2015-12-14 | 2016-08-17 | 乐视网信息技术(北京)股份有限公司 | Information processing method and apparatus |
CN107644036B (en) * | 2016-07-21 | 2021-06-29 | 阿里巴巴集团控股有限公司 | Method, device and system for pushing data object |
CN108073568B (en) | 2016-11-10 | 2020-09-11 | 腾讯科技(深圳)有限公司 | Keyword extraction method and device |
CN108241629A (en) * | 2016-12-23 | 2018-07-03 | 百度在线网络技术(北京)有限公司 | Keyword group technology and device |
CN108388563B (en) * | 2017-02-03 | 2022-11-08 | 北京京东尚科信息技术有限公司 | Information output method and device |
CN108628832B (en) * | 2018-05-08 | 2022-03-18 | 中国联合网络通信集团有限公司 | Method and device for acquiring information keywords |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101446959A (en) * | 2008-12-30 | 2009-06-03 | 深圳市迅雷网络技术有限公司 | Internet-based news recommendation method and system thereof |
KR20100091680A (en) * | 2009-02-11 | 2010-08-19 | 엔에이치엔비즈니스플랫폼 주식회사 | Device and method for extrcting competition pattern of search advertisement |
CN101980210A (en) * | 2010-11-12 | 2011-02-23 | 百度在线网络技术(北京)有限公司 | Marked word classifying and grading method and system |
CN101980211A (en) * | 2010-11-12 | 2011-02-23 | 百度在线网络技术(北京)有限公司 | Machine learning model and establishing method thereof |
CN102012915A (en) * | 2010-11-22 | 2011-04-13 | 百度在线网络技术(北京)有限公司 | Keyword recommendation method and system for document sharing platform |
CN102426572A (en) * | 2011-07-05 | 2012-04-25 | 百度在线网络技术(北京)有限公司 | Method and equipment for classifying business entries |
-
2012
- 2012-05-18 CN CN201210156903.XA patent/CN103425677B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101446959A (en) * | 2008-12-30 | 2009-06-03 | 深圳市迅雷网络技术有限公司 | Internet-based news recommendation method and system thereof |
KR20100091680A (en) * | 2009-02-11 | 2010-08-19 | 엔에이치엔비즈니스플랫폼 주식회사 | Device and method for extrcting competition pattern of search advertisement |
CN101980210A (en) * | 2010-11-12 | 2011-02-23 | 百度在线网络技术(北京)有限公司 | Marked word classifying and grading method and system |
CN101980211A (en) * | 2010-11-12 | 2011-02-23 | 百度在线网络技术(北京)有限公司 | Machine learning model and establishing method thereof |
CN102012915A (en) * | 2010-11-22 | 2011-04-13 | 百度在线网络技术(北京)有限公司 | Keyword recommendation method and system for document sharing platform |
CN102426572A (en) * | 2011-07-05 | 2012-04-25 | 百度在线网络技术(北京)有限公司 | Method and equipment for classifying business entries |
Also Published As
Publication number | Publication date |
---|---|
CN103425677A (en) | 2013-12-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103425677B (en) | Keyword classification model determines method, keyword classification method and device | |
Zhu et al. | Heterogeneous hypergraph embedding for document recommendation | |
CN103870973B (en) | Information push, searching method and the device of keyword extraction based on electronic information | |
Kong et al. | Exploring dynamic research interest and academic influence for scientific collaborator recommendation | |
US11416565B2 (en) | Techniques to leverage machine learning for search engine optimization | |
CN107833117A (en) | A kind of Bayes's personalized ordering for considering label information recommends method | |
CN105787068A (en) | Academic recommendation method and system based on citation network and user proficiency analysis | |
CN111949887A (en) | Item recommendation method and device and computer-readable storage medium | |
Cho et al. | Clustering method using item preference based on RFM for recommendation system in u-commerce | |
Sun et al. | A multi-aspect user-interest model based on sentiment analysis and uncertainty theory for recommender systems | |
CN106294882A (en) | Data digging method and device | |
CN103095849B (en) | A method and a system of spervised web service finding based on attribution forecast and error correction of quality of service (QoS) | |
Flory et al. | A new web personalization decision-support artifact for utility-sensitive customer review analysis | |
CN116205675A (en) | Data acquisition method and device based on thread division | |
CN103544299A (en) | Construction method for commercial intelligent cloud computing system | |
Wang et al. | Knowledge graph attention network with attribute significance for personalized recommendation | |
Ma et al. | Content Feature Extraction-based Hybrid Recommendation for Mobile Application Services. | |
US20130332440A1 (en) | Refinements in Document Analysis | |
Jiang et al. | Cultural tourism attraction recommendation model based on optimized weighted association rule algorithm | |
CN110377841B (en) | Similarity calculation method and system applied to collaborative filtering method | |
CN106649842A (en) | Cross recommendation method based on fusion data, system and vehicle machine | |
Wei et al. | A study of customer behaviour through web mining | |
Chen et al. | Extracting deep semantic information for intelligent recommendation | |
Zhang | Web news data extraction technology based on text keywords | |
Larasatie et al. | From fingerprint to footprint: using point of interest (POI) recommendation system in marketing applications |
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 |