CN105939383A - Location information determining method and server - Google Patents

Abstract

The embodiment of the invention discloses a location information determining method. The method comprises the steps of reading geographical location information of a to-be-determined target network account, wherein the geographical location information is the location information of a user who pays attention to the target network account; obtaining feature coding parameters according to the geographical location information; receiving sample marking parameters of a network account of the user, and determining model training parameters according to the sample marking parameters and the feature coding parameters; determining actual geographical location information according to the model training parameters and the feature coding parameters, wherein the actual geographical location information is used for indicating a geographical range which is served by the target network account currently. The embodiment of the invention also provides a server. According to the method and the server provided by the embodiment of the invention, the dynamic change conditions of the user who pays attention to the target network account can be taken into consideration; the actual geographical location information can be determined according to the geographical location of the user who pays attention to the target network account; therefore, the data distortion possibility is reduced; and the accuracy of determining the actual service information of the network account by the server is improved.

Description

A kind of method that positional information determines and server

Technical field

The present invention relates to field of Internet communication, particularly relate to method and clothes that a kind of positional information determines Business device.

Background technology

Along with the development of Internet technology, network platform open ability constantly promotes, the quantity of network account Increase rapidly, and the type of network account and service progressively variation, different types of network account exists Diversified trend also occurs in regional feature.

In actual application, the geographical position generally network account registered or the ground of desired service Reason position is as the geographical position attribute of this network account, and this category information can directly obtain, or carries out Simple process can directly be applied.

But, easily there is the situations such as disappearance or distortion, Er Qieguan in the geographical position of network account registration The user of note network account is often in the process of a dynamic change, and geography during network account registration The geographical position of position or desired service is fixing, it is impossible to reflects this kind of dynamic change, thus leads Cause data distortion, and the service area that network account is actual can not be judged well.

Summary of the invention

Embodiments provide method and server that a kind of positional information determines, it may be considered that arrive The user paying close attention to this network account is in the situation of dynamically change, it is possible to according to the use paying close attention to this network account The geographical position at family determines actual geographic positional information, thus is substantially reduced the distortion probability of data, More effectively improve server and determine the accuracy of the active service information obtaining network account.

In view of this, first aspect present invention provides a kind of method that positional information determines, including:

Reading the geographical location information of objective network account to be determined, described geographical location information is for paying close attention to institute State the positional information at the user place of objective network account；

According to described geographical location information, obtain the feature coding parameter of described objective network account；

Receive the sample label parameters of user network account, and according to described sample label parameters and described Feature coding parameter determination model training parameter；

According to described model training parameter and described feature coding parameter, determine described objective network account Actual geographic positional information, described actual geographic positional information is used for indicating described objective network account to work as The geographic range of front service.

Second aspect, present aspect embodiment also provides for a kind of server, including:

Read module, for reading the geographical location information of objective network account to be determined, described geographical position Confidence breath is the positional information at the user place paying close attention to described objective network account；

Acquisition module, for the described geographical location information read according to described read module, obtains described The feature coding parameter of objective network account；

Receiver module, for receiving the sample label parameters of user network account, and according to described sample mark The described feature coding parameter determination model training parameter that note parameter and described acquisition module obtain；

Determine module, for the described model training parameter that determines according to described receiver module and described in obtain The described feature coding parameter that delivery block obtains, determines the actual geographic position letter of described objective network account Breath, described actual geographic positional information is for indicating the geographic range of described objective network account current service.

The third aspect, present aspect embodiment also provides for a kind of server, including: memorizer, transceiver, Processor and bus system；

Wherein, described memorizer is used for storing program；

Described processor is used for the program performing in described memorizer, step specific as follows:

Reading the geographical location information of objective network account to be determined, described geographical location information is for paying close attention to institute State the positional information at the user place of objective network account；

According to described geographical location information, obtain the feature coding parameter of described objective network account；

Receive the sample label parameters of user network account, and according to described sample label parameters and described Feature coding parameter determination model training parameter；

According to described model training parameter and described feature coding parameter, determine described objective network account Actual geographic positional information, described actual geographic positional information is used for indicating described objective network account to work as The geographic range of front service.

As can be seen from the above technical solutions, the embodiment of the present invention has the advantage that

In the embodiment of the present invention, it is provided that a kind of method that positional information determines, server first reads to be treated really Set the goal the geographical location information of network account, then according to geographical location information, obtains objective network account Number feature coding parameter, then server receive user network account sample label parameters, and according to Sample label parameters and feature coding parameter determination model training parameter, finally according to model training parameter And feature coding parameter, determine the actual geographic positional information of objective network account, actual geographic position Information is for indicating the geographic range of objective network account current service.By using aforesaid way to determine net The geographic range of network account current service, it may be considered that be in dynamically change to the user paying close attention to this network account Situation about changing, it is possible to determine actual geographic position according to the geographical position of the user paying close attention to this network account Information, thus be substantially reduced the distortion probability of data, more effectively improves server and determines and obtain network The accuracy of the active service information of account.

Accompanying drawing explanation

Fig. 1 is the Organization Chart that in the embodiment of the present invention, positional information determines system；

Fig. 2 is the mutual embodiment schematic diagram of method one that in the embodiment of the present invention, positional information determines；

Fig. 3 is one embodiment schematic diagram of method that in the embodiment of the present invention, positional information determines；

Fig. 4 is geographical attribute categorizing system training pattern figure in the embodiment of the present invention；

Fig. 5 is grader structural representation in the embodiment of the present invention；

Fig. 6 is one embodiment schematic diagram of server in the embodiment of the present invention；

Fig. 7 is another embodiment schematic diagram of server in the embodiment of the present invention；

Fig. 8 is another embodiment schematic diagram of server in the embodiment of the present invention；

Fig. 9 is another embodiment schematic diagram of server in the embodiment of the present invention；

Figure 10 is another embodiment schematic diagram of server in the embodiment of the present invention；

Figure 11 is another embodiment schematic diagram of server in the embodiment of the present invention；

Figure 12 is another embodiment schematic diagram of server in the embodiment of the present invention；

Figure 13 is another embodiment schematic diagram of server in the embodiment of the present invention；

Figure 14 is one structural representation of server in the embodiment of the present invention.

Detailed description of the invention

Embodiments provide method and server that a kind of positional information determines, it may be considered that arrive The user paying close attention to this network account is in the situation of dynamically change, it is possible to according to the use paying close attention to this network account The geographical position at family determines actual geographic positional information, thus is substantially reduced the distortion probability of data, More effectively improve server and determine the accuracy of the active service information obtaining network account.

Term " first " in description and claims of this specification and above-mentioned accompanying drawing, " second ", " Three ", the (if present) such as " the 4th " be for distinguishing similar object, specific without being used for describing Order or precedence.Should be appreciated that the data of so use can be exchanged in the appropriate case, in order to this In the embodiments of the invention that describe such as can suitable with in addition to those here illustrating or describing Sequence is implemented.Additionally, term " includes " and " having " and their any deformation, it is intended that cover not Exclusive comprises, such as, contain series of steps or the process of unit, method, system, product or Equipment is not necessarily limited to those steps or the unit clearly listed, but can include the most clearly listing Or for intrinsic other step of these processes, method, product or equipment or unit.

Should be understood that in the embodiment of the present invention that the control method of information pushing is applied to positional information and determines system In, referring to Fig. 1, Fig. 1 is the Organization Chart that in the embodiment of the present invention, positional information determines system, such as figure institute Showing, setting up communication connection between the master mobile terminal of service and server for providing, master mobile terminal carries For objective network account, the secondary mobile terminal of the main mobile Information Mobile Service of multiple acquisitions has paid close attention to objective network account Number, it is also possible to being known as " vermicelli " of objective network account, the service of offer is uploaded to by master mobile terminal Server, other secondary mobile terminal obtains, by paying close attention to master mobile terminal, the clothes that master mobile terminal provides Business device.

And different types of objective network account also occurs in that diversified trend in regional feature, have one A little objective network accounts vermicelli body that orients towards the whole country provides indiscriminating service, and other objective network accounts are then Vermicelli for target province or objective area provides the orientation clothes towards this locality life and local information Business, these objective network accounts all occur in that bigger difference in the geographic range that service covers, adjoint Carrying out in a deep going way of the business such as platform business promotion, search and advertisement, it is therefore desirable to understand objective network account The geographical range information of number active service.

It should be noted that master mobile terminal and secondary mobile terminal in the present invention program can be intelligence handss Machine, panel computer, personal digital assistant (English full name: Personal Digital Assistant, English contracting Write: PDA) or vehicle-mounted computer etc., should not be construed as limitation of the invention herein.

In order to make it easy to understand, refer to Fig. 2, Fig. 2 is the method that in the embodiment of the present invention, positional information determines One mutual embodiment schematic diagram, as it can be seen, in step 101, mobile terminal first sends to server The geographical location information of objective network account, objective network account can be one for providing the user clothes Public's account of business, multiple mobile terminals send the geographical location information at its place, example respectively to server As in which provinces and cities, or the information such as concrete longitude and latitude.In step 102, server can be from multiple shiftings The geographical location information that dynamic terminal provides extracts characteristic of correspondence coding parameter.The most in step 103, User can also send, to server, the sample that one or more groups user network account is corresponding by mobile terminal Label parameters, the server method by machine learning, utilize these sample label parameters to calculate model Training parameter, i.e. generates grader.Finally, server in step 105, utilizes the grader generated, And from the geographical location information of objective network account extract feature coding parameter, determine target network The actual geographic positional information of network account.

Below by from the angle of server, the method determining positional information in the present invention is introduced, please Refering to Fig. 3, one embodiment of the method that in the embodiment of the present invention, positional information determines includes:

201, reading the geographical location information of objective network account to be determined, geographical location information is that pass is gazed at The positional information at the user place of mark network account；

In the present embodiment, first server reads the geographical location information of objective network account to be determined, its In, objective network account can be the public's account providing server for multi-user, such as a microblogging account Number or wechat public number etc..And the position that geographical location information is the user place paying close attention to objective network account Confidence ceases, it is believed that be the position at objective network account " vermicelli " place.

Objective network account registration when, network operator can select place geographical position coordinates point or The country at person place, province and urban information, as the geographical position attribute of objective network account.This kind of Information is typically objective network account owner registration or the place of desired service, takes with objective network account The geographic range that pragmatic border covers often has bigger difference, and the actual geographic of objective network account vermicelli Distribution, more can reflect the geographic range that objective network account service reality covers objectively.

Wherein, the user geographic location information obtaining concern objective network account can be by with lower section Formula is extracted, such as can from Wireless Fidelity (English full name: Wireless-Fidelity, english abbreviation: WiFi), Agreement (English full name: Internet Protocol, english abbreviation: the IP) address of interconnection between network, base In location-based service (English full name: Location Based Service, english abbreviation: LBS) and user Attribute extracts user geographic location information, herein in the information such as the city at place, province or country It is not construed as limiting.

202, according to geographical location information, the feature coding parameter of objective network account is obtained；

In the present embodiment, server, according to getting the geographical location information of follower, can carry the most again Take and calculate the feature coding parameter of these geographical location information.

203, the sample label parameters of user network account is received, and according to sample label parameters and feature Coding parameter determines model training parameter；

In the present embodiment, server is in order to generate the user institute that can be used for determining concern objective network account Positional information belong to the model of any Regional Distribution type, first will receive that user sends one group Organize the sample label parameters of user network account more or, according to machine learning algorithm, utilize sample mark ginseng Number and feature coding parameter training go out a group model training parameter, i.e. obtain a classifiers.

204, according to model training parameter and feature coding parameter, objective network account is determined practically Reason positional information, actual geographic positional information is for indicating the geographic range of objective network account current service.

In the present embodiment, the model training parameter that server obtains according to training, and from paying close attention to target network The user geographic location information of network account is extracted feature coding parameter, is calculated objective network account Number actual geographic positional information, actual geographic positional information is used for indicating objective network account current service Geographic range.

Specifically, server obtains a classifiers according to model training parameter, is then joined by feature coding Number inputs to grader, and objective network account is divided into by grader according to its Regional Distribution paying close attention to crowd Several types.Calculate objective network account by this method and pay close attention to the geographical distribution attribute of user crowd, Network account is paid close attention to user crowd's geographical distribution attribute geographical position attribute as objective network account.

In the embodiment of the present invention, it is provided that a kind of method that positional information determines, server first reads to be treated really Set the goal the geographical location information of network account, then according to geographical location information, obtains objective network account Number feature coding parameter, then server receive user network account sample label parameters, and according to Sample label parameters and feature coding parameter determination model training parameter, finally according to model training parameter And feature coding parameter, determine the actual geographic positional information of objective network account, actual geographic position Information is for indicating the geographic range of objective network account current service.By using aforesaid way to determine net The geographic range of network account current service, it may be considered that be in dynamically change to the user paying close attention to this network account Situation about changing, it is possible to determine actual geographic position according to the geographical position of the user paying close attention to this network account Information, thus be substantially reduced the distortion probability of data, more effectively improves server and determines and obtain network The accuracy of the active service information of account.

Alternatively, on the basis of the embodiment that above-mentioned Fig. 3 is corresponding, the position that the embodiment of the present invention provides In first alternative embodiment of the method that information determines, feature coding parameter includes normalized vector, scale Coefficient and coefficient of kurtosis；

According to geographical location information, obtain the feature coding parameter of objective network account, may include that

According to geographical positional information calculation scale coefficient；

According to geographical positional information calculation normalized vector；

Coefficient of kurtosis is calculated according to normalized vector.

In the present embodiment, server, according to geographical location information, obtains the feature coding of objective network account Parameter is concrete it may be that server is first according to geographical positional information calculation scale coefficient, then according to geography Positional information and scale coefficient calculations obtain normalized vector, are calculated finally by normalized vector Coefficient of kurtosis,

For the ease of introducing, referring to Fig. 4, Fig. 4 is geographical attribute categorizing system instruction in the embodiment of the present invention Practicing illustraton of model, as it can be seen, block arrows represents the flow direction of data, dotted arrow presentation class device controls The flow direction of information.Geographical attribute categorizing system training pattern mainly comprises four modules, respectively features Coding module, sample labeling module, model training module and classifier modules.Wherein, feature coding mould Feature coding parameter, for obtaining the feature coding parameter of objective network account, is then separately input into by block Model training module and grader.

Secondly, in the embodiment of the present invention, the feature coding parameter bag illustrating objective network account is schematically illustrated Include scale coefficient, normalized vector and coefficient of kurtosis, and server is according to geographical location information meter Calculation scale coefficient and normalized vector, go out coefficient of kurtosis further according to normalization neighborhood calculation.Use above-mentioned Mode lifting scheme feasibility in actual applications.

Alternatively, on the basis of first corresponding for above-mentioned Fig. 3 or Fig. 3 embodiment, the present invention implements In second alternative embodiment of method that the positional information that example provides determines, according to geographical positional information calculation Scale coefficient, may include that

Calculating scale coefficient as follows:

$M_f=\underset{i=0}{\overset{n}{\Σ}}{C}_{fi}$

M_fRepresent objective network account f scale coefficient

C_fiRepresent the objective network account f number in numbered i region, wherein, C_fi≥C_f(i+1)；

For represent objective network account f at numbered total number of persons from i=0 to i=n region, N represents overall area quantity.

In the present embodiment, server uses the feature coding module in geographical attribute categorizing system training pattern Calculating scale coefficient, these scale coefficients are the user's current geographic position information paying close attention to objective network account Corresponding scale coefficient, therefore can obtain at least one scale coefficient in Practical Calculation.Here for It is easy to explanation, is only introduced as a example by calculating a scale coefficient.

Assume that current collection is as shown in table 1 below to the various places concern number of objective network account:

Table 1

Account	Province	Number
			Account f	Guangdong Province	1465
Account f	Hainan Province	474
			Account f	Shanghai City	14848
Account f	Beijing	124
			Account f	Hunan Province	1250
Account f	Jiangxi Province	187
			Account f	Henan Province	779
Account f	Heilongjiang Province	812

According to upper table, according to Regional Property, these data can be polymerized, generate one group of variable C_f1,C_f2,C_f3,C_f4,...,C_fM, wherein, C_fiRepresent the objective network account f people in numbered i region Number, i region can be specifically the region divided according to province, and n then represents the sum in region.

In order to extract more accurately pay close attention to objective network account user distribution feature, need employing number from The fewest arrangement mode, i.e. C_fi≥C_f(i+1)

The scale coefficient of calculating account f as follows:

$\begin{array}{c}{M}_f=\underset{i=0}{\overset{n}{\Σ}}{C}_{fi}\\ =1465+474+14848+124+1250+187+779+812=19939\end{array}$

Again, in the embodiment of the present invention, server can use formula to calculate scale coefficient, passes through formula Be calculated rational scale coefficient, with practicality and the feasibility of this lifting scheme.

Alternatively, on the basis of first or second embodiment that above-mentioned Fig. 3, Fig. 3 are corresponding, the present invention In the 3rd alternative embodiment of method that the positional information that embodiment provides determines, according to geographical location information Calculate normalized vector, may include that

Calculate normalized vector as follows:

$k_{fi}=C_{fi}/\underset{i=0}{\overset{n}{\Σ}}{C}_{fi}$

k_fiRepresent the normalized value corresponding to the objective network account f number in numbered i region, Duo Gegui One change value composition normalized vector, k_fiSpan be more than 0 and less than 1.

In the present embodiment, it is assumed that being calculated current scale coefficient according to table 1 is 19939, then The normalized value in each region can be calculated in conjunction with the data in table 1.

Calculate the normalized value in Guangdong Province as follows:

$\begin{array}{c}{k}_{fi}=C_{fi}/\underset{i=0}{\overset{n}{\Σ}}{C}_{fi}\\ =1465/19939=0.07347\end{array}$

The normalized value in other regions can also use aforesaid way to calculate, and multiple normalized values are last Normalized vector can be formed.

Further, in the embodiment of the present invention, on the basis of obtaining scale coefficient, it is also possible to utilize public affairs Formula is calculated normalized vector, promotes server with this and calculates the feasibility and rationally of normalized vector Property.

Alternatively, on the basis of embodiment any one of corresponding first to the 3rd of above-mentioned Fig. 3, Fig. 3, In the 4th alternative embodiment of method that the positional information that the embodiment of the present invention provides determines, according to normalization Vector calculates coefficient of kurtosis, may include that

Calculate coefficient of kurtosis as follows:

$\stackrel{\‾}{k_f}=Sum(k_{f1},k_{f2},k_{f3},k_{f4},\mathrm{...},k_{fN})/N;$

${\mathrm{Kurt}}_f=\frac{\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{(k_{fi}-\stackrel{\‾}{k_f})}^4}{{\left(\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{(k_{fi}-\stackrel{\‾}{k_f})}^2\right)}^2};$

Represent the meansigma methods of normalized value in the normalized vector that objective network account f is preset；

N represents default parameter, and the span of N is more than 0 and less than or equal to described i；

k_fNRepresent the normalized value in n-th normalized vector, (k_f1,k_f2,k_f3,k_f4,...,k_fN) represent and return One changes one group of sub-normalized vector in vector；

Kurt_fRepresenting the coefficient of kurtosis of objective network account f, coefficient of kurtosis is used for representing the steep slow degree of distribution；

Pay close attention to what the user of objective network account was distributed Fourth-order moment；

Pay close attention to what the user of objective network account was distributed second moment.

In the present embodiment, it is assumed that being calculated current scale coefficient according to table 1 is 19939, and extensively The normalized value of Dong Sheng is 0.07347, then can calculate each region continuing with the data in table 1 Coefficient of kurtosis.

First according to each data in table 1, employing equation below:

$\stackrel{\‾}{k_f}=Sum(k_{f1},k_{f2},k_{f3},k_{f4},\mathrm{...},k_{fN})/N$

It is calculated the meansigma methods of normalized value in the normalized vector that objective network account f is preset, then adopts It is calculated by equation below, the coefficient of kurtosis of objective network account f:

${\mathrm{Kurt}}_f=\frac{\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{(k_{fi}-\stackrel{\‾}{k_f})}^4}{{\left(\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{(k_{fi}-\stackrel{\‾}{k_f})}^2\right)}^2}$

Further, in the embodiment of the present invention, server be calculated scale coefficient and normalization to On the basis of amount, it is possible to use formula continues to be calculated coefficient of kurtosis, promote server with this and calculate peak The feasibility of degree coefficient and reasonability.

Alternatively, on the basis of embodiment any one of corresponding first to fourth of above-mentioned Fig. 3, Fig. 3, In the 5th alternative embodiment of method that the positional information that the embodiment of the present invention provides determines, receive user network The sample label parameters of network account, may include that

Receive one group of user network account；

One group of sample label parameters corresponding to user network account is generated according to location distribution type.

In the present embodiment, the mark sample module in geographical attribute categorizing system training pattern provides the user The interface module of one data mark, user can be by this interface by one group of user network account f⁽⁰⁾,f⁽¹⁾,f⁽²⁾,f⁽³⁾,...,f⁽ⁿ⁾Input mark sample module, mark sample module will be according to Regional Distribution Type generates sample label parametersThen by sample label parameters Pass to model training module, be used for carrying out model training.

Secondly, in the embodiment of the present invention, server, can be according to after receiving one group of user network account Location distribution type generates one group of sample label parameters corresponding to user network account.By above-mentioned side Formula can get the user network account that user provides neatly, and generates sample label parameters, if The data that user provides are abundant, and obtained sample label parameters is the most, so that train Model-fitting degree is the highest, thus the accuracy of lifting scheme.

Alternatively, on the basis of embodiment any one of corresponding first to the 5th of above-mentioned Fig. 3, Fig. 3, In the 6th alternative embodiment of method that the positional information that the embodiment of the present invention provides determines, according to sample mark Note parameter and feature coding parameter determination model training parameter, may include that

According to scale coefficient by the distribution of sample label parameters to different labeled data subclass；

Each labeled data subclass is carried out model training, and obtains the result of model training；

Result according to model training determines model training parameter.

In the present embodiment, server is according to sample label parameters and feature coding parameter determination model training The step of parameter is it may be that sample labeled data is first distributed to the most different marks by server according to coefficient of kurtosis Note data subset closes, particularly as follows:

Model training module in geographical attribute categorizing system training pattern receives the output of sample labeling module Sample label parametersWith feature coding module output normalization to Amount k_f1,k_f2,k_f3,k_f4,...,k_fN, scale coefficient M_fWith coefficient of kurtosis Kurt_f, model training module is permissible According to scale coefficient M_fBy sample label parametersIt is assigned to different In set, these collection are combined into s_i, i=0,1,2,3,4,5..n.The method of salary distribution is as follows:

I i.e. represents each labeled data subclass, by the way labeled data set is divided into multiple mark Note data subset closes, and the objective network account of different scales assigns to each labeled data in different set Set s_iModel training can be carried out independently, and for the objective network account of different " vermicelli " scales Stand-alone training model, can improve the accuracy rate of category of model.Each labeled data subclass is being carried out After model training, can obtain the result of model training, the result finally according to model training determines model Training parameter.Each labeled data subclass is carried out the process of model training by the 7th enforcement below Example is specifically introduced.

Again, in the embodiment of the present invention, server is true according to sample label parameters and feature coding parameter Cover half type training parameter, can be first to distribute sample label parameters to the most different marks according to scale coefficient Data subset closes, and then each labeled data subclass is carried out model training, and obtains model training As a result, the result finally according to model training determines model training parameter.By said method, server Can independently be trained for the objective network account of different " vermicelli " scales, improve model with this and divide The accuracy rate of class.

Alternatively, on the basis of embodiment any one of corresponding first to the 6th of above-mentioned Fig. 3, Fig. 3, In the 7th alternative embodiment of method that the positional information that the embodiment of the present invention provides determines, labeled data Set includes labeled data test set and labeled data training set；

Labeled data test set share in the first labeled data in labeled data subclass is iterated meter Calculate；

Labeled data training set is share in the second labeled data in labeled data subclass is carried out model instruction Practice.

Model training mould in the present embodiment, in the geographical attribute categorizing system training pattern that server includes Block, can carry out model training to each labeled data subclass, first can be drawn by labeled data subclass It is divided into labeled data test set and labeled data training set,

Below to single labeled data subclass s_iThe mode of model training illustrates.

Model training module uses the mode of supervised learning, can use common machine learning algorithm here, Can be decision tree, Bayes, linear discriminent, logistic regression etc..

Wherein, decision tree is a kind of diagram method intuitively using probability analysis, sends out in known various situations On the basis of raw probability, ask for the expected value of the net present value (NPV) probability more than or equal to zero by constituting decision tree, Assessment item risk, it is judged that its feasibility.In machine learning, decision tree is a forecast model, it Represent is a kind of mapping relations between object properties and object value.

Original judgement is modified providing effective means by Bayes for utilizing the information collected.Adopting Before sample, have a judgement to various hypothesis, i.e. prior probability, about the distribution of prior probability, generally Can determine by root micro-judgment, when without any information, generally assume that each prior probability is identical, more complicated essence Really available includes that maximum-entropy technique or the limit method such as distribution density and mutual information principle determine Prior probability distribution.

The basic thought of linear discriminent is that the pattern sample of higher-dimension is projected to best discriminant technique vector space, To reach to extract classification information and the effect of compressive features space dimensionality, after projection, Assured Mode sample is newly Subspace have the between class distance of maximum and minimum inter-object distance, i.e. pattern has optimal within this space Separability.

Logistic regression is a kind of generalized linear regression, has a lot of something in common with multiple linear regression analysis. Their model form is substantially the same.

It should be noted that in actual applications, it is also possible to there is other machine learning algorithm, herein It is not construed as limiting.

It follows that the algorithm that model training module is chosen can use Y_iRepresenting, model training module will mark Data subset closes s_iIt is divided into labeled data test set s⁽⁰⁾ _iWith labeled data training set s⁽¹⁾ _i, will be multiple Labeled data subclass is divided into the first labeled data and the second labeled data, the set structure of the first labeled data Become labeled data test set, and the set of the second labeled data has constituted labeled data training set. Labeled data training set s⁽¹⁾ _iFor the parameter of training pattern, labeled data test set s⁽⁰⁾ _iFor surveying The accuracy of die trial type, model is trained through successive ignition, until the accuracy rate of model converges to certain shape State.

Model training is for different scales s_iTrain different models, the model training of these models ginseng NumberIt is transferred to grader as control signal；Model training module training produces manifold classification device and controls Parameter, different vermicelli size target network accounts are classified by this component class respectively.

Further, in the embodiment of the present invention, labeled data subclass can be divided into labeled data and survey Examination set and labeled data training set, wherein, labeled data test set share in labeled data The first labeled data in set is iterated calculating, and labeled data training set is share in labeled data The second labeled data in set carries out model training.By the way, degree of fitting can be trained more High model, and constantly sophisticated model in the test of labeled data and training, lift scheme dynamic State adaptability.

Alternatively, on the basis of embodiment any one of corresponding first to the 7th of above-mentioned Fig. 3, Fig. 3, In the 8th alternative embodiment of method that the positional information that the embodiment of the present invention provides determines, instruct according to model Practice parameter and feature coding parameter, determine the actual geographic positional information of objective network account, can wrap Include:

According to scale coefficient, by the feature coding parametric distribution of objective network account to sub-classifier, sub Grader is for the geographical position attribute classification according to feature coding parameter output objective network account, geographical Position attribution classification includes the one in concentrated, Topical Dispersion type and overall situation decentralized；

The actual geographic positional information of objective network account is determined according to geographical position attribution classification.

In the present embodiment, model training parameter is used for generating grader, and referring to Fig. 5, Fig. 5 is the present invention Grader structural representation in embodiment, as it can be seen, grader includes block function and many height Grader Y₁To Y_n, sub-classifier generally may refer to machine learning algorithm, and different sub-classifiers is permissible Select different machine learning algorithms, such as decision tree, Bayes, linear discriminent or logistic regression etc..

Block function is according to the scale coefficient in feature coding parameter, by feature coding parametric distribution to corresponding Sub-classifier in, sub-classifier receives the normalized vector in feature coding parameter and coefficient of kurtosis, so The geographical position attribute classification of rear output objective network account.

Geographical position attribute classification includes the one in concentrated, Topical Dispersion type and overall situation decentralized, Wherein, concentrated represents that concern user's integrated distribution of objective network account, in single city, locally divides What scattered type represented objective network account pays close attention to user's integrated distribution in the city in a certain province or geographical position Adjacent group of cities, overall situation decentralized represents the concern user distribution of objective network account in China Multiple cities, these cities are not adjacent on geographical position.

For the objective network account of concentrated, then may further determine that what this objective network account was concentrated City, i.e. uses statistic law to determine.For the objective network account of Topical Dispersion, can be further Calculate the province at its discrete areas place.

Secondly, in the embodiment of the present invention, server according to model training parameter and feature coding parameter, Determine the actual geographic positional information of objective network account, specifically can be first according to scale coefficient, by target The feature coding parametric distribution of network account is in sub-classifier, and sub-classifier is for joining according to feature coding Number output objective network accounts geographical position attribute classifications, geographical position attribute classification include concentrated, One in Topical Dispersion type and overall situation decentralized, determines target finally according to geographical position attribute classification The actual geographic positional information of network account.Use aforesaid way, specifically construct one and can be used for exporting The model of objective network account actual geographic positional information, the grader in training pattern introduces many height Grader, has fully taken into account the multiformity of objective network account scale, trains multiple from data plane Sub-classifier, each sub-classifier is just for the objective network account of some scale, such model energy Enough with training data preferably your matching, the identification ability of raising model.

For ease of understanding, with a concrete application scenarios, positional information a kind of in the present invention is determined below Method is described in detail, particularly as follows:

First company has offered public's account A, has the user of 30 zoness of different to pay close attention to first company at present Public's account A, the situation of concern is as shown in table 2:

Table 2

Now it needs to be determined that the area of public's account A active service, then server uses geographical attribute to divide Feature coding module in class systematic training model calculates scale coefficient, it may be assumed that

$\begin{array}{c}{M}_f=\underset{i=0}{\overset{n}{\Σ}}{C}_{fi}\\ =3807+2141+2095+1865+331+323+268+264+257+256+246+244+241+\\ 241+237+236+232+229+210+194+193+192+191+183+181+179+175\\ =15211\end{array}$

And above-mentioned calculating has met C_fi≥C_f(i+1)Condition.

Calculate the normalized value in each region the most respectively, as a example by Chongqing City, it may be assumed that

$\begin{array}{c}{k}_{fi}=C_{fi}/\underset{i=0}{\overset{n}{\Σ}}{C}_{fi}\\ =3807/15211\\ =0.250279\end{array}$

By that analogy, calculate the normalized value in each area, form according to the normalized value in each area Normalized vector, as shown in table 2.

It follows that calculate the meansigma methods of normalized value in normalized vector, it may be assumed that

$\begin{array}{c}\stackrel{\‾}{k_f}=Sum(k_{f1},k_{f2},k_{f3},k_{f4},\mathrm{...},k_{fN})/N\\ =(0.250279403+0.140753402+0.137729275+0.122608638+0.021760568+\\ 0.021234633+0.017618828+0.017355861+0.016895668+0.016829926+\\ 0.016172507+0.016041023+0.015843797+0.015843797+0.01558083\\ +0.015515088+0.01525212+0.015054894+0.013805798+0.012753928+\\ 0.012688186+0.012622444+0.012556702+0.012030767+0.011899283+0.0117678\\ +0.011504832)/30\\ =0.037037\end{array}$

Finally, the coefficient of kurtosis of objective network account is calculated, it may be assumed that

$\begin{array}{c}{\mathrm{Kurt}}_f=\frac{\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{(k_{fi}-\stackrel{\‾}{k_f})}^4}{{\left(\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{(k_{fi}-\stackrel{\‾}{k_f})}^2\right)}^2}\\ =0.02661293\end{array}$

By the grader in these data input to model, owing to grader will according to scale coefficient 15211 Characteristic is assigned in sub-classifier Y, and sub-classifier Y uses logistic regression algorithm.Sub-classifier Y Receiving normalized vector and coefficient of kurtosis, output public's account A finally geographical position attribution classification is local Decentralized, it is possible to the province obtaining discrete areas place further is Chongqing, Beijing, Shanghai and Tianjin.

Then the active service region of public's account A is defined as Chongqing.

Below the server in the present invention is described in detail, refers to Fig. 6, described server bag Include:

Read module 301, for reading the geographical location information of objective network account to be determined, described geography Positional information is the positional information at the user place paying close attention to described objective network account；

Acquisition module 302, for the described geographical location information read according to described read module 301, obtains Take the feature coding parameter of described objective network account；

Receiver module 303, for receiving the sample label parameters of user network account, and according to described sample The described feature coding parameter determination model training parameter that label parameters and described acquisition module 302 obtain；

Determine module 304, for the described model training parameter that determines according to described receiver module 303 and The described feature coding parameter that described acquisition module 302 obtains, determines the reality of described objective network account Geographical location information, described actual geographic positional information is used for indicating described objective network account current service Geographic range.

In the present embodiment, read module 301 reads the geographical location information of objective network account to be determined, Described geographical location information is the positional information at the user place paying close attention to described objective network account, obtains mould The described geographical location information that block 302 reads according to described read module 301, obtains described objective network The feature coding parameter of account, receiver module 303 receives the sample label parameters of user network account, and The described feature coding parameter determination obtained according to described sample label parameters and described acquisition module 302 Model training parameter, determines the described model training ginseng that module 304 determines according to described receiver module 303 The described feature coding parameter that several and described acquisition module 302 obtains, determines described objective network account Actual geographic positional information, described actual geographic positional information is used for indicating described objective network account to work as The geographic range of front service.

In the embodiment of the present invention, it is provided that a kind of method that positional information determines, server first reads to be treated really Set the goal the geographical location information of network account, then according to geographical location information, obtains objective network account Number feature coding parameter, then server receive user network account sample label parameters, and according to Sample label parameters and feature coding parameter determination model training parameter, finally according to model training parameter And feature coding parameter, determine the actual geographic positional information of objective network account, actual geographic position Information is for indicating the geographic range of objective network account current service.By using aforesaid way to determine net The geographic range of network account current service, it may be considered that be in dynamically change to the user paying close attention to this network account Situation about changing, it is possible to determine actual geographic position according to the geographical position of the user paying close attention to this network account Information, thus be substantially reduced the distortion probability of data, more effectively improves server and determines and obtain network The accuracy of the active service information of account.

Alternatively, on the basis of the embodiment corresponding to above-mentioned Fig. 6, referring to Fig. 7, the present invention implements In another embodiment of the server that example provides,

Described feature coding parameter includes normalized vector, scale coefficient and coefficient of kurtosis；

Described acquisition module 302 includes:

First computing unit 3021, for calculating described scale coefficient according to described geographical location information；

Second computing unit 3022, for calculating described normalized vector according to described geographical location information；

3rd computing unit 3023, for the described normalization calculated according to described second computing unit 3022 Vector calculates described coefficient of kurtosis.

Secondly, in the embodiment of the present invention, the feature coding parameter bag illustrating objective network account is schematically illustrated Include scale coefficient, normalized vector and coefficient of kurtosis, and server is according to geographical location information meter Calculation scale coefficient and normalized vector, go out coefficient of kurtosis further according to normalization neighborhood calculation.Use above-mentioned Mode lifting scheme feasibility in actual applications.

Alternatively, on the basis of the embodiment corresponding to above-mentioned Fig. 7, referring to Fig. 8, the present invention implements In another embodiment of the server that example provides,

Described first computing unit 3021 includes:

First computation subunit 30211, for calculating described scale coefficient as follows:

$M_f=\underset{i=0}{\overset{n}{\Σ}}{C}_{fi}$

Described M_fRepresent scale coefficient described in described objective network account f

Described C_fiRepresent the described objective network account f number in numbered i region, wherein, C_fi≥C_f(i+1)；

For representing that described objective network account f is numbered total people from i=0 to i=n region Number, n represents overall area quantity.

Again, in the embodiment of the present invention, server can use formula to calculate scale coefficient, passes through formula Be calculated rational scale coefficient, with practicality and the feasibility of this lifting scheme.

Alternatively, on the basis of the embodiment corresponding to above-mentioned Fig. 8, referring to Fig. 9, the present invention implements In another embodiment of the server that example provides,

Described second computing unit 3022 includes:

Second computation subunit 30221, for calculating described normalized vector as follows:

$k_{fi}=C_{fi}/\underset{i=0}{\overset{n}{\Σ}}{C}_{fi}$

Described k_fiRepresent the normalization corresponding to the described objective network account f number in numbered i region Value, multiple described normalized values form described normalized vector, described k_fiSpan be more than 0 and Less than 1.

Further, in the embodiment of the present invention, on the basis of obtaining scale coefficient, it is also possible to utilize public affairs Formula is calculated normalized vector, promotes server with this and calculates the feasibility and rationally of normalized vector Property.

Alternatively, on the basis of the embodiment corresponding to above-mentioned Fig. 8, referring to Figure 10, the present invention is real Execute in another embodiment of the server that example provides,

Described 3rd computing unit 3023 includes:

3rd computation subunit 30231, for calculating described coefficient of kurtosis as follows:

$\stackrel{\‾}{k_f}=Sum(k_{f1},k_{f2},k_{f3},k_{f4},\mathrm{...},k_{fN})/N;$

${\mathrm{Kurt}}_f=\frac{\frac{1}{n}\underset{i=1}{\overset{N}{\Σ}}{(k_{fi}-\stackrel{\‾}{k_f})}^4}{{\left(\frac{1}{n}\underset{i=1}{\overset{N}{\Σ}}{(k_{fi}-\stackrel{\‾}{k_f})}^2\right)}^2};$

DescribedRepresent normalized value described in the described normalized vector that described objective network account f is preset Meansigma methods；

Described N represents default parameter, and the span of N is more than 0 and less than or equal to described i；

Described k_fNRepresent the normalized value in normalized vector described in n-th, described (k_f1,k_f2,k_f3,k_f4,...,k_fN) represent one group of sub-normalized vector in described normalized vector, described Sum represents summation operation；

Described Kurt_fRepresenting the described coefficient of kurtosis of described objective network account f, described coefficient of kurtosis is used for Represent the steep slow degree of distribution；

DescribedPay close attention to what the user of described objective network account was distributed quadravalence Square；

DescribedPay close attention to what the user of described objective network account was distributed second order Square.

Further, in the embodiment of the present invention, server be calculated scale coefficient and normalization to On the basis of amount, it is possible to use formula continues to be calculated coefficient of kurtosis, promote server with this and calculate peak The feasibility of degree coefficient and reasonability.

Alternatively, on the basis of the embodiment corresponding to above-mentioned Fig. 6, referring to Figure 11, the present invention is real Execute in another embodiment of the server that example provides,

Described receiver module 303 includes:

Receive unit 3031, for user network account described in reception one group；

Signal generating unit 3032, receives for generating described reception unit 3031 according to location distribution type Described one group described in described sample label parameters corresponding to user network account.

Secondly, in the embodiment of the present invention, server, can be according to after receiving one group of user network account Location distribution type generates one group of sample label parameters corresponding to user network account.By above-mentioned side Formula can get the user network account that user provides neatly, and generates sample label parameters, if The data that user provides are abundant, and obtained sample label parameters is the most, so that train Model-fitting degree is the highest, thus the accuracy of lifting scheme.

Alternatively, on the basis of the embodiment corresponding to above-mentioned Figure 11, referring to Figure 12, the present invention is real Execute in another embodiment of the server that example provides,

Described receiver module 303 includes:

Allocation unit 3033, for distributing described sample label parameters to different according to described scale coefficient Labeled data subclass；

Training unit 3034, for each described labeled data subset to the distribution of described allocation unit 3033 Conjunction carries out model training, and obtains the result of described model training；

First determines unit 3035, for the described model instruction obtained according to the training of described training unit 3034 The result practiced determines described model training parameter.

Again, in the embodiment of the present invention, server is true according to sample label parameters and feature coding parameter Cover half type training parameter, can be first to distribute sample label parameters to the most different marks according to scale coefficient Data subset closes, and then each labeled data subclass is carried out model training, and obtains model training As a result, the result finally according to model training determines model training parameter.By said method, server Can independently be trained for the objective network account of different " vermicelli " scales, improve model with this and divide The accuracy rate of class.

Alternatively, on the basis of the embodiment corresponding to above-mentioned Figure 11, the clothes that the embodiment of the present invention provides In another embodiment of business device,

Described labeled data subclass includes labeled data test set and labeled data training set；

Described labeled data test set share in entering the first labeled data in described labeled data subclass Row iteration calculates；

Described labeled data training set is share in entering the second labeled data in described labeled data subclass Row model training.

Further, in the embodiment of the present invention, labeled data subclass can be divided into labeled data and survey Examination set and labeled data training set, wherein, labeled data test set share in labeled data The first labeled data in set is iterated calculating, and labeled data training set is share in labeled data The second labeled data in set carries out model training.By the way, degree of fitting can be trained more High model, and constantly sophisticated model in the test of labeled data and training, lift scheme dynamic State adaptability.

Alternatively, on the basis of embodiment corresponding any one of above-mentioned Fig. 6 to Figure 10, refer to Figure 13, in another embodiment of the server that the embodiment of the present invention provides,

Described determine that module 304 includes:

Output unit 3041, for according to described scale coefficient, compiling the feature of described objective network account Code parametric distribution is in sub-classifier, and described sub-classifier is for exporting institute according to described feature coding parameter State the geographical position attribute classification of objective network account, described geographical position attribute classification include concentrated, One in Topical Dispersion type and overall situation decentralized；

Second determines unit 3042, for the described geographical position Attribute class exported according to described output unit Do not determine the actual geographic positional information of described objective network account.

Secondly, in the embodiment of the present invention, server according to model training parameter and feature coding parameter, Determine the actual geographic positional information of objective network account, specifically can be first according to scale coefficient, by target The feature coding parametric distribution of network account is in sub-classifier, and sub-classifier is for joining according to feature coding Number output objective network accounts geographical position attribute classifications, geographical position attribute classification include concentrated, One in Topical Dispersion type and overall situation decentralized, determines target finally according to geographical position attribute classification The actual geographic positional information of network account.Use aforesaid way, specifically construct one and can be used for exporting The model of objective network account actual geographic positional information, the grader in training pattern introduces many height Grader, has fully taken into account the multiformity of objective network account scale, trains multiple from data plane Sub-classifier, each sub-classifier is just for the objective network account of some scale, such model energy Enough with training data preferably your matching, the identification ability of raising model.

Figure 14 is a kind of server architecture schematic diagram that the embodiment of the present invention provides, and this server 400 can be because of Configuration or performance are different and produce bigger difference, can include one or more central processing units (English full name: central processing units, english abbreviation: CPU) 422 (such as, one or one Individual above processor) and memorizer 432, one or more storage application program 442 or data 444 Storage medium 430 (such as one or more mass memory units).Wherein, memorizer 432 He Storage medium 430 can be of short duration storage or persistently store.The program being stored in storage medium 430 is permissible Including one or more modules (diagram does not marks), each module can include in server Series of instructions operates.Further, central processing unit 422 could be arranged to lead to storage medium 430 Letter, performs a series of command operatings in storage medium 430 on server 400.

Server 400 can also include one or more power supplys 426, one or more wired or Radio network interface 450, one or more input/output interfaces 458, and/or, one or one with Upper operating system 441, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM etc..

Above-described embodiment can be tied based on the server shown in this Figure 14 by the step performed by server Structure.

Wherein, central processing unit 422 is used for,

Reading the geographical location information of objective network account to be determined, described geographical location information is for paying close attention to institute State the positional information at the user place of objective network account；

According to described geographical location information, obtain the feature coding parameter of described objective network account；

Receive the sample label parameters of user network account, and according to described sample label parameters and described Feature coding parameter determination model training parameter；

According to described model training parameter and described feature coding parameter, determine described objective network account Actual geographic positional information, described actual geographic positional information is used for indicating described objective network account to work as The geographic range of front service.

Wherein, central processing unit 422 specifically for,

Described scale coefficient is calculated according to described geographical location information；

Described normalized vector is calculated according to described geographical location information；

Described coefficient of kurtosis is calculated according to described normalized vector.

Wherein, central processing unit 422 specifically for,

Calculate described scale coefficient as follows:

$M_f=\underset{i=0}{\overset{n}{\Σ}}{C}_{fi}$

Described M_fRepresent scale coefficient described in described objective network account f

Described C_fiRepresent the described objective network account f number in numbered i region；

For representing that described objective network account f is numbered total people from i=0 to i=n region Number, n represents overall area quantity.

Wherein, central processing unit 422 specifically for,

Calculate described normalized vector as follows:

$k_{fi}=C_{fi}/\underset{i=0}{\overset{n}{\Σ}}{C}_{fi}$

Described k_fiRepresent the normalization corresponding to the described objective network account f number in numbered i region Value, described k_fiSpan be more than 0 and less than 1.

Wherein, central processing unit 422 specifically for,

Calculate described coefficient of kurtosis as follows:

$\stackrel{\‾}{k_f}=(k_{f1},k_{f2},k_{f3},k_{f4},\mathrm{...},k_{fN})/N;$

${\mathrm{Kurt}}_f=\frac{\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{(k_{fi}-\stackrel{\‾}{k_f})}^4}{{\left(\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{(f_{fi}-\stackrel{\‾}{k_f})}^2\right)}^2};$

DescribedRepresent the normalized vector meansigma methods that described objective network account f is preset；

Described N represents default parameter, and the span of N is more than 0 and less than or equal to described i；

Described k_fNRepresent the normalized value in normalized vector described in n-th, described (k_f1,k_f2,k_f3,k_f4,...,k_fN) represent one group of sub-normalized vector in described normalized vector；

Described Kurt_fRepresenting the described coefficient of kurtosis of described objective network account f, described coefficient of kurtosis is used for Represent the steep slow degree of distribution；

DescribedPay close attention to what the user of described objective network account was distributed quadravalence Square；

DescribedPay close attention to what the user of described objective network account was distributed second order Square.

Wherein, central processing unit 422 specifically for,

Receive user network account described in a group；

According to the described sample that user network account described in described one group of location distribution type generation is corresponding Label parameters.

Wherein, central processing unit 422 specifically for,

According to described coefficient of kurtosis by the distribution of described sample label parameters to different labeled data subclass；

Each described labeled data subclass is carried out model training, and obtains the result of described model training；

Result according to described model training determines described model training parameter.

Wherein, central processing unit 422 specifically for,

According to described scale coefficient, by the feature coding parametric distribution of described objective network account to subclassification In device, described sub-classifier for exporting the ground of described objective network account according to described feature coding parameter Reason position attribution classification, described geographical position attribute classification includes concentrated, Topical Dispersion type and the overall situation One in decentralized；

The actual geographic position letter of described objective network account is determined according to described geographical position attribute classification Breath.

Those skilled in the art is it can be understood that arrive, and for convenience and simplicity of description, above-mentioned retouches The specific works process of the system stated, device and unit, is referred to the correspondence in preceding method embodiment Process, does not repeats them here.

In several embodiments provided herein, it should be understood that disclosed system, device and Method, can realize by another way.Such as, device embodiment described above is only shown Meaning property, such as, the division of described unit, be only a kind of logic function and divide, actual can when realizing There to be other dividing mode, the most multiple unit or assembly can in conjunction with or be desirably integrated into another System, or some features can ignore, or do not perform.Another point, shown or discussed each other Coupling direct-coupling or communication connection can be the INDIRECT COUPLING by some interfaces, device or unit Or communication connection, can be electrical, machinery or other form.

The described unit illustrated as separating component can be or may not be physically separate, makees The parts shown for unit can be or may not be physical location, i.e. may be located at a place, Or can also be distributed on multiple NE.Can select according to the actual needs part therein or The whole unit of person realizes the purpose of the present embodiment scheme.

It addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit In, it is also possible to it is that unit is individually physically present, it is also possible to two or more unit are integrated in one In individual unit.Above-mentioned integrated unit both can realize to use the form of hardware, it would however also be possible to employ software merit The form of energy unit realizes.

If described integrated unit realizes and as independent production marketing using the form of SFU software functional unit Or when using, can be stored in a computer read/write memory medium.Based on such understanding, this The part that the most in other words prior art contributed of technical scheme of invention or this technical scheme Completely or partially can embody with the form of software product, this computer software product is stored in one In storage medium, including some instructions with so that computer equipment (can be personal computer, Server, or the network equipment etc.) perform completely or partially walking of method described in each embodiment of the present invention Suddenly.And aforesaid storage medium includes: USB flash disk, portable hard drive, read only memory (English full name: Read-Only Memory, english abbreviation: ROM), random access memory (English full name: Random Access Memory, english abbreviation: RAM), magnetic disc or CD etc. are various can store The medium of program code.

The above, above example only in order to technical scheme to be described, is not intended to limit； Although being described in detail the present invention with reference to previous embodiment, those of ordinary skill in the art should Work as understanding: the technical scheme described in foregoing embodiments still can be modified by it, or to it Middle part technical characteristic carries out equivalent；And these amendments or replacement, do not make appropriate technical solution Essence depart from various embodiments of the present invention technical scheme spirit and scope.

Claims (19)

1. the method that a positional information determines, it is characterised in that including:

Reading the geographical location information of objective network account to be determined, described geographical location information is for paying close attention to institute State the positional information at the user place of objective network account；

According to described geographical location information, obtain the feature coding parameter of described objective network account；

Receive the sample label parameters of user network account, and according to described sample label parameters and described Feature coding parameter determination model training parameter；

According to described model training parameter and described feature coding parameter, determine described objective network account Actual geographic positional information, described actual geographic positional information is used for indicating described objective network account to work as The geographic range of front service.

Method the most according to claim 1, it is characterised in that described feature coding parameter includes returning One changes vector, scale coefficient and coefficient of kurtosis；

Described obtain the feature coding parameter of described objective network account according to described geographical location information, Including:

Described scale coefficient is calculated according to described geographical location information；

Described normalized vector is calculated according to described geographical location information；

Described coefficient of kurtosis is calculated according to described normalized vector.

Method the most according to claim 2, it is characterised in that described believe according to described geographical position Breath calculates described scale coefficient, including:

Calculate described scale coefficient as follows:

$M_f=\underset{i=0}{\overset{n}{\Σ}}{C}_{fi}$

Described M_fRepresent scale coefficient described in described objective network account f

Described C_fiRepresent the described objective network account f number in numbered i region, wherein, C_fi≥C_f(i+1)；

For representing that described objective network account f is numbered total people from i=0 to i=n region Number, n represents overall area quantity.

Method the most according to claim 3, it is characterised in that described believe according to described geographical position Breath calculates described normalized vector, including:

Calculate described normalized vector as follows:

$k_{fi}=C_{fi}/\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{C}_{fi}$

Described k_fiRepresent the normalization corresponding to the described objective network account f number in numbered i region Value, multiple described normalized values form described normalized vector, described k_fiSpan be more than 0 and Less than 1.

Method the most according to claim 4, it is characterised in that described according to described normalized vector Calculate described coefficient of kurtosis, including:

Calculate described coefficient of kurtosis as follows:

$\stackrel{\‾}{k_f}=Sum(k_{f1},k_{f2},k_{f3},k_{f4},...,k_{fN})/N;$

${\mathrm{Kurt}}_f=\frac{\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{(k_{fi}-\stackrel{\‾}{k_f})}^4}{{\left(\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{\left(k_{fi}-\stackrel{\‾}{k_f}\right)}^2\right)}^2};$

DescribedRepresent normalized value described in the described normalized vector that described objective network account f is preset Meansigma methods；

Described N represents default parameter, and the span of N is more than 0 and less than or equal to described i；

Described k_fNRepresent the normalized value in normalized vector described in n-th, described (k_f1,k_f2,k_f3,k_f4,...,k_fN) represent one group of sub-normalized vector in described normalized vector, described Sum represents summation operation；

Described Kurt_fRepresenting the described coefficient of kurtosis of described objective network account f, described coefficient of kurtosis is used for Represent the steep slow degree of distribution；

DescribedPay close attention to what the user of described objective network account was distributed quadravalence Square；

DescribedPay close attention to what the user of described objective network account was distributed two Rank square.

Method the most according to claim 1, it is characterised in that described reception user network account Sample label parameters, including:

Receive user network account described in a group；

According to the described sample that user network account described in described one group of location distribution type generation is corresponding Label parameters.

Method the most according to claim 6, it is characterised in that described according to described sample mark ginseng Several and described feature coding parameter determination model training parameter, including:

According to described scale coefficient by the distribution of described sample label parameters to different labeled data subclass；

Each described labeled data subclass is carried out model training, and obtains the result of described model training；

Result according to described model training determines described model training parameter.

Method the most according to claim 7, it is characterised in that described labeled data subclass includes Labeled data test set and labeled data training set；

Described labeled data test set share in entering the first labeled data in described labeled data subclass Row iteration calculates；

Described labeled data training set is share in entering the second labeled data in described labeled data subclass Row model training.

Method the most according to any one of claim 1 to 5, it is characterised in that described according to institute State model training parameter and described feature coding parameter, determine the actual geographic of described objective network account Positional information, including:

According to described scale coefficient, by the feature coding parametric distribution of described objective network account to subclassification In device, described sub-classifier for exporting the ground of described objective network account according to described feature coding parameter Reason position attribution classification, described geographical position attribute classification includes concentrated, Topical Dispersion type and the overall situation One in decentralized；

The actual geographic position letter of described objective network account is determined according to described geographical position attribute classification Breath.

10. a server, it is characterised in that including:

Read module, for reading the geographical location information of objective network account to be determined, described geographical position Confidence breath is the positional information at the user place paying close attention to described objective network account；

Acquisition module, for the described geographical location information read according to described read module, obtains described The feature coding parameter of objective network account；

Receiver module, for receiving the sample label parameters of user network account, and according to described sample mark The described feature coding parameter determination model training parameter that note parameter and described acquisition module obtain；

Determine module, for the described model training parameter that determines according to described receiver module and described in obtain The described feature coding parameter that delivery block obtains, determines the actual geographic position letter of described objective network account Breath, described actual geographic positional information is for indicating the geographic range of described objective network account current service.

11. servers according to claim 10, it is characterised in that described feature coding parameter bag Include normalized vector, scale coefficient and coefficient of kurtosis；

Described acquisition module includes:

First computing unit, for calculating described scale coefficient according to described geographical location information；

Second computing unit, for calculating described normalized vector according to described geographical location information；

3rd computing unit, calculates for the described normalized vector calculated according to described second computing unit Described coefficient of kurtosis.

12. servers according to claim 11, it is characterised in that described first computing unit bag Include:

First computation subunit, for calculating described scale coefficient as follows:

$M_f=\underset{i=0}{\overset{n}{\Σ}}{C}_{fi}$

Described M_fRepresent scale coefficient described in described objective network account f

Described C_fiRepresent the described objective network account f number in numbered i region, wherein, C_fi≥C_f(i+1)；

For representing that described objective network account f is numbered total people from i=0 to i=n region Number, n represents overall area quantity.

13. servers according to claim 12, it is characterised in that described second computing unit bag Include:

Second computation subunit, for calculating described normalized vector as follows:

$k_{fi}=C_{fi}/\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{C}_{fi}$

Described k_fiRepresent the normalization corresponding to the described objective network account f number in numbered i region Value, multiple described normalized values form described normalized vector, described k_fiSpan be more than 0 and Less than 1.

14. servers according to claim 13, it is characterised in that described 3rd computing unit bag Include:

3rd computation subunit, for calculating described coefficient of kurtosis as follows:

$\stackrel{\‾}{k_f}=Sum(k_{f1},k_{f2},k_{f3},k_{f4},...,k_{fN})/N;$

${\mathrm{Kurt}}_f=\frac{\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{(k_{fi}-\stackrel{\‾}{k_f})}^4}{{\left(\frac{1}{n}\underset{i=1}{\overset{n}{\Σ}}{\left(k_{fi}-\stackrel{\‾}{k_f}\right)}^2\right)}^2};$

DescribedRepresent normalized value described in the described normalized vector that described objective network account f is preset Meansigma methods；

Described N represents default parameter, and the span of N is more than 0 and less than or equal to described i；

Described k_fNRepresent the normalized value in normalized vector described in n-th, described (k_f1,k_f2,k_f3,k_f4,...,k_fN) represent one group of sub-normalized vector in described normalized vector, described Sum represents summation operation；

Described Kurt_fRepresenting the described coefficient of kurtosis of described objective network account f, described coefficient of kurtosis is used for Represent the steep slow degree of distribution；

DescribedPay close attention to what the user of described objective network account was distributed quadravalence Square；

DescribedPay close attention to what the user of described objective network account was distributed two Rank square.

15. servers according to claim 10, it is characterised in that described receiver module includes:

Receive unit, for user network account described in reception one group；

Signal generating unit, for generating described the one of described reception unit reception according to location distribution type Organize the described sample label parameters that described user network account is corresponding.

16. servers according to claim 15, it is characterised in that described receiver module includes:

Allocation unit, for distributing described sample label parameters to the most different marks according to described scale coefficient Note data subset closes；

Training unit, carries out mould for each described labeled data subclass distributing described allocation unit Type training, and obtain the result of described model training；

First determines unit, the result of the described model training for obtaining according to the training of described training unit Determine described model training parameter.

17. servers according to claim 16, it is characterised in that described labeled data subclass Including labeled data test set and labeled data training set；

Described labeled data test set share in entering the first labeled data in described labeled data subclass Row iteration calculates；

Described labeled data training set is share in entering the second labeled data in described labeled data subclass Row model training.

18. according to the server according to any one of claim 10 to 14, it is characterised in that described Determine that module includes:

Output unit, for according to described scale coefficient, joining the feature coding of described objective network account Number distributes to sub-classifier, and described sub-classifier is for exporting described mesh according to described feature coding parameter The geographical position attribute classification of mark network account, described geographical position attribute classification includes concentrated, locally One in decentralized and overall situation decentralized；

Second determines unit, true for the described geographical position attribute classification exported according to described output unit The actual geographic positional information of fixed described objective network account.

19. 1 kinds of servers, it is characterised in that including: memorizer, transceiver, processor and total Wire system；

Wherein, described memorizer is used for storing program；

Described processor is used for the program performing in described memorizer, step specific as follows:

Reading the geographical location information of objective network account to be determined, described geographical location information is for paying close attention to institute State the positional information at the user place of objective network account；

According to described geographical location information, obtain the feature coding parameter of described objective network account；

Receive the sample label parameters of user network account, and according to described sample label parameters and described Feature coding parameter determination model training parameter；

According to described model training parameter and described feature coding parameter, determine described objective network account Actual geographic positional information, described actual geographic positional information is used for indicating described objective network account to work as The geographic range of front service.