CN110008973A - A kind of model training method, the method and device that target user is determined based on model - Google Patents
A kind of model training method, the method and device that target user is determined based on model Download PDFInfo
- Publication number
- CN110008973A CN110008973A CN201811403695.2A CN201811403695A CN110008973A CN 110008973 A CN110008973 A CN 110008973A CN 201811403695 A CN201811403695 A CN 201811403695A CN 110008973 A CN110008973 A CN 110008973A
- Authority
- CN
- China
- Prior art keywords
- user
- matching degree
- scene
- value
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Data Mining & Analysis (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Game Theory and Decision Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Marketing (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Business, Economics & Management (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
It discloses a kind of model training method, determine the method and device of target user based on model.The training method of the matching degree prediction model of a kind of user and target scene, this method comprises: obtaining user's training sample set；Determine initial matching degree prediction model, processing is iterated using following steps, until the accuracy rate of the predicted value of model output reaches preset requirement: concentrating the characteristic value of any sample to input current matching degree prediction model user's training sample, the matching degree predicted value exported；Use the accuracy rate of the label value verification predicted value of institute's input sample；Judge whether the accuracy rate reaches preset requirement；If not up to, optimizing processing to current matching degree prediction model according to the accuracy rate；After iteration, current matching degree prediction model is determined as to be used for target user's prediction model of target scene.
Description
Technical field
This specification embodiment is related to technical field of internet application more particularly to a kind of model training method, based on mould
Type determines the method and device of target user.
Background technique
Businessman is when to user's recommendation information (such as advertisement, preferential), if screened in advance to user group, determines
The interested target user of information may be recommended, the rate of return on investment of businessman can be improved, and avoid user's quilt
Recommend uninterested information, promotes user experience.
In the prior art, it can be screened based on the various features of user, for example, it is assumed that target scene is businessman's needs
Recommend certain promotional card to user group, and it is desirable that user gets, using the promotional card, then it can be after small-scale recommend, really
Surely several users (can be described as seed user) of the promotional card are actually used, it is believed that of this certain customers and target scene
It is higher with spending, various features value of the seed user in target scene is calculated, if the one or more characteristic values of certain of seed user
Be significantly higher than non-seed user, then can will one or more higher users of characteristic value in non-recommended user, being determined as can
With the target user recommended.
In this scheme, the value of the various features of user is only individually calculated, and obtained characteristic value is only individually or simple
For screening after ground combination, thus target user can not be determined based on the relevance between each feature, the selection result is more rough.
Summary of the invention
In view of the above technical problems, this specification embodiment provides a kind of model training method, determines target based on model
The method and device of user, technical solution are as follows:
According to this specification embodiment in a first aspect, providing the matching degree prediction model of a kind of user and target scene
Training method, this method comprises:
Obtain user's training sample set；User's training sample is concentrated using seed user as positive sample and with non-kind
For child user as negative sample, the characteristic value of any user training sample includes several identity characteristics and several behavioural characteristics；
It determines initial matching degree prediction model, is iterated processing using following steps, until the prediction of model output
The accuracy rate of value reaches preset requirement:
It concentrates the characteristic value of any sample to input current matching degree prediction model user's training sample, is exported
Matching degree predicted value；It wherein, include: the user vector operation rule current according to model according to the method that characteristic value obtains predicted value
Then, characteristic value is carried out that user vector is calculated；Obtain the scene vector of several feature calculations previously according to target scene；
Default vector operation is carried out to user vector and scene vector and obtains predicted value；
Use the accuracy rate of the label value verification predicted value of institute's input sample；
Judge whether the accuracy rate reaches preset requirement；If not up to, according to the accuracy rate to current matching
Degree prediction model optimizes processing；The optimization processing includes at least: current user vector operation rule in Optimized model；
After iteration, the target user that current matching degree prediction model is determined as being used for target scene is predicted into mould
Type.
According to the second aspect of this specification embodiment, a kind of prediction target based on the matching degree prediction model is provided
The method of user, this method comprises:
The value of the preset several identity characteristics of alternative user and the value of several behavioural characteristics are extracted, the spy of alternative user is obtained
Value indicative；
Extracted characteristic value is inputted into the matching degree prediction model, obtains the matching degree predicted value of model output；
According to the matching degree predicted value, judge whether the alternative user meets preset condition；If so, described in determining
Alternative user is the target user of the target scene.
According to the third aspect of this specification embodiment, the matching degree prediction model of a kind of user and target scene is provided
Training device, the device include:
Sample obtains module, for obtaining user's training sample set；User's training sample concentration is made with seed user
For positive sample and using non-seed user as negative sample, the characteristic value of any user training sample include several identity characteristics with
Several behavioural characteristics；
Model determining module, for determining initial matching degree prediction model；
Optimization processing module, for being iterated processing using following submodule, until the standard of the predicted value of model output
True rate reaches preset requirement:
Predicted value obtains submodule, for concentrating user's training sample the characteristic value of any sample to input current matching
Spend prediction model, the matching degree predicted value exported；It wherein, include: according to mould according to the device that characteristic value obtains predicted value
The current user vector operation rule of type, carries out characteristic value user vector is calculated；It obtains previously according to target scene
The scene vector of several feature calculations；Default vector operation is carried out to user vector and scene vector and obtains predicted value；
Accuracy rate judging submodule, the accuracy rate of the label value verification predicted value of input sample for using；
Optimization processing submodule, for judging whether the accuracy rate reaches preset requirement；If not up to, according to
Accuracy rate optimizes processing to current matching degree prediction model；The optimization processing includes at least: current in Optimized model
User vector operation rule；
Output module, for being determined as current matching degree prediction model to be used for the mesh of target scene after iteration
Mark user in predicting model.
According to the fourth aspect of this specification embodiment, provide a kind of based on the matching of any one of claim 9 to 12
The device of the prediction target user of prediction model is spent, which includes:
Characteristics extraction module, for extracting the value and several behavioural characteristics of the preset several identity characteristics of alternative user
Value, obtains the characteristic value of alternative user；
Predicted value obtains module, and for extracted characteristic value to be inputted the matching degree prediction model, it is defeated to obtain model
Matching degree predicted value out；
Target user's determining module, for it is pre- to judge whether the alternative user meets according to the matching degree predicted value
If condition；If so, determining that the alternative user is the target user of the target scene.
Technical solution provided by this specification embodiment, first training are for determining that prediction user matches with target scene
The model of degree, using seed user as positive sample when training pattern, using non-seed user as negative sample, and in training pattern and base
During model is predicted, the multiple identity characteristics and behavioural characteristic of user's sample are extracted, thus comprehensive alternative user
Various features predict whether it is target user.
It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not
This specification embodiment can be limited.
In addition, any embodiment in this specification embodiment does not need to reach above-mentioned whole effects.
Detailed description of the invention
In order to illustrate more clearly of this specification embodiment or technical solution in the prior art, below will to embodiment or
Attached drawing needed to be used in the description of the prior art is briefly described, it should be apparent that, the accompanying drawings in the following description is only
The some embodiments recorded in this specification embodiment for those of ordinary skill in the art can also be attached according to these
Figure obtains other attached drawings.
Fig. 1 is the flow diagram of the matching degree prediction model construction method of this specification embodiment；
Fig. 2 is the process of target user's prediction technique based on constructed matching degree prediction model of this specification embodiment
Schematic diagram；
Fig. 3 is the flow diagram that user vector is calculated based on User DN N of this specification embodiment；
Fig. 4 is the flow diagram that scene vector is calculated based on scene DNN of this specification embodiment；
Fig. 5 is the flow diagram of the method based on DNN building matching degree prediction model of this specification embodiment；
Fig. 6 is the structural schematic diagram of the matching degree prediction model construction device of this specification embodiment；
Fig. 7 is the structure of target user's prediction meanss based on constructed matching degree prediction model of this specification embodiment
Schematic diagram；
Fig. 8 is the structural schematic diagram for configuring a kind of equipment of this specification embodiment device.
Specific embodiment
In order to make those skilled in the art more fully understand the technical solution in this specification embodiment, below in conjunction with this
Attached drawing in specification embodiment is described in detail the technical solution in this specification embodiment, it is clear that described
Embodiment is only a part of the embodiment of this specification, instead of all the embodiments.The embodiment of base in this manual,
Those of ordinary skill in the art's every other embodiment obtained, all should belong to the range of protection.
Businessman is when recommending marketing message to user, it may be predetermined that the targeted user population of the marketing scene, user
Group is more accurately determining, and the success rate of marketing may be higher.For recommending insurance products to user, for different insurances
Product, operation personnel can be determined according to the different characteristics of different product, respectively each product corresponding targeted user population into
Row is recommended, to promote the click and conversion in marketing process, excavates potential user with higher efficiency, improves marketing behavior
Rate of return on investment.Therefore, before being marketed, it is necessary first to determine targeted user population.
In a kind of scheme of the prior art, laid in user group is directed to by operation personnel in advance, according to user data
Add one or more labels for each user, when needing to determine target user for certain marketing scene, operation personnel further according to
The feature of the scene determines one or more target labels, so that the user with target labels is determined as the marketing scene
Target user.
When obviously, based on the labeled targets user manually added, the accuracy of target user largely depends on fortune
Battalion be user addition label and be scene determination target labels accuracy.Also, certain characteristics of user and scene,
Possibly the accuracy of target user in this programme can not be more made to deviate by label accurate description, positioning.
In another scheme of the prior art, in the case where having carried out a degree of marketing, obtains and realize marketing
The information of the seed user of purpose, statistics seed user are significantly higher than the one or more of non-seed user in the marketing scene
Characteristic value (can become conspicuousness label), so as to will equally have conspicuousness label in non-recommended user, i.e., this
Or the equally higher user of multinomial characteristic value, it is determined as the target user that can be recommended.
For example, the TGI value (Target of the various features of seed user and non-seed user in marketing scene can be calculated
Gro up Index, target group's index), so that it is determined that seed user TGI value is significantly higher than one or more of non-seed user
Item feature, then calculates the TGI value of the individual features of non-recommended user, and the calculated higher user of TGI value is determined as mesh
Mark user.
But in this scheme, the value of various features is individually to calculate, and used using conspicuousness label filtration target
When family, each feature is merely able to use individually or after simple combination, thus can not be based on the inner link between each feature, more
Accurately excavate target user.
In view of the above-mentioned problems, this specification embodiment provides a kind of target user's prediction technique, the program includes two ranks
Section: the building stage of the matching degree prediction model of user and target scene, and alternative user is predicted based on matching degree prediction model
With the stage of target scene matching degree.
The building stage of the matching degree prediction model of user and target scene is introduced first, and shown in Figure 1, this method can
With the following steps are included:
S101 obtains user's training sample set；User's training sample concentrate using seed user as positive sample and with
For non-seed user as negative sample, the characteristic value of any user training sample includes several identity characteristics and several behavioural characteristics；
In the scheme that this specification embodiment provides, the marketing of certain scale can be first carried out.For example, with certain promotional card
For, according to aforementioned 2 kinds of prior arts or other schemes, determines a batch target user, push the advertisement of the promotional card
Information or the electronic card packet etc. that promotional card is directly issued to target user, the certain customers in this crowd of target user may
Using the promotional card, this certain customers is the seed user of target scene.
Obviously, the matching degree of seed user and target scene is higher, and remaining users and the matching degree of target scene are lower,
Therefore, the model construction stage of this programme is to carry out the matching degree of user and target scene using seed user as positive sample
Prediction model training.Specifically, the data of seed user, can be by the client that markets, according to the user accumulated
Data provide；It can also be provided by the platform marketed according to the user data in marketing result；This specification does not limit
It is fixed.
In addition, in the model construction stage of this programme, be progress model training using non-seed user as negative sample, here
Non-seed user may include all types of user except seed user.For example, for the user group for having carried out marketing, seed
Remaining users except user can be considered as non-seed user, and can carry out this some residual user as negative sample
Model training；In another example in the marketing of the above-mentioned certain scale first carried out, the non-targeted user being determined as except target user,
Non-seed user can also be considered as；It is all with per family can be in addition to seed user for another example in the user group laid in
It is considered as non-seed user；Etc., it individually or can combine the example above or other all types of user, obtain in this programme
To several negative samples for training, without limitation, those skilled in the art can be according to the actual situation for this specification embodiment
It is selected.
The seed user concentrated for user's training sample and non-seed user can divide from identity and the dimension of behavior
Several features are indescribably taken to obtain the label value of sample.Identity characteristic may include the gender of user, age (section), health status etc.
The feature for characterizing natural quality also may include the feature that educational background, city, occupation, hobby etc. are used to characterize social property,
Etc..Behavioural characteristic may include the historical behaviors such as the search of user, click, such as search for certain commodity, activity, click advertisement,
Preferential activity etc.；It also may include purchase, the payment etc. on user's line and under line.
S102 determines initial matching degree prediction model；
It is understood that matching degree prediction model initial determined by here, can be initialization before training for the first time
Obtained model is also possible to the model obtained after the completion of last time training.
For example, client after market for the first time and obtain batch of seeds user, can be determined just by this programme
The model that beginningization obtains, and subsequent training is carried out based on first seed user；The matching degree prediction model obtained after training, can
To market for second, and second batch seed user is obtained, then the model after can determining training for the first time is initial
Continue training pattern with degree prediction model, and based on second batch seed user；The model obtained after second of training, can be used for
Third time is marketed, and can also carry out third time training, and the model after second is determined as initial matching degree and is predicted
Model even, can also be in third time training, by the model after first time training if third time marketing is ineffective
It is determined as initial matching degree prediction model.Therefore, this specification does not limit identified initial matching when secondary trained
Spend prediction model.
User's training sample is concentrated the characteristic value of any sample to input current matching degree prediction model, obtained by S103
The matching degree predicted value of output；It wherein, include: the user vector current according to model according to the method that characteristic value obtains predicted value
Operation rule carries out characteristic value user vector is calculated；Obtain the field of several feature calculations previously according to target scene
Scape vector；Default vector operation is carried out to user vector and scene vector and obtains predicted value；
It is to be calculated separately for user with scene to the measurement mode of user and scene matching degree in this specification scheme
For the vector of characterization, then by the operation result of user vector and scene vector, as the predicted value to matching degree.
In the user vector, the characteristic in terms of M of the M dimension to describe user, but the characteristic in terms of this M can use
The meaning in reality must not necessarily be corresponded to.Similarly, in scene vector, in terms of can use N of N number of dimension to describe user
Characteristic, but the characteristic in terms of this N must not necessarily correspond to the meaning in reality, and M and the numerical value of N and whether it is equal can be by ability
Field technique personnel determine according to demand.
The detailed process for calculating user vector, can be first according to the characteristic value of user's training sample of input model, base
It is calculated in the current user vector operation rule of model, obtains user vector.For example, user vector operation rule can be
One DNN (Deep Neural Networks, deep neural network), the characteristic value of input can then export after the DNN
User vector.
In addition, for scene vector target can also be extracted first according to several feature calculations of target scene, specifically
Several default features of scene, obtain the characteristic value of target scene, preset scene vector operation rule are then obtained, according to institute
Scene vector operation rule is stated, the characteristic value of target scene is calculated, scene vector is obtained.
It is understood that root can be used in advance since target scene and its feature are generally changeless
Scene vector is calculated based on scene vector operation rule according to the feature of target scene.Scene vector operation rule can be
Independent rule can also be the current scene vector operation of matching degree prediction model in the same manner as user vector operation rule
Rule.
Also, scene vector operation rule can be fixed and invariable, to obtain the same scene vector for subsequent
Operation is also possible to user vector operation rule likewise it is possible to processing be optimized during model training, to make
It is more accurate based on rule scene vector calculated.For example, can be also a DNN, the characteristic value of target scene is defeated
Scene vector can be exported after entering the DNN.
After user vector and scene vector is calculated, default vector operation carried out to the two, in this specification embodiment
Specific vector mode is not limited.
For example, dot product can be carried out to the two, to obtain the scalar value between [0,1].Also, the scalar that dot product obtains
Value is closer to 1, and can indicating user vector, the angle in vector space is smaller with scene vector, i.e., closer to parallel, Ke Yili
Solution is higher for the matching degree of user and scene；And the scalar value that dot product obtains closer to 0 when, can indicate user vector and scene
Vector angle in vector space is bigger, i.e., closer to vertical, it can be understood as the matching degree of user and scene is lower.
It is worth noting that, when carrying out different vector operations from scene vector to user vector, to the attribute of the two
There are different requirements.For example, then needing the corresponding ranks of the two if necessary to carry out dot product to user vector and scene vector
Number is identical.
In addition, carried out during default vector operation obtains predicted value to user vector and scene vector
Default vector operation, can also be the combination of multiple operations not just for a certain operation.For example, to 2 vectors into
After row dot product, discretization can also be carried out again, and the successive value that dot product is obtained is converted into corresponding discrete value, in order to subsequent
It is compared with label value.
S104 uses the accuracy rate of the label value verification predicted value of institute's input sample；
S105, judges whether the accuracy rate reaches preset requirement；If not up to, according to the accuracy rate to current
Matching degree prediction model optimizes processing, and returns to S103；The optimization processing includes at least: current use in Optimized model
Family vector operation rule；S106 is continued to execute if reaching；
For ease of description, S104 is illustrated in conjunction with S105.
It can be accomplished in several ways using the accuracy rate of label value verification predicted value.
For example, can make the label value 1 of positive sample, the label value of negative sample is 0, and to the user vector of seed user
Can be according to the numerical value and 1 difference after the dot product both calculated obtains 1 numerical value with the scene vector of target scene, determination
The accuracy rate of this output predicted value, similarly the difference according to non-seed user with 0 determines accuracy rate；In another example can be with
Indicate positive sample using some value, such as A, 10 indicate negative sample, such as C, 11 using some value, and by user vector with
The operation result of scene vector is converted into predicted value relevant to the value of positive negative sample is indicated by preset corresponding relationship, from
And by comparing the gap of predicted value and label value, obtain the accuracy rate of the two；Etc..
S106, after iteration, the target user for being determined as being used for target scene for current matching degree prediction model is pre-
Survey model.
Referring to shown in Fig. 2, based on the method for the matching degree prediction model prediction target user constructed above, may include with
Lower step:
S201 extracts the value of the preset several identity characteristics of alternative user and the value of several behavioural characteristics, is alternatively used
The characteristic value at family；
Specifically, alternative user collection can be obtained first, include several alternative users in the set, then for described standby
Any alternative user for selecting family to concentrate: the identity data and behavioral data of the alternative user are obtained；According to the identity data
The value of the preset several identity characteristics of the alternative user is extracted, and if to extract the alternative user according to the behavioral data preset
The value of dry behavioural characteristic, obtains the characteristic value of alternative user.
The alternative user that obtained alternative user is concentrated can be all or most of user laid in, but directly will
A large amount of user's input model is predicted, the efficiency of entire prediction process may be reduced, accordingly it is also possible to based on being laid in
User, after carrying out preliminary rough but efficient screening, alternative user collection is formed by user qualified after screening, then will
It in alternative user input model, is more accurately predicted, to precisely and efficiently predict target user.
Specifically, initial user collection can be obtained first, include several unscreened initial users in the set, then
The every screening conditions concentrated using the screening conditions, each initial user concentrated to the initial user are screened,
Obtain the alternative user collection being made of the user for meeting every screening conditions.
Screening conditions concentration may include one or more screening conditions, and every screening conditions can be simultaneously or sequentially
It is applied in combination.
For example, may include label filtration condition, for whether there are target labels to screen as condition to user.
As already mentioned it is possible to be in advance that laid in user adds one or more labels by operation personnel, such as women, 20 generations, north
Capital (residence), programmer, etc., then label filtration condition whether can have with user or not have some or it is more
A target labels judge whether user can be alternative user as condition.
In another example may include behavior screening conditions, for whether there is goal behavior to sieve as condition to user
Choosing.By this condition, whether there can be or not have goal behavior according to user, judge that user whether may be used as condition
Think alternative user, if whether user search for, clicked relevant advertisement, favor information, whether user bought, it is similar to pay
Commodity, service, even user's whether in the near future logged platform, etc. for facing of marketing.
For another example may include seed screening conditions, for whether to be that seed user as condition screens user.
As previously described, same marketing scene, may will do it multiple promotion and model training, to generate and use more batches of kinds
Child user.The seed user generated in marketing before, due to its label, behavior and with the matching degree of target scene etc., it is clear that
It is satisfactory, is predicted so if directly inputting model, it will be targeted user.And in order to save marketing
Cost, and user experience is improved, the user of seed user will can be had become before, from the target user's mistake marketed again
It filters.
Extracted characteristic value is inputted the matching degree prediction model by S202, obtains the matching degree prediction of model output
Value；
It is understood that obtain the process of predicted value according to user characteristics value in this step, be also related to user vector with
The operation of scene vector, but due to target scene and it is characterized in that being basically unchanged, can directly use model training
Last time iteration when used scene vector or recalculate prediction after a scene vector in multiple alternative users
In the process using or every time prediction alternative user when recalculate, etc., this specification embodiment does not limit this, and
And used mode should not be used as the restriction based on model prediction when institute usage mode when model construction.
S203 judges whether the alternative user meets preset condition according to the matching degree predicted value；If so, really
The fixed alternative user is the target user of the target scene.
The matching degree predicted value of model output, can be discrete, for example 2 or multiple discrete values, if wherein 1
A or multiple values are as eligible, can be determined as target user；It is also possible to continuously, and is arranged according to predicted value
Sequence, determines within the scope of certain numerical value according to ranking results or a certain number of users are target user；This specification embodiment is not
It limits.
After model prediction matching degree and determining target user based on building, it can be directed to target user, based on certain
Marketing platform market, and obtained seed user after marketing, and can be used as positive sample for training again, optimizes
Prediction model, after repeatedly marketing, model will more and more precisely, so that marketing to the prediction effect of target user
And the user of marketing message is received, obtain good experience.
Below with reference to one, more specifically example, the () method provided this specification are illustrated.
Assuming that certain payment platform needs to promote a new software to the user of the platform, and carried out primary small-scale
Marketing, obtain the seed user that a batch installs the software by the connection downloading of push, it is assumed that quantity totally 100 ten thousand.
1) the matching degree prediction model of user and target scene are constructed.
1,000,000 obtained seed users will be marketed for the first time as positive sample, and from the remaining user group of platform
1,000,000 users of random selection are as negative sample.The platform is used according to the user information and user registering, share in platform
Or the behavioral datas such as login, search, payment of association platform progress, extract several identity characteristics of the user as positive negative sample
With behavioural characteristic, obtain the characteristic value of each user, and with 1 for positive sample label value, with 0 for negative sample label value.
For target scene, determine feature relevant to marketing in the scene, such as characterize the software target user,
Several features of the service range of the software etc. extract these features and obtain characteristic value.
1 User DN N and 1 scene DNN is constructed by research staff in advance, and initialization is completed to 2 DNN, then
Carry out the iteration of model training.
In each iteration, as shown in figure 3, the characteristic value of 1 user's training sample is inputted User DN N, to obtain
The user vector of User DN N output；Also, as shown in figure 4, by the characteristic value input scene DNN of target scene, to must show up
The scene vector of scape DNN output.
Then as shown in figure 5, the user vector that User DN N is exported and the scene vector of scene DNN output carry out dot product
Operation obtains the scalar value of one [0,1], the predicted value for the matching degree prediction model output as trained, using to application
The label value (0 or 1) of family training sample verifies the accuracy rate of the predicted value.Specifically, if label value is 1, according to pre-
Measured value and 1 gap calculate accuracy rate；And if label value is 0, accuracy rate is calculated according to the gap of predicted value and 0.
User DN N and scene DNN are optimized according to the accuracy rate of predicted value, then carry out next iteration, until
The accuracy rate of predicted value reaches preset requirement, can stop iteration, obtains the matching degree prediction mould that last time iteration obtains
Type.
2) based on constructed matching degree prediction model, the target user of target scene is predicted.
It, can user to payment platform and target after training obtains the matching degree prediction model of user and target scene
The matching degree of scene is predicted, and the target user of this marketing is determined according to prediction result.
But in actual conditions, the number of users of payment platform is larger (such as 1,000,000,000), and scheduled target user's number is opposite
Less (such as 10,000,000), the efficiency all predicted with input model per family is lower, therefore can first based on it is some more
Rough condition screens platform user.
It may for instance be considered that the long period was not logged in the user of the payment platform, even if being pushed this marketing letter
Breath, the probability for becoming seed user is also smaller, that is, the user and target scene can be determined by not needing input prediction model
Matching degree is lower, therefore can only retain user logged in nearly 30 days；In another example, it is assumed that the software is automatic for certain money of arranging in pairs or groups
The software that vending machine uses, and this vending machine is only launched in Beijing, the Shanghai city Deng Jizuo, then it can be according to user's
City label is screened, and only Hold sticker is the user in the cities such as Beijing, Shanghai；Etc..
It, can also be with even if not input prediction model in addition, be determined as a collection of user of seed user in first time marketing
Determination is higher with the matching degree of target scene, therefore can filter out this certain customers, it is not necessary to input prediction model, thus into one
Step improves forecasting efficiency.
By above-mentioned preliminary screening, a large number of users of the payment platform can be limited to the alternative of less quantity
User (such as 20,000,000) using the platform or is associated with what platform carried out according to the user information and user registered, shared in platform
The behavioral datas such as login, search, payment, extract several identity characteristics and behavioural characteristic of alternative user, obtain each user's
Characteristic value obtains the user vector of alternative user after characteristic value is inputted User DN N shown in Fig. 3.
User vector and scene vector (can directly be used in model training in last time iteration, according to scene DNN
Obtained vector) point multiplication operation is done, the predicted value of each alternative user can be obtained, the predicted value of alternative user is ranked up,
Preceding 10,000,000 users therein are taken, the target user of this marketing scene is determined as.
As it can be seen that using above scheme, multinomial identity characteristic and multinomial behavioural characteristic and target scene based on user
Multinomial feature obtains user and target on the basis of considering relevance internal between each feature using depth learning technology
Matching degree between scene, so that identified target group are more accurate.
Corresponding to above method embodiment, this specification embodiment also provides a kind of user and the matching degree of target scene is pre-
The training device of model is surveyed, it is shown in Figure 6, the apparatus may include:
Sample obtains module 110, for obtaining user's training sample set；User's training sample is concentrated with seed user
As positive sample and using non-seed user as negative sample, the characteristic value of any user training sample includes several identity characteristics
With several behavioural characteristics；
Model determining module 120, for determining initial matching degree prediction model；
Optimization processing module, for being iterated processing using following submodule, until the standard of the predicted value of model output
True rate reaches preset requirement:
Predicted value obtains submodule 131, for concentrating user's training sample the characteristic value of any sample to input currently
Matching degree prediction model, the matching degree predicted value exported；Wherein, the predicted value obtains submodule 131 and is specifically used for: root
According to the current user vector operation rule of model, characteristic value is carried out user vector is calculated；It obtains previously according to target field
The scene vector of several feature calculations of scape；Default vector operation is carried out to user vector and scene vector and obtains predicted value；
Accuracy rate judging submodule 132, the accuracy rate of the label value verification predicted value of input sample for using；
Optimization processing submodule 133, for judging whether the accuracy rate reaches preset requirement；If not up to, basis
The accuracy rate optimizes processing to current matching degree prediction model；The optimization processing includes at least: in Optimized model
Current user vector operation rule；
Output module 140, for being determined as current matching degree prediction model to be used for target scene after iteration
Target user's prediction model.
In a kind of specific embodiment that this specification provides, the predicted value obtain submodule specifically can use with
Lower unit is according to several feature calculation scene vectors of target scene:
Feature extraction unit obtains the characteristic value of target scene for extracting several default features of target scene；
Regular obtaining unit, for obtaining preset scene vector operation rule；
Vector calculation unit, for calculating the characteristic value of target scene according to the scene vector operation rule,
Obtain scene vector.
In a kind of specific embodiment that this specification provides, the rule obtaining unit specifically can be used for:
Obtain the current scene vector operation rule of the matching degree prediction model.
In a kind of specific embodiment that this specification provides, optimization that the optimization processing submodule carries out model
Handling to include:
Optimize scene vector operation rule current in Optimized model.
This specification embodiment also provides the device of prediction target user based on the matching degree prediction model a kind of, ginseng
As shown in Figure 7, the apparatus may include:
Characteristics extraction module 210, the value and several behaviors for extracting the preset several identity characteristics of alternative user are special
The value of sign obtains the characteristic value of alternative user；
Predicted value obtains module 220, for extracted characteristic value to be inputted the matching degree prediction model, obtains model
The matching degree predicted value of output；
Target user's determining module 230, for judging whether the alternative user meets according to the matching degree predicted value
Preset condition；If so, determining that the alternative user is the target user of the target scene.
In a kind of specific embodiment that this specification provides, the characteristics extraction module 210 may include:
Alternative user collection obtains submodule, includes several alternative users in the set for obtaining alternative user collection；
Any alternative user for concentrating for the alternative user: characteristics extraction submodule obtains the alternative use
The identity data and behavioral data at family；The value of the preset several identity characteristics of the alternative user is extracted according to the identity data,
And the value of the preset several behavioural characteristics of the alternative user is extracted according to the behavioral data, obtain the characteristic value of alternative user.
In a kind of specific embodiment that this specification provides, the alternative user collection obtains submodule, may include:
Initial user collection obtaining unit includes several unscreened initial for obtaining initial user collection, in the set
User；
Screening conditions collection determination unit, for determining the screening conditions collection for being directed to the initial user collection；
Initial user screening unit, every screening conditions for being concentrated using the screening conditions, to the initial use
Each initial user that family is concentrated is screened, and the alternative user collection being made of the user for meeting every screening conditions is obtained.
In a kind of specific embodiment that this specification provides, the screening conditions collection may include:
Label filtration condition, for whether there are target labels to screen as condition to user；And/or
Behavior screening conditions, for whether there is goal behavior to screen as condition to user；And/or
Seed screening conditions, for whether to be that seed user as condition screens user.
The function of modules and the realization process of effect are specifically detailed in the above method and correspond to step in above-mentioned apparatus
Realization process, details are not described herein.
This specification embodiment also provides a kind of computer equipment, includes at least memory, processor and is stored in
On reservoir and the computer program that can run on a processor, wherein processor realizes model above-mentioned when executing described program
Training method, the method that target user is determined based on model.This method includes at least:
A kind of training method of the matching degree prediction model of user and target scene, this method comprises:
Obtain user's training sample set；User's training sample is concentrated using seed user as positive sample and with non-kind
For child user as negative sample, the characteristic value of any user training sample includes several identity characteristics and several behavioural characteristics；
It determines initial matching degree prediction model, is iterated processing using following steps, until the prediction of model output
The accuracy rate of value reaches preset requirement:
It concentrates the characteristic value of any sample to input current matching degree prediction model user's training sample, is exported
Matching degree predicted value；It wherein, include: the user vector operation rule current according to model according to the method that characteristic value obtains predicted value
Then, characteristic value is carried out that user vector is calculated；Obtain the scene vector of several feature calculations previously according to target scene；
Default vector operation is carried out to user vector and scene vector and obtains predicted value；
Use the accuracy rate of the label value verification predicted value of institute's input sample；
Judge whether the accuracy rate reaches preset requirement；If not up to, according to the accuracy rate to current matching
Degree prediction model optimizes processing；The optimization processing includes at least: current user vector operation rule in Optimized model；
After iteration, the target user that current matching degree prediction model is determined as being used for target scene is predicted into mould
Type.
A method of the prediction target user based on the matching degree prediction model, this method comprises:
The value of the preset several identity characteristics of alternative user and the value of several behavioural characteristics are extracted, the spy of alternative user is obtained
Value indicative；
Extracted characteristic value is inputted into the matching degree prediction model, obtains the matching degree predicted value of model output；
According to the matching degree predicted value, judge whether the alternative user meets preset condition；If so, described in determining
Alternative user is the target user of the target scene.
Fig. 8 shows one kind provided by this specification embodiment and more specifically calculates device hardware structural schematic diagram,
The equipment may include: processor 1010, memory 1020, input/output interface 1030, communication interface 1040 and bus
1050.Wherein processor 1010, memory 1020, input/output interface 1030 and communication interface 1040 are real by bus 1050
The now communication connection inside equipment each other.
Processor 1010 can use general CPU (Central Processing Unit, central processing unit), micro- place
Reason device, application specific integrated circuit (Application Specific Integrated Circuit, ASIC) or one
Or the modes such as multiple integrated circuits are realized, for executing relative program, to realize technical side provided by this specification embodiment
Case.
Memory 1020 can use ROM (Read Only Memory, read-only memory), RAM (Random Access
Memory, random access memory), static storage device, the forms such as dynamic memory realize.Memory 1020 can store
Operating system and other applications are realizing technical solution provided by this specification embodiment by software or firmware
When, relevant program code is stored in memory 1020, and execution is called by processor 1010.
Input/output interface 1030 is for connecting input/output module, to realize information input and output.Input and output/
Module can be used as component Configuration (not shown) in a device, can also be external in equipment to provide corresponding function.Wherein
Input equipment may include keyboard, mouse, touch screen, microphone, various kinds of sensors etc., output equipment may include display,
Loudspeaker, vibrator, indicator light etc..
Communication interface 1040 is used for connection communication module (not shown), to realize the communication of this equipment and other equipment
Interaction.Wherein communication module can be realized by wired mode (such as USB, cable etc.) and be communicated, can also be wirelessly
(such as mobile network, WIFI, bluetooth etc.) realizes communication.
Bus 1050 include an access, equipment various components (such as processor 1010, memory 1020, input/it is defeated
Outgoing interface 1030 and communication interface 1040) between transmit information.
It should be noted that although above equipment illustrates only processor 1010, memory 1020, input/output interface
1030, communication interface 1040 and bus 1050, but in the specific implementation process, which can also include realizing normal fortune
Other assemblies necessary to row.In addition, it will be appreciated by those skilled in the art that, it can also be only comprising real in above equipment
Component necessary to existing this specification example scheme, without including all components shown in figure.
This specification embodiment also provides a kind of computer readable storage medium, is stored thereon with computer program, the journey
The method realized model training method above-mentioned when sequence is executed by processor, determine target user based on model.This method is at least
Include:
A kind of training method of the matching degree prediction model of user and target scene, this method comprises:
Obtain user's training sample set；User's training sample is concentrated using seed user as positive sample and with non-kind
For child user as negative sample, the characteristic value of any user training sample includes several identity characteristics and several behavioural characteristics；
It determines initial matching degree prediction model, is iterated processing using following steps, until the prediction of model output
The accuracy rate of value reaches preset requirement:
It concentrates the characteristic value of any sample to input current matching degree prediction model user's training sample, is exported
Matching degree predicted value；It wherein, include: the user vector operation rule current according to model according to the method that characteristic value obtains predicted value
Then, characteristic value is carried out that user vector is calculated；Obtain the scene vector of several feature calculations previously according to target scene；
Default vector operation is carried out to user vector and scene vector and obtains predicted value；
Use the accuracy rate of the label value verification predicted value of institute's input sample；
Judge whether the accuracy rate reaches preset requirement；If not up to, according to the accuracy rate to current matching
Degree prediction model optimizes processing；The optimization processing includes at least: current user vector operation rule in Optimized model；
After iteration, the target user that current matching degree prediction model is determined as being used for target scene is predicted into mould
Type.
A method of the prediction target user based on the matching degree prediction model, this method comprises:
The value of the preset several identity characteristics of alternative user and the value of several behavioural characteristics are extracted, the spy of alternative user is obtained
Value indicative；
Extracted characteristic value is inputted into the matching degree prediction model, obtains the matching degree predicted value of model output；
According to the matching degree predicted value, judge whether the alternative user meets preset condition；If so, described in determining
Alternative user is the target user of the target scene.
Computer-readable medium includes permanent and non-permanent, removable and non-removable media can be by any method
Or technology come realize information store.Information can be computer readable instructions, data structure, the module of program or other data.
The example of the storage medium of computer includes, but are not limited to phase change memory (PRAM), static random access memory (SRAM), moves
State random access memory (DRAM), other kinds of random access memory (RAM), read-only memory (ROM), electric erasable
Programmable read only memory (EEPROM), flash memory or other memory techniques, read-only disc read only memory (CD-ROM) (CD-ROM),
Digital versatile disc (DVD) or other optical storage, magnetic cassettes, tape magnetic disk storage or other magnetic storage devices
Or any other non-transmission medium, can be used for storage can be accessed by a computing device information.As defined in this article, it calculates
Machine readable medium does not include temporary computer readable media (transitory media), such as the data-signal and carrier wave of modulation.
As seen through the above description of the embodiments, those skilled in the art can be understood that this specification
Embodiment can be realized by means of software and necessary general hardware platform.Based on this understanding, this specification is implemented
Substantially the part that contributes to existing technology can be embodied in the form of software products the technical solution of example in other words,
The computer software product can store in storage medium, such as ROM/RAM, magnetic disk, CD, including some instructions are to make
It is each to obtain computer equipment (can be personal computer, server or the network equipment etc.) execution this specification embodiment
Method described in certain parts of a embodiment or embodiment.
System, device, module or the unit that above-described embodiment illustrates can specifically realize by computer chip or entity,
Or it is realized by the product with certain function.A kind of typically to realize that equipment is computer, the concrete form of computer can
To be personal computer, laptop computer, cellular phone, camera phone, smart phone, personal digital assistant, media play
In device, navigation equipment, E-mail receiver/send equipment, game console, tablet computer, wearable device or these equipment
The combination of any several equipment.
All the embodiments in this specification are described in a progressive manner, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for device reality
For applying example, since it is substantially similar to the method embodiment, so describing fairly simple, related place is referring to embodiment of the method
Part explanation.The apparatus embodiments described above are merely exemplary, wherein described be used as separate part description
Module may or may not be physically separated, can be each module when implementing this specification example scheme
Function realize in the same or multiple software and or hardware.Can also select according to the actual needs part therein or
Person's whole module achieves the purpose of the solution of this embodiment.Those of ordinary skill in the art are not the case where making the creative labor
Under, it can it understands and implements.
The above is only the specific embodiment of this specification embodiment, it is noted that for the general of the art
For logical technical staff, under the premise of not departing from this specification embodiment principle, several improvements and modifications can also be made, this
A little improvements and modifications also should be regarded as the protection scope of this specification embodiment.
Claims (17)
1. a kind of training method of the matching degree prediction model of user and target scene, this method comprises:
Obtain user's training sample set；User's training sample is concentrated using seed user as positive sample and with non-seed use
As negative sample, the characteristic value of any user training sample includes several identity characteristics and several behavioural characteristics at family；
It determines initial matching degree prediction model, is iterated processing using following steps, until the predicted value of model output
Accuracy rate reaches preset requirement:
The characteristic value of any sample is concentrated to input current matching degree prediction model, the matching exported user's training sample
Spend predicted value；It wherein, include: the user vector operation rule current according to model according to the method that characteristic value obtains predicted value,
Characteristic value is carried out user vector is calculated；Obtain the scene vector of several feature calculations previously according to target scene；It is right
User vector and scene vector carry out default vector operation and obtain predicted value；
Use the accuracy rate of the label value verification predicted value of institute's input sample；
Judge whether the accuracy rate reaches preset requirement；If not up to, pre- to current matching degree according to the accuracy rate
It surveys model and optimizes processing；The optimization processing includes at least: current user vector operation rule in Optimized model；
After iteration, current matching degree prediction model is determined as to be used for target user's prediction model of target scene.
2. according to the method described in claim 1, the method for several feature calculation scene vectors according to target scene, tool
Body includes:
Several default features for extracting target scene, obtain the characteristic value of target scene；
Obtain preset scene vector operation rule；
According to the scene vector operation rule, the characteristic value of target scene is calculated, scene vector is obtained.
3. according to the method described in claim 2, described obtain preset scene vector operation rule, comprising:
Obtain the current scene vector operation rule of the matching degree prediction model.
4. according to the method described in claim 3, the optimization processing further include:
Optimize scene vector operation rule current in Optimized model.
5. a kind of method of the prediction target user based on any one of Claims 1-4 matching degree prediction model, this method
Include:
The value of the preset several identity characteristics of alternative user and the value of several behavioural characteristics are extracted, the feature of alternative user is obtained
Value；
Extracted characteristic value is inputted into the matching degree prediction model, obtains the matching degree predicted value of model output；
According to the matching degree predicted value, judge whether the alternative user meets preset condition；If so, determining described alternative
User is the target user of the target scene.
6. according to the method described in claim 5, the value for extracting the preset several identity characteristics of alternative user and several rows
The value being characterized obtains the characteristic value of alternative user, comprising:
Alternative user collection is obtained, includes several alternative users in the set；
Any alternative user concentrated for the alternative user:
Obtain the identity data and behavioral data of the alternative user；
The value of the preset several identity characteristics of the alternative user is extracted according to the identity data, and is mentioned according to the behavioral data
The value for taking the preset several behavioural characteristics of the alternative user, obtains the characteristic value of alternative user.
7. according to the method described in claim 6, described obtain the alternative user collection including several alternative users, comprising:
Initial user collection is obtained, includes several unscreened initial users in the set；
Determine the screening conditions collection for being directed to the initial user collection；
The every screening conditions concentrated using the screening conditions, each initial user concentrated to the initial user are sieved
Choosing, obtains the alternative user collection being made of the user for meeting every screening conditions.
8. according to the method described in claim 7, the screening conditions collection, comprising:
Label filtration condition, for whether there are target labels to screen as condition to user；And/or
Behavior screening conditions, for whether there is goal behavior to screen as condition to user；And/or
Seed screening conditions, for whether to be that seed user as condition screens user.
9. a kind of training device of the matching degree prediction model of user and target scene, the device include:
Sample obtains module, for obtaining user's training sample set；User's training sample is concentrated using seed user as just
Sample and using non-seed user as negative sample, the characteristic value of any user training sample include several identity characteristics with it is several
Behavioural characteristic；
Model determining module, for determining initial matching degree prediction model；
Optimization processing module, for being iterated processing using following submodule, until the accuracy rate of the predicted value of model output
Reach preset requirement:
Predicted value obtains submodule, pre- for concentrating user's training sample the characteristic value of any sample to input current matching degree
Survey model, the matching degree predicted value exported；Wherein, the predicted value obtains submodule and is specifically used for: current according to model
User vector operation rule, characteristic value is carried out user vector is calculated；Obtain several spies previously according to target scene
Levy the scene vector calculated；Default vector operation is carried out to user vector and scene vector and obtains predicted value；
Accuracy rate judging submodule, the accuracy rate of the label value verification predicted value of input sample for using；
Optimization processing submodule, for judging whether the accuracy rate reaches preset requirement；If not up to, according to described accurate
Rate optimizes processing to current matching degree prediction model；The optimization processing includes at least: current use in Optimized model
Family vector operation rule；
Output module, for after iteration, the target that current matching degree prediction model is determined as being used for target scene to be used
Family prediction model.
10. device according to claim 9, the predicted value obtains submodule and specifically utilizes with lower unit according to target field
Several feature calculation scene vectors of scape:
Feature extraction unit obtains the characteristic value of target scene for extracting several default features of target scene；
Regular obtaining unit, for obtaining preset scene vector operation rule；
Vector calculation unit, for calculating the characteristic value of target scene, obtaining according to the scene vector operation rule
Scene vector.
11. device according to claim 10, the rule obtaining unit, are specifically used for:
Obtain the current scene vector operation rule of the matching degree prediction model.
12. device according to claim 11, the optimization processing that the optimization processing submodule carries out model further include:
Optimize scene vector operation rule current in Optimized model.
13. a kind of device of the prediction target user based on any one of claim 9 to the 12 matching degree prediction model, the dress
It sets and includes:
Characteristics extraction module, for extracting the value of the preset several identity characteristics of alternative user and the value of several behavioural characteristics,
Obtain the characteristic value of alternative user；
Predicted value obtains module, for extracted characteristic value to be inputted the matching degree prediction model, obtains model output
Matching degree predicted value；
Target user's determining module, for judging whether the alternative user meets default item according to the matching degree predicted value
Part；If so, determining that the alternative user is the target user of the target scene.
14. device according to claim 13, the characteristics extraction module, comprising:
Alternative user collection obtains submodule, includes several alternative users in the set for obtaining alternative user collection；
Any alternative user for concentrating for the alternative user: characteristics extraction submodule obtains the alternative user
Identity data and behavioral data；The value of the preset several identity characteristics of the alternative user, and root are extracted according to the identity data
The value that the preset several behavioural characteristics of the alternative user are extracted according to the behavioral data, obtains the characteristic value of alternative user.
15. device according to claim 14, the alternative user collection obtains submodule, comprising:
Initial user collection obtaining unit includes several unscreened initial users in the set for obtaining initial user collection；
Screening conditions collection determination unit, for determining the screening conditions collection for being directed to the initial user collection；
Initial user screening unit, every screening conditions for being concentrated using the screening conditions, to the initial user collection
In each initial user screened, obtain the alternative user collection being made of the user for meeting every screening conditions.
16. device according to claim 15, the screening conditions collection, comprising:
Label filtration condition, for whether there are target labels to screen as condition to user；And/or
Behavior screening conditions, for whether there is goal behavior to screen as condition to user；And/or
Seed screening conditions, for whether to be that seed user as condition screens user.
17. a kind of computer equipment including memory, processor and stores the meter that can be run on a memory and on a processor
Calculation machine program, wherein the processor realizes method as claimed in any one of claims 1 to 8 when executing described program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811403695.2A CN110008973B (en) | 2018-11-23 | 2018-11-23 | Model training method, method and device for determining target user based on model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811403695.2A CN110008973B (en) | 2018-11-23 | 2018-11-23 | Model training method, method and device for determining target user based on model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110008973A true CN110008973A (en) | 2019-07-12 |
| CN110008973B CN110008973B (en) | 2023-05-02 |
Family
ID=67165014
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811403695.2A Active CN110008973B (en) | 2018-11-23 | 2018-11-23 | Model training method, method and device for determining target user based on model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110008973B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110647622A (en) * | 2019-09-29 | 2020-01-03 | 北京金山安全软件有限公司 | Interactive data validity identification method and device |
| CN110852755A (en) * | 2019-11-06 | 2020-02-28 | 支付宝(杭州)信息技术有限公司 | User identity identification method and device for transaction scene |
| CN111222923A (en) * | 2020-01-13 | 2020-06-02 | 秒针信息技术有限公司 | Method and device for judging potential customer, electronic equipment and storage medium |
| CN111414540A (en) * | 2020-03-20 | 2020-07-14 | 张明 | Online learning recommendation method and device, online learning system and server |
| CN111429214A (en) * | 2020-03-13 | 2020-07-17 | 贝壳技术有限公司 | Transaction data-based buyer and seller matching method and device |
| CN111708810A (en) * | 2020-06-17 | 2020-09-25 | 北京世纪好未来教育科技有限公司 | Model optimization recommendation method and device and computer storage medium |
| CN113378067A (en) * | 2021-07-13 | 2021-09-10 | 深圳前海微众银行股份有限公司 | Message recommendation method, device, medium, and program product based on user mining |
| CN113688326A (en) * | 2021-10-26 | 2021-11-23 | 腾讯科技（深圳）有限公司 | Recommendation method, device, equipment and computer readable storage medium |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150294246A1 (en) * | 2014-04-10 | 2015-10-15 | International Business Machines Corporation | Selecting optimal training data set for service contract prediction |
| CN105183781A (en) * | 2015-08-14 | 2015-12-23 | 百度在线网络技术（北京）有限公司 | Information recommendation method and apparatus |
| CN105447730A (en) * | 2015-12-25 | 2016-03-30 | 腾讯科技（深圳）有限公司 | Target user orientation method and device |
| CN107220852A (en) * | 2017-05-26 | 2017-09-29 | 北京小度信息科技有限公司 | Method, device and server for determining target recommended user |
| CN107424043A (en) * | 2017-06-15 | 2017-12-01 | 北京三快在线科技有限公司 | A kind of Products Show method and device, electronic equipment |
| CN107578332A (en) * | 2017-09-22 | 2018-01-12 | 深圳乐信软件技术有限公司 | A kind of method, apparatus, equipment and storage medium for recommending cash commodity |
| CN108022116A (en) * | 2016-11-01 | 2018-05-11 | 北京京东尚科信息技术有限公司 | To the method, system and terminal device of user modeling |
| CN108021931A (en) * | 2017-11-20 | 2018-05-11 | 阿里巴巴集团控股有限公司 | A kind of data sample label processing method and device |
| CN108074116A (en) * | 2016-11-09 | 2018-05-25 | 阿里巴巴集团控股有限公司 | Information providing method and device |
| CN108334887A (en) * | 2017-01-19 | 2018-07-27 | 腾讯科技（深圳）有限公司 | A kind of user's choosing method and device |
| CN108510336A (en) * | 2017-02-23 | 2018-09-07 | 北京京东尚科信息技术有限公司 | Method, apparatus, electronic equipment and storage medium for determining user data information |
| CN108648093A (en) * | 2018-04-23 | 2018-10-12 | 腾讯科技（深圳）有限公司 | Data processing method, device and equipment |
-
2018
- 2018-11-23 CN CN201811403695.2A patent/CN110008973B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150294246A1 (en) * | 2014-04-10 | 2015-10-15 | International Business Machines Corporation | Selecting optimal training data set for service contract prediction |
| CN105183781A (en) * | 2015-08-14 | 2015-12-23 | 百度在线网络技术（北京）有限公司 | Information recommendation method and apparatus |
| CN105447730A (en) * | 2015-12-25 | 2016-03-30 | 腾讯科技（深圳）有限公司 | Target user orientation method and device |
| CN108022116A (en) * | 2016-11-01 | 2018-05-11 | 北京京东尚科信息技术有限公司 | To the method, system and terminal device of user modeling |
| CN108074116A (en) * | 2016-11-09 | 2018-05-25 | 阿里巴巴集团控股有限公司 | Information providing method and device |
| CN108334887A (en) * | 2017-01-19 | 2018-07-27 | 腾讯科技（深圳）有限公司 | A kind of user's choosing method and device |
| CN108510336A (en) * | 2017-02-23 | 2018-09-07 | 北京京东尚科信息技术有限公司 | Method, apparatus, electronic equipment and storage medium for determining user data information |
| CN107220852A (en) * | 2017-05-26 | 2017-09-29 | 北京小度信息科技有限公司 | Method, device and server for determining target recommended user |
| CN107424043A (en) * | 2017-06-15 | 2017-12-01 | 北京三快在线科技有限公司 | A kind of Products Show method and device, electronic equipment |
| CN107578332A (en) * | 2017-09-22 | 2018-01-12 | 深圳乐信软件技术有限公司 | A kind of method, apparatus, equipment and storage medium for recommending cash commodity |
| CN108021931A (en) * | 2017-11-20 | 2018-05-11 | 阿里巴巴集团控股有限公司 | A kind of data sample label processing method and device |
| CN108648093A (en) * | 2018-04-23 | 2018-10-12 | 腾讯科技（深圳）有限公司 | Data processing method, device and equipment |
Non-Patent Citations (3)
| Title |
|---|
| KAI SHU等: "User identity linkage across online social networks: A review" * |
| 冯鸣等: "交互设计思维在UGC产品设计中的应用" * |
| 张旭等: "电信行业基于种子用户群扩展技术的定向营销研究与应用" * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110647622A (en) * | 2019-09-29 | 2020-01-03 | 北京金山安全软件有限公司 | Interactive data validity identification method and device |
| CN110852755A (en) * | 2019-11-06 | 2020-02-28 | 支付宝(杭州)信息技术有限公司 | User identity identification method and device for transaction scene |
| CN111222923A (en) * | 2020-01-13 | 2020-06-02 | 秒针信息技术有限公司 | Method and device for judging potential customer, electronic equipment and storage medium |
| CN111222923B (en) * | 2020-01-13 | 2023-12-15 | 秒针信息技术有限公司 | Method and device for judging potential clients, electronic equipment and storage medium |
| CN111429214A (en) * | 2020-03-13 | 2020-07-17 | 贝壳技术有限公司 | Transaction data-based buyer and seller matching method and device |
| CN111429214B (en) * | 2020-03-13 | 2023-06-09 | 贝壳技术有限公司 | Transaction data-based buyer and seller matching method and device |
| CN111414540B (en) * | 2020-03-20 | 2021-01-15 | 重庆探程数字科技有限公司 | Online learning recommendation method and device, online learning system and server |
| CN111414540A (en) * | 2020-03-20 | 2020-07-14 | 张明 | Online learning recommendation method and device, online learning system and server |
| CN111708810B (en) * | 2020-06-17 | 2022-05-27 | 北京世纪好未来教育科技有限公司 | Model optimization recommendation method and device and computer storage medium |
| CN111708810A (en) * | 2020-06-17 | 2020-09-25 | 北京世纪好未来教育科技有限公司 | Model optimization recommendation method and device and computer storage medium |
| CN113378067A (en) * | 2021-07-13 | 2021-09-10 | 深圳前海微众银行股份有限公司 | Message recommendation method, device, medium, and program product based on user mining |
| CN113688326A (en) * | 2021-10-26 | 2021-11-23 | 腾讯科技（深圳）有限公司 | Recommendation method, device, equipment and computer readable storage medium |
| CN113688326B (en) * | 2021-10-26 | 2022-02-08 | 腾讯科技（深圳）有限公司 | Recommendation method, device, equipment and computer readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110008973B (en) | 2023-05-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110008973A (en) | A kind of model training method, the method and device that target user is determined based on model | |
| US10872298B2 (en) | Machine learning and prediction using graph communities | |
| CN103823908B (en) | Content recommendation method and server based on user preference | |
| CN109961142B (en) | Neural network optimization method and device based on meta learning | |
| CN107273436A (en) | The training method and trainer of a kind of recommended models | |
| US20190385213A1 (en) | System for presenting items in online environment based on previous item selections | |
| CN107784390A (en) | Recognition methods, device, electronic equipment and the storage medium of subscriber lifecycle | |
| CN109299356B (en) | Activity recommendation method and device based on big data, electronic equipment and storage medium | |
| CN104239338A (en) | Information recommendation method and information recommendation device | |
| CN109255644A (en) | For providing the method and apparatus of the marketing management data of optimization distribution and logistics | |
| US20200234218A1 (en) | Systems and methods for entity performance and risk scoring | |
| EP3764303A1 (en) | Information processing device, etc. for calculating prediction data | |
| US11227309B2 (en) | Method and system for optimizing user grouping for advertisement | |
| CN110689402A (en) | Method and device for recommending merchants, electronic equipment and readable storage medium | |
| CN112215448A (en) | Method and device for distributing customer service | |
| CN112380449B (en) | Information recommendation method, model training method and related device | |
| CN104618347B (en) | A kind of game events processing unit and method, the network platform | |
| CN109886769A (en) | A kind of the displaying optimization method and device of virtual objects | |
| CN107741967A (en) | Method, apparatus and electronic equipment for behavioral data processing | |
| JP2019049836A (en) | Estimation device and estimation method and estimation program | |
| CN109711917A (en) | Information-pushing method and device | |
| CN112395499B (en) | Information recommendation method and device, electronic equipment and storage medium | |
| CN113822734A (en) | Method and apparatus for generating information | |
| JP2016192013A (en) | Sales representative candidate extraction system | |
| CN109977979A (en) | Position method, apparatus, electronic equipment and the storage medium of seed user |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20200927 Address after: Cayman Enterprise Centre, 27 Hospital Road, George Town, Grand Cayman Islands Applicant after: Advanced innovation technology Co.,Ltd. Address before: A four-storey 847 mailbox in Grand Cayman Capital Building, British Cayman Islands Applicant before: Alibaba Group Holding Ltd. Effective date of registration: 20200927 Address after: Cayman Enterprise Centre, 27 Hospital Road, George Town, Grand Cayman Islands Applicant after: Innovative advanced technology Co.,Ltd. Address before: Cayman Enterprise Centre, 27 Hospital Road, George Town, Grand Cayman Islands Applicant before: Advanced innovation technology Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |