CN110209926A - Merchant recommendation method, device, electronic equipment and readable storage medium storing program for executing - Google Patents
Merchant recommendation method, device, electronic equipment and readable storage medium storing program for executing Download PDFInfo
- Publication number
- CN110209926A CN110209926A CN201910324959.3A CN201910324959A CN110209926A CN 110209926 A CN110209926 A CN 110209926A CN 201910324959 A CN201910324959 A CN 201910324959A CN 110209926 A CN110209926 A CN 110209926A
- Authority
- CN
- China
- Prior art keywords
- neural network
- network model
- recurrent neural
- businessman
- recognition
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- General Business, Economics & Management (AREA)
- Image Analysis (AREA)
Abstract
The present disclosure discloses a kind of merchant recommendation method, device, electronic equipment and readable storage medium storing program for executing.The method, comprising: obtain the click sequence information of target user;The recommendation score of each alternative businessman is obtained by preset recommended models according to the click sequence information;Based on the recommendation score of the alternative businessman, the target for obtaining the target user recommends businessman and pushes to the target user;Wherein, the recommended models are the built-up pattern by obtain after joint training to the double-deck Recognition with Recurrent Neural Network model and deep neural network model.Thus the technical issues of poor accuracy of existing recommended method is solved.Achieve the beneficial effect for improving the accuracy for recommending businessman.
Description
Technical field
This disclosure relates to machine learning techniques field, and in particular to a kind of merchant recommendation method, device, electronic equipment and can
Read storage medium.
Background technique
With the fast development of internet and machine learning techniques, more and more e-commerce platforms etc. pass through recommendation
System recommends the personalized recommendation businessman for meeting its demand to user.And recommender system can generally be directed to active user, to each
Alternative businessman is ranked up, with the recommendation businessman that determination is final.
The application in scene is recommended to have been achieved for certain effect, such as FNN in machine learning model at present
(Feedforward Neural Network, feedforward neural network), RNN (Recurrent Neural Network, circulation mind
Through network) and DNN (Deep Neural Networks, deep neural network) network on achieve certain effect.Industry at present
Inside also have for sequence information to be added in Xgb (eXtreme Gradient Boosting, extreme gradient are promoted) and DNN and carry out
The method of recommendation, but the feature that individually then training RNN learns RNN is needed to be input in DNN, and individually instructing
The training coordinate system of two models may be different when practicing machine learning model, thus the user's sequence letter for being easy to cause study to arrive
Breath training in machine learning model is invalid, and then influences recommendation results accuracy.It can be seen that existing suggested design is still deposited
The recommendation results poor accuracy the problem of.
Summary of the invention
The disclosure provides a kind of merchant recommendation method, device, electronic equipment and readable storage medium storing program for executing, partly or entirely to solve
The certainly relevant above problem of businessman's recommendation process in the prior art.
According to the disclosure in a first aspect, providing a kind of merchant recommendation method, comprising:
Obtain the click sequence information of target user;
The recommendation score of each alternative businessman is obtained by preset recommended models according to the click sequence information;
Based on the recommendation score of the alternative businessman, the target for obtaining the target user is recommended businessman and is pushed to described
Target user;
Wherein, the recommended models are by joining to the double-deck Recognition with Recurrent Neural Network model and deep neural network model
Close obtained built-up pattern after training, the bilayer Recognition with Recurrent Neural Network model include the first level Recognition with Recurrent Neural Network model and
Second level Recognition with Recurrent Neural Network model, and the input data of the second level Recognition with Recurrent Neural Network model includes described first
The output data of level Recognition with Recurrent Neural Network model.
According to the second aspect of the disclosure, a kind of businessman's recommendation apparatus is provided, comprising:
It clicks sequence information and obtains module, for obtaining the click sequence information of target user;
Recommendation score obtains module, for being obtained each according to the click sequence information by preset recommended models
The recommendation score of alternative businessman;
Target recommends businessman's acquisition module to obtain the target user for the recommendation score based on the alternative businessman
Target recommend businessman and to push to the target user;
Wherein, the recommended models are by joining to the double-deck Recognition with Recurrent Neural Network model and deep neural network model
Close obtained built-up pattern after training, the bilayer Recognition with Recurrent Neural Network model include the first level Recognition with Recurrent Neural Network model and
Second level Recognition with Recurrent Neural Network model, and the input data of the second level Recognition with Recurrent Neural Network model includes described first
The output data of level Recognition with Recurrent Neural Network model.
According to the third aspect of the disclosure, a kind of electronic equipment is provided, comprising:
Processor, memory and it is stored in the computer journey that can be run on the memory and on the processor
Sequence, which is characterized in that the processor realizes merchant recommendation method above-mentioned when executing described program.
According to the fourth aspect of the disclosure, provide a kind of readable storage medium storing program for executing, when the instruction in the storage medium by
When the processor of electronic equipment executes, so that electronic equipment is able to carry out merchant recommendation method above-mentioned.
According to the merchant recommendation method of the disclosure, the click sequence information of available target user;According to the click
Sequence information obtains the recommendation score of each alternative businessman by preset recommended models;Recommendation based on the alternative businessman
Scoring, the target for obtaining the target user recommend businessman and push to the target user;Wherein, the recommended models are logical
It crosses and the built-up pattern obtained after joint training is carried out to the double-deck Recognition with Recurrent Neural Network model and deep neural network model, it is described double
Layer Recognition with Recurrent Neural Network model includes the first level Recognition with Recurrent Neural Network model and the second level Recognition with Recurrent Neural Network model, and institute
The input data for stating the second level Recognition with Recurrent Neural Network model includes the output number of the first level Recognition with Recurrent Neural Network model
According to.Thus the technical issues of poor accuracy of existing recommended method is solved.Achieve the accuracy for improving and recommending businessman
Beneficial effect.
Above description is only the general introduction of disclosed technique scheme, in order to better understand the technological means of the disclosure,
And it can be implemented in accordance with the contents of the specification, and in order to allow above and other objects, features, and advantages of the present disclosure can
It is clearer and more comprehensible, below the special specific embodiment for lifting the disclosure.
Detailed description of the invention
By reading the following detailed description of the preferred embodiment, various other advantages and benefits are common for this field
Technical staff will become clear.The drawings are only for the purpose of illustrating a preferred embodiment, and is not considered as to the disclosure
Limitation.And throughout the drawings, the same reference numbers will be used to refer to the same parts.In the accompanying drawings:
Fig. 1 shows a kind of step flow chart of merchant recommendation method according to an embodiment of the present disclosure;
Fig. 2 shows the step flow charts according to another merchant recommendation method of an embodiment of the present disclosure;
Fig. 3 shows a kind of schematic diagram of click sequence information according to an embodiment of the present disclosure;
Fig. 4 shows a kind of structural schematic diagram of businessman's recommendation apparatus according to an embodiment of the present disclosure;And
Fig. 5 shows the structural schematic diagram of another businessman's recommendation apparatus according to an embodiment of the present disclosure.
Specific embodiment
Exemplary embodiments of the present disclosure are described in more detail below with reference to accompanying drawings.Although showing the disclosure in attached drawing
Exemplary embodiment, it being understood, however, that may be realized in various forms the disclosure without should be by embodiments set forth here
It is limited.On the contrary, these embodiments are provided to facilitate a more thoroughly understanding of the present invention, and can be by the scope of the present disclosure
It is fully disclosed to those skilled in the art.
Embodiment one
A kind of recommended method of embodiment of the present disclosure offer is provided.
Referring to Fig.1, a kind of step flow chart of recommended method in the embodiment of the present disclosure is shown.
Step 110, the click sequence information of target user is obtained.
In the embodiments of the present disclosure, in order to recommend the target for meeting its current state to recommend businessman to target user,
It then needs to obtain to determine that target recommends the coherent reference information of businessman.And in practical applications, user is in historical time
The interior businessman browsed by modes such as clicks can reflect its current demand to a certain extent.Therefore, in disclosure reality
It applies in example, available target user's clicks sequence information as the reference information for determining that its target recommends businessman.
Wherein, clicking sequence information may include that target user is browsed by modes such as clicks within a preset period of time
The characteristic information of each businessman, the user information of respective objects user, user and interactive information of each businessman etc., click sequence
It may include description user, businessman, the feature of context and user and businessman's interactive relation in column information.It is specific to click
The content for including in sequence information, and the data array etc. clicked in sequence information can carry out in advance according to demand
Setting, is not limited this embodiment of the present disclosure.
Step 120, pushing away for each alternative businessman is obtained by preset recommended models according to the click sequence information
Recommend scoring;Wherein, the recommended models are by joining to the double-deck Recognition with Recurrent Neural Network model and deep neural network model
Close obtained built-up pattern after training, the bilayer Recognition with Recurrent Neural Network model include the first level Recognition with Recurrent Neural Network model and
Second level Recognition with Recurrent Neural Network model, and the input data of the second level Recognition with Recurrent Neural Network model includes described first
The output data of level Recognition with Recurrent Neural Network model.
After acquiring the click sequence information of target user, then it can be obtained each by preset recommended models
The recommendation score of a alternative businessman.The content that alternative businessman therein specifically includes can be preset according to demand,
This embodiment of the present disclosure is not limited.
And when acquiring the recommendation score of each alternative businessman, the parameter for inputting recommended models may include clicking
The attribute information of sequence information and each alternative businessman.The content that wherein attribute information is specifically included can also according to demand into
Row is preset, and is not limited to this embodiment of the present disclosure.
Recommend in scene for example, selling businessman outside, can be set includes 7 discrete features and 119 companies in attribute information
Continuous feature.It also may include outside target user clicks within a preset period of time in the click sequence information of corresponding target user
Sell 7 discrete features and 119 continuous features of businessman, and the corresponding vegetable sequence, etc. for taking out businessman.
And then it then can be according to the attribute information of the click sequence information and each alternative businessman of target user, by default
Recommended models, obtain the recommendation score of each alternative businessman.
Recommended models therein are by the double-deck Recognition with Recurrent Neural Network (Recurrent Neural Networks, RNN)
Model and deep neural network (Deep Neural Network, DNN) model carry out the built-up pattern obtained after joint training.
Deep neural network is exactly profound neural network from literal upper understanding.
Recognition with Recurrent Neural Network be it is a kind of with sequence (sequence) data for input, carry out recurrence in the evolution tendency of sequence
(recursion) and all nodes (cycling element) press the recurrent neural network that chain type connects.Recognition with Recurrent Neural Network has memory
Property, parameter sharing and scheme clever complete (Turing completeness), therefore can be with very high efficiency to the non-linear of sequence
Feature is learnt.RNN model may include multiple types, such as LSTM (Long Short Term Memory, shot and long term note
Recall) network model, GRU network model, two-way RNN model, etc..
The double-deck Recognition with Recurrent Neural Network model is then understood that as the nerve as made of two Recognition with Recurrent Neural Network model constructions
Network model, it is assumed for example that include the first level Recognition with Recurrent Neural Network model and the second level in the double-deck Recognition with Recurrent Neural Network model
Recognition with Recurrent Neural Network model, and the input data of the second level Recognition with Recurrent Neural Network model includes the first level circulation
The output data of neural network model.At this point it is possible to which the input terminal of the first level Recognition with Recurrent Neural Network model is arranged as corresponding
The input terminal of the double-deck Recognition with Recurrent Neural Network model, the output end of the first level Recognition with Recurrent Neural Network model and the second level circulation mind
Input terminal connection through network model, the output end of the second level Recognition with Recurrent Neural Network model is as the corresponding double-deck circulation nerve net
The output end of network model.
Deep neural network is exactly profound neural network from literal upper understanding.In this many field, DNN can
Surmount the accuracy rate of the mankind.And the outstanding performance of DNN is derived from it can use that statistical learning method is extracted from original sensorial data
High-level characteristic obtains the Efficient Characterization of the input space in a large amount of data.DNN model may include depth convolutional neural networks
Model, FNN (feedforward neural network, feedforward neural network) model, etc..
In the embodiments of the present disclosure, it since the click sequence information of target user can be sequence form, can set
Set the combination that recommended models are bilayer RNN model and DNN model.Moreover, may include by two RNN model structures in recommended models
At the double-deck RNN model and at least one DNN model, can specifically be preset according to demand, and each RNN
The concrete type of the concrete type of model and the particular number of DNN model and each DNN model can carry out according to demand
It presets, this embodiment of the present disclosure is not limited.
Moreover, at this time when being trained to recommended models, it can be by the double-deck Recognition with Recurrent Neural Network model and depth
The built-up pattern that neural network model obtain after joint training is as recommended models.
Step 130, the recommendation score based on the alternative businessman, the target for obtaining the target user are recommended businessman and are pushed away
It send to the target user.
After the recommendation score for acquiring each alternative businessman, then it can be obtained according to the recommendation score of alternative businessman
It takes the target of the target user to recommend businessman and pushes to the target user.Wherein, in the embodiments of the present disclosure, Ke Yishe
The matching degree for setting the more high then corresponding alternative businessman of recommendation score and target user is higher, then can set according to demand at this time
A recommendation score critical value is set, and then obtains recommendation score and is more than or equal to the alternative businessman of the recommendation score critical value as target
The target of user recommends businessman, and then then businessman can be recommended to push to target user target;Alternatively, if setting is recommended to comment
Divide the matching degree of more high then corresponding alternative businessman and target user lower, then then available recommendation score is less than or equal to
The alternative businessman of the recommendation score critical value recommends businessman as the target of target user, and then then target can be recommended businessman
Push to target user.
Wherein it is possible to target recommendation businessman is sent to target user using any available means, moreover, target is recommended
The concrete mode that businessman is sent to target user can also be preset according to demand, this embodiment of the present disclosure is not added
To limit.
It, then can will be corresponding if it is a certain take-away businessman that target, which recommends businessman, for example, for taking out platform
The electronic business card etc. for taking out businessman is sent to target user, if target user clicks the electronic business card, can jump to phase
The take-away businessman answered.
In the embodiments of the present disclosure, pass through the click sequence information of acquisition target user;According to the click sequence information,
By preset recommended models, the recommendation score of each alternative businessman is obtained;Based on the recommendation score of the alternative businessman, obtain
The target of the target user recommends businessman and pushes to the target user;Wherein, the recommended models are by bilayer
Recognition with Recurrent Neural Network model and deep neural network model carry out the built-up pattern obtained after joint training, the double-deck circulation mind
Include the first level Recognition with Recurrent Neural Network model and the second level Recognition with Recurrent Neural Network model, and the second layer through network model
The input data of grade Recognition with Recurrent Neural Network model includes the output data of the first level Recognition with Recurrent Neural Network model.It achieves
Improve the beneficial effect for recommending businessman's accuracy.
Embodiment two
A kind of recommended method of embodiment of the present disclosure offer is provided.
Referring to Fig. 2, a kind of step flow chart of recommended method in the embodiment of the present disclosure is shown.
Step 210, the characteristic information that the target user clicks businessman within a preset period of time is obtained, and is based on the spy
Levy the click sequence information of target user described in information architecture.
It has been observed that being that the click sequence signature based on target user obtains each alternative businessman in the embodiments of the present disclosure
Recommendation score, therefore click sequence information acquisition it is most important.In the embodiments of the present disclosure, sequence letter is clicked in order to improve
The accuracy and completeness of breath can be clicked within a preset period of time the characteristic information of businessman with target user, and be based on the spy
Levy the click sequence information of target user described in information architecture.
Wherein, the particular content that characteristic information is included can be preset according to demand, be implemented to this disclosure
Example is not limited.For example, for taking out businessman and recommending scene, click businessman at this time can be to take out businessman, can be with
Characteristic information is set including but not limited to businessman feature, the vegetable feature, user businessman cross feature, target user for taking out businessman
User characteristics, in the user characteristics of other users, etc. accordingly taken out businessman and have consumer record.Businessman feature therein is again
It can include but is not limited to Merchant name, businessman's generic, merchant location, the sales volume of businessman, the evaluation of businessman, businessman
Per capita consuming level, etc.;Vegetable feature can include but is not limited to menu name, vegetable price, vegetable quantity, vegetable pin again
Amount, vegetable evaluation, etc.;User characteristics can include but is not limited to user's gender, occupation, age, consumption propensity, real-time position again
Set, etc.;User businessman cross feature then may include the operation note that user is directed to corresponding businessman, and corresponding businessman is directed to user
Feedback of operation, etc..
Moreover, putting in order for each characteristic information can also be set in advance according to demand in clicking sequence information
It sets, this embodiment of the present disclosure is not limited.In order to improve model accuracy, for same recommended models, it can be set
Putting in order for each characteristic information is consistent in corresponding click sequence information, may be set to be if necessary certainly
It is inconsistent, this embodiment of the present disclosure is not limited.
For example, it is assumed that 7 discrete features and 119 can be extracted for each click businessman for taking out in scene
Continuous feature, in addition it can extract to obtain each vegetable sequence for clicking businessman, if that sequence information is being clicked in setting
In putting in order as vegetable sequence+discrete features+continuous feature for each characteristic information for clicking businessman, then then can be with
Obtain click sequence information as shown in Figure 3.
Optionally, in the embodiments of the present disclosure, the click sequence information includes fisrt feature sequence and second feature sequence
Column;Include vegetable feature in the fisrt feature sequence, includes businessman feature, user characteristics and use in the second feature sequence
At least one of family businessman's cross feature.
Preset time period therein can be preset according to demand, be not limited to this embodiment of the present disclosure.
For example, can choose the characteristic information of the click businessman in different time sections according to different business demands, such as can choose and work as
It clicks the characteristic information of businessman, also can choose target user from all click businessmans' in the previous moon at current time
Characteristic information, etc..
In practical applications, under common electric business environment, the demand of only simple Recommendations.And scene is sold outside
Under, the description of businessman can directly be considered from businessman's angle, businessman can also be described from vegetable angle, so in disclosure reality
It applies in example, corresponding businessman is described based on vegetable feature for convenience, vegetable feature can be split from click sequence information
Individually learnt out.So can be set at this time and click sequence information includes fisrt feature sequence and second feature sequence,
And include vegetable feature in the fisrt feature sequence, include in the second feature sequence businessman feature, user characteristics and
At least one of user's businessman's cross feature.
Step 220, according to preset training sample data, joint training is by the double-deck Recognition with Recurrent Neural Network model and institute
State the recommended models that deep neural network model combines;Wherein, in each training process, the double-deck circulation nerve net
The input data of network model includes the sample sequence information in the training sample data, the deep neural network model it is defeated
Enter the output data that data include the double-deck Recognition with Recurrent Neural Network model, the corresponding sample businessman's of the sample sequence information
Characteristic sequence and the corresponding label value of the sample businessman.
In order to use recommended models to obtain the recommendation score of each alternative businessman, then need first to train corresponding recommendation
Model, specifically can be according to preset training sample data, and joint training is by the double-deck Recognition with Recurrent Neural Network model and institute
State the recommended models that deep neural network model combines.
It wherein, may include the sample sequence letter of a sample of users history click businessman in every training sample data
Breath, the characteristic sequence of the corresponding sample businessman of respective sample user and respective sample user are directed to the mark of respective sample businessman
Label value.The meaning that the label value indicates in different business scenarios is different, such as in prediction CTR (Click-Through-
Rate, click-through-rate) scene in the label value mean that click whether.In prediction CVR (Conversion Rate, conversion
Rate) label value means that whether buy in scene.In general, can be set if performing click or purchase operation
Corresponding label value is 1, and it is 0 that corresponding label value, which otherwise can be set,.Sample sequence information can then refer to above-mentioned click
Sequence information, not in this to go forth.
Moreover, needing to include positive and negative sample data in training sample data, every to improve the accuracy of recommended models
In Positive training sample data, respective sample businessman is the businessman that respective sample user finally selectes, then respective sample at this time
The label value of businessman can be 1;And in every negative training sample data, respective sample businessman can be any one corresponding sample
The businessman that this user does not select finally, then the label value of respective sample businessman can be 0 at this time.
So during carrying out joint training to recommended models using every training sample data, bilayer circulation nerve
The input data of network model includes the sample sequence information in the training sample data, the deep neural network model
Input data includes the output data of the double-deck Recognition with Recurrent Neural Network model, the corresponding sample businessman of the sample sequence information
Characteristic sequence and the sample businessman label value.
Step 230, by the click sequence information input double-deck Recognition with Recurrent Neural Network model, and the bilayer is obtained
The output vector of Recognition with Recurrent Neural Network model.
Optionally, in the embodiments of the present disclosure, the step 230 can further include:
Sub-step 231 by the first level Recognition with Recurrent Neural Network model described in the fisrt feature sequence inputting, and obtains institute
State the first output vector of the first level Recognition with Recurrent Neural Network model;
Second level described in first output vector and the second feature sequence inputting is recycled mind by sub-step 232
Through network model, and the second output vector of the second level Recognition with Recurrent Neural Network model is obtained, as the double-deck circulation
The output vector of neural network model.
It has been observed that under common electric business environment, only simple Recommendations demand, and sold under scene outside to businessman's
Description can directly consider from businessman's angle, can also describe businessman from vegetable angle, so in the embodiments of the present disclosure, design
For describing the RNN model of businessman's vegetable, therefore the double-deck Recognition with Recurrent Neural Network model in the embodiments of the present disclosure includes the
One level Recognition with Recurrent Neural Network model and the second level Recognition with Recurrent Neural Network model.
It so at this time then can be by fisrt feature sequence inputting the first level Recognition with Recurrent Neural Network mould including vegetable feature
Type, and obtain the first output vector of the first level Recognition with Recurrent Neural Network model, so by first output vector and
The second level of second feature sequence inputting Recognition with Recurrent Neural Network model, and obtain the second level Recognition with Recurrent Neural Network mould
Second output vector of type, the output vector as the double-deck Recognition with Recurrent Neural Network model.Moreover, by fisrt feature sequence
It, can be defeated by vegetable feature according to the clicking rate sequence from low to high of corresponding vegetable when being input in the first level RNN model
Enter into the first level RNN model, so that the high vegetable of clicking rate can more indicate the characteristic of corresponding businessman.It can certainly
Vegetable feature is inputted in other orders, can specifically be preset according to demand, this embodiment of the present disclosure is not added
To limit.
Optionally, in the embodiments of the present disclosure, the first level Recognition with Recurrent Neural Network model and second level are followed
Ring neural network model may each comprise but be not limited to dynamic shot and long term memory network model (bucketing LSTM/dynamic
LSTM).Moreover, if the first level Recognition with Recurrent Neural Network model and the second level Recognition with Recurrent Neural Network model are LSTM
Model, then bilayer Recognition with Recurrent Neural Network model above-mentioned is referred to as the double-deck dynamic LSTM model.
In practical applications, RNN model its be responsible for handling the sample sequence information in every training sample data, and sample
The length for the characteristic sequence for including in sequence information can change, then for general RNN model, input terminal
Mouth is fixed, if that the length of the corresponding characteristic sequence of sample sequence information and the input port length of RNN model are not
When matching, when especially the length of characteristic sequence is less than the input port length of RNN model, then it can be directed to corresponding characteristic sequence
Zero padding is carried out, had not only increased the occupancy of resource, but also be easy to make the precision of recommended models to reduce because introducing noise.
Therefore, in the embodiments of the present disclosure, the first level Recognition with Recurrent Neural Network model and second level can be set
Recognition with Recurrent Neural Network model includes dynamic shot and long term memory network model.The dynamic LSTM is can be according to list entries
Length adaptively adjust.For example, if input sequence length be 5, namely in sample sequence information comprising 5 click
The characteristic sequence of businessman, then the input port of dynamic LSTM model just becomes 5 sequences, with correspondingly received each feature sequence
Column;And if input is 10 sequence datas, dynamic LSTM if, can adaptively become 10.
Alternatively, the first level Recognition with Recurrent Neural Network model and second level circulation mind can also only be arranged according to demand
It is dynamic LSTM model through at least one of network model, this embodiment of the present invention is not limited.
Optionally, in the embodiments of the present disclosure, the sub-step 232, can further include:
Sub-step 2321, based on preset encoder matrix and first output vector and the second feature sequence
In coordinate of each discrete features relative to the encoder matrix, obtain the discrete features relative to the encoder matrix
First insertion vector;
Sub-step 2322, according to the first of each offline feature the insertion vector and first output vector and described
Continuous feature in second feature sequence, obtain the second of first output vector and the second feature sequence be embedded in
Amount;
The second insertion vector is inputted the second level Recognition with Recurrent Neural Network model, and obtained by sub-step 2323
Second output vector of the second level Recognition with Recurrent Neural Network model, the output as the double-deck Recognition with Recurrent Neural Network model
Vector.
It has been observed that the second level RNN model is also in order to handle outside the other feature other than vegetable characteristic sequence
Need support user characteristics, user's businessman's cross feature, the input of businessman feature, so for the defeated of the second level RNN model
Enter data to be converted, makes it that can meet real business demand.
It specifically can be based in preset encoder matrix and first output vector and the second feature sequence
Coordinate of each discrete features relative to the encoder matrix, obtain of the discrete features relative to the encoder matrix
One insertion vector, and then according to the first of each offline feature the insertion vector and first output vector and described second
Continuous feature in characteristic sequence obtains the second insertion vector of first output vector and the second feature sequence, most
The second insertion vector is inputted into the second level Recognition with Recurrent Neural Network model afterwards, and obtains the second level circulation nerve net
Second output vector of network model, the output vector as the double-deck Recognition with Recurrent Neural Network model.
Encoder matrix therein can be preset according to demand, be not limited to this embodiment of the present disclosure, and
And each discrete features can also be preset according to demand relative to the coordinate of the encoder matrix, to this disclosure reality
Example is applied also to be not limited.
For example, if a certain discrete features correspond to the i-th row jth column in encoder matrix, then available coding
The i-th row vector in matrix, and then the row vector gone out for various discrete feature extraction is attached (concat) either
Be averaged the operation such as (average), thus obtain obtaining the discrete features relative to the first of the encoder matrix be embedded in
Amount.Certainly in the embodiments of the present disclosure, if a certain discrete features correspond to the i-th row jth column in encoder matrix, can also set
Setting the obtained in encoder matrix can specifically be set with regard to j column vector with setting up the first insertion vector in advance according to demand
It sets, this embodiment of the present disclosure is not limited.
Further, according to the first of each offline feature the insertion vector and first output vector and described the
Continuous feature in two characteristic sequences obtains the second insertion vector of first output vector and the second feature sequence.
It specifically can be by the first insertion vector connection corresponding with discrete features of each continuous feature, to obtain the first output vector
Vector is embedded in the second of the second feature sequence.Certainly, according to the first of each offline feature the insertion vector and described
Continuous feature in first output vector and the second feature sequence, the concrete mode for obtaining the second insertion vector can also root
It is preset according to demand, this embodiment of the present disclosure is not limited.
Alternatively, in the embodiments of the present disclosure, can also just for each discrete features in second feature sequence relative to
The coordinate of the encoder matrix obtains first insertion vector of the discrete features relative to the encoder matrix;And then basis
First insertion vector of each offline feature and continuous feature in the second feature sequence and described first export to
Amount obtains the second insertion vector of first output vector and the second feature sequence.It at this time only need to be for for second
Coordinate of each discrete features relative to the encoder matrix in characteristic sequence, obtains the discrete features relative to the volume
First insertion vector of code matrix, without being operated to the first output vector.
Step 240, according to the output vector and the characteristic information of the alternative businessman, pass through the depth nerve net
Network model obtains the recommendation score of each alternative businessman.
Step 250, the recommendation score based on the alternative businessman, the target for obtaining the target user are recommended businessman and are pushed away
It send to the target user.
In the embodiments of the present disclosure, pass through the click sequence information of acquisition target user;According to the click sequence information,
By preset recommended models, the recommendation score of each alternative businessman is obtained;Based on the recommendation score of the alternative businessman, obtain
The target of the target user recommends businessman and pushes to the target user;Wherein, the recommended models are by bilayer
Recognition with Recurrent Neural Network model and deep neural network model carry out the built-up pattern obtained after joint training, the double-deck circulation mind
Include the first level Recognition with Recurrent Neural Network model and the second level Recognition with Recurrent Neural Network model, and the second layer through network model
The input data of grade Recognition with Recurrent Neural Network model includes the output data of the first level Recognition with Recurrent Neural Network model.It achieves
Improve the beneficial effect for recommending businessman's accuracy.
Moreover, in the embodiments of the present disclosure, the target user can also be obtained and click businessman's within a preset period of time
Characteristic information, and construct based on the characteristic information click sequence information of the target user.And by the click sequence
Bilayer Recognition with Recurrent Neural Network model described in information input, and obtain the output vector of the double-deck Recognition with Recurrent Neural Network model;Root
Each institute is obtained by the deep neural network model according to the output vector and the characteristic information of the alternative businessman
State the recommendation score of alternative businessman.The click sequence information includes fisrt feature sequence and second feature sequence;Described first
Include vegetable feature in characteristic sequence, includes that businessman feature, user characteristics and user businessman are intersected in the second feature sequence
At least one of feature.Moreover, by the first level of fisrt feature sequence inputting Recognition with Recurrent Neural Network model, and obtain institute
State the first output vector of the first level Recognition with Recurrent Neural Network model;By first output vector and the second feature sequence
Input the second level Recognition with Recurrent Neural Network model, and obtain the second of the second level Recognition with Recurrent Neural Network model export to
Amount, the output vector as the double-deck Recognition with Recurrent Neural Network model.Based on preset encoder matrix and first output
Coordinate of each discrete features relative to the encoder matrix in second feature sequence described in vector sum obtains the discrete spy
It levies and is embedded in vector relative to the first of the encoder matrix;According to the first of each offline feature the insertion vector and described the
Continuous feature in one output vector and the second feature sequence obtains first output vector and the second feature sequence
Second insertion vector of column;The second insertion vector is inputted into the second level Recognition with Recurrent Neural Network model, and obtains described the
Second output vector of two level Recognition with Recurrent Neural Network models, the output vector as the double-deck Recognition with Recurrent Neural Network model.
So as to further increase the accuracy of recommendation results.
In addition, in the embodiments of the present disclosure, according to preset training sample data, joint training is by the double-deck circulation mind
The recommended models combined through network model and the deep neural network model;Wherein, described in each training process
The input data of the double-deck Recognition with Recurrent Neural Network model includes the sample sequence information in the training sample data, the depth mind
Input data through network model includes the output data of the double-deck Recognition with Recurrent Neural Network model, the sample sequence information pair
The characteristic sequence of the sample businessman answered and the label value of the sample businessman.The first circulation neural network model and institute
Stating second circulation neural network model includes dynamic shot and long term memory network model.It equally can be further improved recommended models
Accuracy.
For embodiment of the method, for simple description, therefore, it is stated as a series of action combinations, but this field
Technical staff should be aware of, and the embodiment of the present disclosure is not limited by the described action sequence, because implementing according to the disclosure
Example, some steps may be performed in other sequences or simultaneously.Secondly, those skilled in the art should also know that, specification
Described in embodiment belong to preferred embodiment, necessary to the related movement not necessarily embodiment of the present disclosure.
Embodiment three
A kind of businessman's recommendation apparatus of embodiment of the present disclosure offer is provided.
Referring to Fig. 4, a kind of structural schematic diagram of businessman's recommendation apparatus in the embodiment of the present disclosure is shown.
It clicks sequence information and obtains module 310, for obtaining the click sequence information of target user.
Recommendation score obtains module 320, for being obtained according to the click sequence information by preset recommended models
The recommendation score of each alternative businessman;Wherein, the recommended models are by the double-deck Recognition with Recurrent Neural Network model and depth mind
The built-up pattern obtained after network model carries out joint training, the bilayer Recognition with Recurrent Neural Network model include that the first level is followed
Ring neural network model and the second level Recognition with Recurrent Neural Network model, and the input of the second level Recognition with Recurrent Neural Network model
Data include the output data of the first level Recognition with Recurrent Neural Network model.
Target recommends businessman to obtain module 330, for the recommendation score based on the alternative businessman, obtains the target and uses
The target at family recommends businessman and pushes to the target user.
In the embodiments of the present disclosure, pass through the click sequence information of acquisition target user;According to the click sequence information,
By preset recommended models, the recommendation score of each alternative businessman is obtained;Based on the recommendation score of the alternative businessman, obtain
The target of the target user recommends businessman and pushes to the target user;Wherein, the recommended models are by bilayer
Recognition with Recurrent Neural Network model and deep neural network model carry out the built-up pattern obtained after joint training, the double-deck circulation mind
Include the first level Recognition with Recurrent Neural Network model and the second level Recognition with Recurrent Neural Network model, and the second layer through network model
The input data of grade Recognition with Recurrent Neural Network model includes the output data of the first level Recognition with Recurrent Neural Network model.It achieves
Improve the beneficial effect for recommending businessman's accuracy.
Example IV
A kind of businessman's recommendation apparatus of embodiment of the present disclosure offer is provided.
Referring to Fig. 5, a kind of structural schematic diagram of businessman's recommendation apparatus in the embodiment of the present disclosure is shown.
It clicks sequence information and obtains module 410, for obtaining the click sequence information of target user.
Optionally, in the embodiments of the present disclosure, the click sequence information obtains module 410, can further include:
Sequence information acquisition submodule 411 is clicked, clicks businessman within a preset period of time for obtaining the target user
Characteristic information, and construct based on the characteristic information click sequence information of the target user.
Recommended models training module 420, for according to preset training sample data, joint training to be by the double-deck circulation
The recommended models that neural network model and the deep neural network model combine;Wherein, in each training process, institute
The input data for stating the double-deck Recognition with Recurrent Neural Network model includes sample sequence information in the training sample data, the depth
The input data of neural network model includes the output data of the double-deck Recognition with Recurrent Neural Network model, the sample sequence information
The characteristic sequence of corresponding sample businessman and the label value of the sample businessman.
Recommendation score obtains module 430, for being obtained according to the click sequence information by preset recommended models
The recommendation score of each alternative businessman;Wherein, the recommended models are by the double-deck Recognition with Recurrent Neural Network model and depth mind
The built-up pattern obtained after network model carries out joint training, the bilayer Recognition with Recurrent Neural Network model include that the first level is followed
Ring neural network model and the second level Recognition with Recurrent Neural Network model, and the input of the second level Recognition with Recurrent Neural Network model
Data include the output data of the first level Recognition with Recurrent Neural Network model.
Optionally, in the embodiments of the present disclosure, the click sequence information includes fisrt feature sequence and second feature sequence
Column;Include vegetable feature in the fisrt feature sequence, includes businessman feature, user characteristics and use in the second feature sequence
At least one of family businessman's cross feature.
Optionally, in the embodiments of the present disclosure, the recommendation score obtains module 430, can further include:
First input submodule 431, for the click sequence information to be inputted the double-deck Recognition with Recurrent Neural Network model,
And obtain the output vector of the double-deck Recognition with Recurrent Neural Network model;
Optionally, in the embodiments of the present disclosure, first input submodule 431, can further include:
Fisrt feature sequence inputting unit, for the first level described in the fisrt feature sequence inputting to be recycled nerve net
Network model, and obtain the first output vector of the first level Recognition with Recurrent Neural Network model;
Second feature sequence inputting unit, being used for will be described in first output vector and the second feature sequence inputting
Second level Recognition with Recurrent Neural Network model, and the second output vector of the second level Recognition with Recurrent Neural Network model is obtained, make
For the output vector of the double-deck Recognition with Recurrent Neural Network model.
Optionally, in the embodiments of the present disclosure, the first level Recognition with Recurrent Neural Network model and second level are followed
Ring neural network model includes dynamic shot and long term memory network model.
Optionally, in the embodiments of the present disclosure, the second feature sequence inputting unit, can further include:
First insertion vector obtains subelement, for based on preset encoder matrix and first output vector and
Coordinate of each discrete features relative to the encoder matrix in the second feature sequence, it is opposite to obtain the discrete features
First in the encoder matrix is embedded in vector;
Second insertion vector obtains subelement, for according to the first of each offline feature the insertion vector and described the
Continuous feature in one output vector and the second feature sequence obtains first output vector and the second feature sequence
Second insertion vector of column;
Output vector obtains subelement, for the second insertion vector to be inputted the second level Recognition with Recurrent Neural Network mould
Type, and the second output vector of the second level Recognition with Recurrent Neural Network model is obtained, as the double-deck Recognition with Recurrent Neural Network
The output vector of model.
Recommendation score acquisition submodule 432, for the characteristic information according to the output vector and the alternative businessman,
By the deep neural network model, the recommendation score of each alternative businessman is obtained.
Target recommends businessman to obtain module 440, for the recommendation score based on the alternative businessman, obtains the target and uses
The target at family recommends businessman and pushes to the target user.
In the embodiments of the present disclosure, pass through the click sequence information of acquisition target user;According to the click sequence information,
By preset recommended models, the recommendation score of each alternative businessman is obtained;Based on the recommendation score of the alternative businessman, obtain
The target of the target user recommends businessman and pushes to the target user;Wherein, the recommended models are by bilayer
Recognition with Recurrent Neural Network model and deep neural network model carry out the built-up pattern obtained after joint training, the double-deck circulation mind
Include the first level Recognition with Recurrent Neural Network model and the second level Recognition with Recurrent Neural Network model, and the second layer through network model
The input data of grade Recognition with Recurrent Neural Network model includes the output data of the first level Recognition with Recurrent Neural Network model.It achieves
Improve the beneficial effect for recommending businessman's accuracy.
Moreover, in the embodiments of the present disclosure, the target user can also be obtained and click businessman's within a preset period of time
Characteristic information, and construct based on the characteristic information click sequence information of the target user.And by the click sequence
Bilayer Recognition with Recurrent Neural Network model described in information input, and obtain the output vector of the double-deck Recognition with Recurrent Neural Network model;Root
Each institute is obtained by the deep neural network model according to the output vector and the characteristic information of the alternative businessman
State the recommendation score of alternative businessman.The click sequence information includes fisrt feature sequence and second feature sequence;Described first
Include vegetable feature in characteristic sequence, includes that businessman feature, user characteristics and user businessman are intersected in the second feature sequence
At least one of feature.Moreover, by the first level of fisrt feature sequence inputting Recognition with Recurrent Neural Network model, and obtain institute
State the first output vector of the first level Recognition with Recurrent Neural Network model;By first output vector and the second feature sequence
Input the second level Recognition with Recurrent Neural Network model, and obtain the second of the second level Recognition with Recurrent Neural Network model export to
Amount, the output vector as the double-deck Recognition with Recurrent Neural Network model.Based on preset encoder matrix and first output
Coordinate of each discrete features relative to the encoder matrix in second feature sequence described in vector sum obtains the discrete spy
It levies and is embedded in vector relative to the first of the encoder matrix;According to the first of each offline feature the insertion vector and described the
Continuous feature in one output vector and the second feature sequence obtains first output vector and the second feature sequence
Second insertion vector of column;The second insertion vector is inputted into the second level Recognition with Recurrent Neural Network model, and obtains described the
Second output vector of two level Recognition with Recurrent Neural Network models, the output vector as the double-deck Recognition with Recurrent Neural Network model.
So as to further increase the accuracy of recommendation results.
In addition, in the embodiments of the present disclosure, according to preset training sample data, joint training is by the double-deck circulation mind
The recommended models combined through network model and the deep neural network model;Wherein, described in each training process
The input data of the double-deck Recognition with Recurrent Neural Network model includes the sample sequence information in the training sample data, the depth mind
Input data through network model includes the output data of the double-deck Recognition with Recurrent Neural Network model, the sample sequence information pair
The characteristic sequence of the sample businessman answered and the label value of the sample businessman.It is described bilayer Recognition with Recurrent Neural Network model include
Dynamic shot and long term memory network model.It equally can be further improved the accuracy of recommended models.
A kind of electronic equipment is also disclosed in the embodiment of the present disclosure, comprising:
Processor, memory and it is stored in the computer journey that can be run on the memory and on the processor
Sequence, which is characterized in that the processor realizes any one merchant recommendation method above-mentioned when executing described program.
A kind of readable storage medium storing program for executing is also disclosed in the embodiment of the present disclosure, when the instruction in the storage medium is set by electronics
When standby processor executes, so that electronic equipment is able to carry out any one merchant recommendation method above-mentioned.
For device embodiment, since it is basically similar to the method embodiment, related so being described relatively simple
Place illustrates referring to the part of embodiment of the method.
Algorithm and display are not inherently related to any particular computer, virtual system, or other device provided herein.
Various general-purpose systems can also be used together with teachings based herein.As described above, it constructs required by this kind of system
Structure be obvious.In addition, the disclosure is also not for any particular programming language.It should be understood that can use various
Programming language realizes content of this disclosure described herein, and the description done above to language-specific is to disclose this public affairs
The preferred forms opened.
In the instructions provided here, numerous specific details are set forth.It is to be appreciated, however, that the implementation of the disclosure
Example can be practiced without these specific details.In some instances, well known method, structure is not been shown in detail
And technology, so as not to obscure the understanding of this specification.
Similarly, it should be understood that in order to simplify the disclosure and help to understand one or more of the various inventive aspects,
Above in the description of the exemplary embodiment of the disclosure, each feature of the disclosure is grouped together into single implementation sometimes
In example, figure or descriptions thereof.However, the disclosed method should not be interpreted as reflecting the following intention: i.e. required to protect
The disclosure of shield requires features more more than feature expressly recited in each claim.More precisely, as following
Claims reflect as, inventive aspect is all features less than single embodiment disclosed above.Therefore,
Thus the claims for following specific embodiment are expressly incorporated in the specific embodiment, wherein each claim itself
All as the separate embodiments of the disclosure.
Those skilled in the art will understand that can be carried out adaptively to the module in the equipment in embodiment
Change and they are arranged in one or more devices different from this embodiment.It can be the module or list in embodiment
Member or component are combined into a module or unit or component, and furthermore they can be divided into multiple submodule or subelement or
Sub-component.Other than such feature and/or at least some of process or unit exclude each other, it can use any
Combination is to all features disclosed in this specification (including adjoint claim, abstract and attached drawing) and so disclosed
All process or units of what method or apparatus are combined.Unless expressly stated otherwise, this specification is (including adjoint power
Benefit require, abstract and attached drawing) disclosed in each feature can carry out generation with an alternative feature that provides the same, equivalent, or similar purpose
It replaces.
In addition, it will be appreciated by those of skill in the art that although some embodiments described herein include other embodiments
In included certain features rather than other feature, but the combination of the feature of different embodiments means to be in the disclosure
Within the scope of and form different embodiments.For example, in the following claims, embodiment claimed is appointed
Meaning one of can in any combination mode come using.
The various component embodiments of the disclosure can be implemented in hardware, or to run on one or more processors
Software module realize, or be implemented in a combination thereof.It will be understood by those of skill in the art that can be used in practice
Microprocessor or digital signal processor (DSP) come realize some in businessman's recommendation apparatus according to the embodiment of the present disclosure or
The some or all functions of person's whole component.The disclosure is also implemented as one for executing method as described herein
Point or whole device or device programs (for example, computer program and computer program product).Such this public affairs of realization
The program opened can store on a computer-readable medium, or may be in the form of one or more signals.It is such
Signal can be downloaded from an internet website to obtain, and is perhaps provided on the carrier signal or is provided in any other form.
The disclosure is limited it should be noted that above-described embodiment illustrates rather than the disclosure, and ability
Field technique personnel can be designed alternative embodiment without departing from the scope of the appended claims.In the claims,
Any reference symbol between parentheses should not be configured to limitations on claims.Word "comprising" does not exclude the presence of not
Element or step listed in the claims.Word "a" or "an" located in front of the element does not exclude the presence of multiple such
Element.The disclosure can be by means of including the hardware of several different elements and being come by means of properly programmed computer real
It is existing.In the unit claims listing several devices, several in these devices can be by the same hardware branch come
It embodies.The use of word first, second, and third does not indicate any sequence.These words can be construed to title.
Claims (11)
1. a kind of merchant recommendation method characterized by comprising
Obtain the click sequence information of target user;
The recommendation score of each alternative businessman is obtained by preset recommended models according to the click sequence information;
Based on the recommendation score of the alternative businessman, the target for obtaining the target user recommends businessman and pushes to the target
User;
Wherein, the recommended models are by carrying out joint instruction to the double-deck Recognition with Recurrent Neural Network model and deep neural network model
The built-up pattern obtained after white silk, the bilayer Recognition with Recurrent Neural Network model includes the first level Recognition with Recurrent Neural Network model and second
Level Recognition with Recurrent Neural Network model, and the input data of the second level Recognition with Recurrent Neural Network model includes first level
The output data of Recognition with Recurrent Neural Network model.
2. the method according to claim 1, wherein described according to the click sequence information, by preset
The step of recommended models, the recommendation score of each alternative businessman of acquisition, comprising:
By the click sequence information input double-deck Recognition with Recurrent Neural Network model, and obtain the double-deck Recognition with Recurrent Neural Network
The output vector of model;
It is obtained according to the output vector and the characteristic information of the alternative businessman by the deep neural network model
The recommendation score of each alternative businessman.
3. method according to claim 1, which is characterized in that the click sequence information includes fisrt feature sequence and second
Characteristic sequence;Include vegetable feature in the fisrt feature sequence, includes businessman feature, Yong Hute in the second feature sequence
It seeks peace at least one of user's businessman's cross feature.
4. according to the method described in claim 3, it is characterized in that, described follow the click sequence information input bilayer
Ring neural network model, and the step of obtaining the output vector of the double-deck Recognition with Recurrent Neural Network model, comprising:
By the first level Recognition with Recurrent Neural Network model described in the fisrt feature sequence inputting, and obtain the first level circulation
First output vector of neural network model;
By the second level Recognition with Recurrent Neural Network model described in first output vector and the second feature sequence inputting, and obtain
The second output vector for taking the second level Recognition with Recurrent Neural Network model, as the defeated of the double-deck Recognition with Recurrent Neural Network model
Outgoing vector.
5. according to the method described in claim 4, it is characterized in that, described by first output vector and the second feature
Second level Recognition with Recurrent Neural Network model described in sequence inputting, and obtain the second of the second level Recognition with Recurrent Neural Network model
The step of output vector, output vector as the double-deck Recognition with Recurrent Neural Network model, comprising:
Based on each discrete features in preset encoder matrix and first output vector and the second feature sequence
Relative to the coordinate of the encoder matrix, first insertion vector of the discrete features relative to the encoder matrix is obtained;
According in the first of each offline feature the insertion vector and first output vector and the second feature sequence
Continuous feature obtains the second insertion vector of first output vector and the second feature sequence;
The second insertion vector is inputted into the second level Recognition with Recurrent Neural Network model, and obtains the second level circulation
Second output vector of neural network model, the output vector as the double-deck Recognition with Recurrent Neural Network model.
6. method according to any one of claims 1-5, which is characterized in that the first level Recognition with Recurrent Neural Network mould
Type and the second level Recognition with Recurrent Neural Network model include dynamic shot and long term memory network model.
7. method according to any one of claims 1-5, which is characterized in that the click sequence for obtaining target user
The step of information, comprising:
The characteristic information that the target user clicks businessman within a preset period of time is obtained, and institute is constructed based on the characteristic information
State the click sequence information of target user.
8. method according to any one of claims 1-5, which is characterized in that believed described according to the click sequence
Breath, by preset recommended models, before the step of obtaining the recommendation score of each alternative businessman, further includes:
According to preset training sample data, joint training is by the double-deck Recognition with Recurrent Neural Network model and the depth nerve net
The recommended models that network model combines;
Wherein, in each training process, the input data of the bilayer Recognition with Recurrent Neural Network model includes the training sample
Sample sequence information in data, the input data of the deep neural network model include the double-deck Recognition with Recurrent Neural Network mould
The output data of type, the characteristic sequence of the corresponding sample businessman of the sample sequence information and the sample businessman are corresponding
Label value.
9. a kind of businessman's recommendation apparatus characterized by comprising
It clicks sequence information and obtains module, for obtaining the click sequence information of target user;
Recommendation score obtains module, for being obtained each alternative according to the click sequence information by preset recommended models
The recommendation score of businessman;
Target recommends businessman's acquisition module to obtain the mesh of the target user for the recommendation score based on the alternative businessman
Mark recommends businessman and pushes to the target user;
Wherein, the recommended models are by carrying out joint instruction to the double-deck Recognition with Recurrent Neural Network model and deep neural network model
The built-up pattern obtained after white silk, the bilayer Recognition with Recurrent Neural Network model includes the first level Recognition with Recurrent Neural Network model and second
Level Recognition with Recurrent Neural Network model, and the input data of the second level Recognition with Recurrent Neural Network model includes first level
The output data of Recognition with Recurrent Neural Network model.
10. a kind of electronic equipment characterized by comprising
Processor, memory and it is stored in the computer program that can be run on the memory and on the processor,
It is characterized in that, the processor realizes the businessman as described in any one of claim 1-8 when executing the computer program
Recommended method.
11. a kind of readable storage medium storing program for executing, which is characterized in that when the instruction in the storage medium is held by the processor of electronic equipment
When row, so that electronic equipment is able to carry out the merchant recommendation method as described in any one of claim 1-8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910324959.3A CN110209926A (en) | 2019-04-22 | 2019-04-22 | Merchant recommendation method, device, electronic equipment and readable storage medium storing program for executing |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910324959.3A CN110209926A (en) | 2019-04-22 | 2019-04-22 | Merchant recommendation method, device, electronic equipment and readable storage medium storing program for executing |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110209926A true CN110209926A (en) | 2019-09-06 |
Family
ID=67786116
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910324959.3A Pending CN110209926A (en) | 2019-04-22 | 2019-04-22 | Merchant recommendation method, device, electronic equipment and readable storage medium storing program for executing |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110209926A (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110992076A (en) * | 2019-11-01 | 2020-04-10 | 北京三快在线科技有限公司 | Merchant quality evaluation method and device, electronic equipment and readable storage medium |
CN111538907A (en) * | 2020-06-05 | 2020-08-14 | 支付宝(杭州)信息技术有限公司 | Object recommendation method, system and device |
CN111831630A (en) * | 2019-12-30 | 2020-10-27 | 北京骑胜科技有限公司 | Data processing method and device, electronic equipment and computer readable storage medium |
CN112016999A (en) * | 2020-08-28 | 2020-12-01 | 北京百度网讯科技有限公司 | Method, apparatus, device and storage medium for processing data |
CN112287238A (en) * | 2020-12-30 | 2021-01-29 | 腾讯科技(深圳)有限公司 | User characteristic determination method and device, storage medium and electronic equipment |
CN112464094A (en) * | 2020-11-30 | 2021-03-09 | 泰康保险集团股份有限公司 | Information recommendation method and device, electronic equipment and storage medium |
CN112633978A (en) * | 2020-12-22 | 2021-04-09 | 重庆大学 | Graph neural network model construction method, and method, device and equipment for commodity recommendation |
CN112801744A (en) * | 2020-12-01 | 2021-05-14 | 珠海必要工业科技股份有限公司 | Activity recommendation method and device, electronic equipment and storage medium |
CN113129053A (en) * | 2021-03-29 | 2021-07-16 | 北京沃东天骏信息技术有限公司 | Information recommendation model training method, information recommendation method and storage medium |
CN113362218A (en) * | 2021-05-21 | 2021-09-07 | 北京百度网讯科技有限公司 | Data processing method and device, electronic equipment and storage medium |
CN114118882A (en) * | 2022-01-27 | 2022-03-01 | 太平金融科技服务(上海)有限公司 | Service data processing method, device, equipment and medium based on combined model |
-
2019
- 2019-04-22 CN CN201910324959.3A patent/CN110209926A/en active Pending
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110992076A (en) * | 2019-11-01 | 2020-04-10 | 北京三快在线科技有限公司 | Merchant quality evaluation method and device, electronic equipment and readable storage medium |
CN111831630A (en) * | 2019-12-30 | 2020-10-27 | 北京骑胜科技有限公司 | Data processing method and device, electronic equipment and computer readable storage medium |
CN111831630B (en) * | 2019-12-30 | 2023-12-05 | 北京骑胜科技有限公司 | Data processing method, device, electronic equipment and computer readable storage medium |
CN111538907A (en) * | 2020-06-05 | 2020-08-14 | 支付宝(杭州)信息技术有限公司 | Object recommendation method, system and device |
CN111538907B (en) * | 2020-06-05 | 2023-04-18 | 支付宝(杭州)信息技术有限公司 | Object recommendation method, system and device |
CN112016999A (en) * | 2020-08-28 | 2020-12-01 | 北京百度网讯科技有限公司 | Method, apparatus, device and storage medium for processing data |
CN112464094B (en) * | 2020-11-30 | 2024-03-15 | 泰康保险集团股份有限公司 | Information recommendation method and device, electronic equipment and storage medium |
CN112464094A (en) * | 2020-11-30 | 2021-03-09 | 泰康保险集团股份有限公司 | Information recommendation method and device, electronic equipment and storage medium |
CN112801744B (en) * | 2020-12-01 | 2022-04-29 | 珠海必要工业科技股份有限公司 | Activity recommendation method and device, electronic equipment and storage medium |
CN112801744A (en) * | 2020-12-01 | 2021-05-14 | 珠海必要工业科技股份有限公司 | Activity recommendation method and device, electronic equipment and storage medium |
CN112633978A (en) * | 2020-12-22 | 2021-04-09 | 重庆大学 | Graph neural network model construction method, and method, device and equipment for commodity recommendation |
CN112633978B (en) * | 2020-12-22 | 2024-03-08 | 重庆大学 | Method, device and equipment for building graphic neural network model and method, device and equipment for recommending commodities |
CN112287238B (en) * | 2020-12-30 | 2021-04-09 | 腾讯科技(深圳)有限公司 | User characteristic determination method and device, storage medium and electronic equipment |
CN112287238A (en) * | 2020-12-30 | 2021-01-29 | 腾讯科技(深圳)有限公司 | User characteristic determination method and device, storage medium and electronic equipment |
CN113129053A (en) * | 2021-03-29 | 2021-07-16 | 北京沃东天骏信息技术有限公司 | Information recommendation model training method, information recommendation method and storage medium |
CN113129053B (en) * | 2021-03-29 | 2024-05-21 | 北京沃东天骏信息技术有限公司 | Information recommendation model training method, information recommendation method and storage medium |
CN113362218A (en) * | 2021-05-21 | 2021-09-07 | 北京百度网讯科技有限公司 | Data processing method and device, electronic equipment and storage medium |
CN114118882A (en) * | 2022-01-27 | 2022-03-01 | 太平金融科技服务(上海)有限公司 | Service data processing method, device, equipment and medium based on combined model |
CN114118882B (en) * | 2022-01-27 | 2022-05-27 | 太平金融科技服务(上海)有限公司 | Service data processing method, device, equipment and medium based on combined model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110209926A (en) | Merchant recommendation method, device, electronic equipment and readable storage medium storing program for executing | |
CN108446374B (en) | User's Intention Anticipation method, apparatus, electronic equipment, storage medium | |
CN110457452A (en) | Rationale for the recommendation generation method, device, electronic equipment and readable storage medium storing program for executing | |
CN110909862A (en) | Attention weight calculation method and device based on convolutional neural network | |
CN109785968A (en) | A kind of event prediction method, apparatus, equipment and program product | |
CN109446341A (en) | The construction method and device of knowledge mapping | |
CN111784455A (en) | Article recommendation method and recommendation equipment | |
CN113688304A (en) | Training method for search recommendation model, and method and device for sequencing search results | |
CN110019793A (en) | A kind of text semantic coding method and device | |
CN107861938A (en) | A kind of POI official documents and correspondences generation method and device, electronic equipment | |
CN110765273B (en) | Recommended document generation method and device, electronic equipment and readable storage medium | |
CN102937976B (en) | A kind of drop-down reminding method based on input prefix and device | |
US20190286978A1 (en) | Using natural language processing and deep learning for mapping any schema data to a hierarchical standard data model (xdm) | |
CN108062303A (en) | The recognition methods of refuse messages and device | |
Shen et al. | A voice of the customer real-time strategy: An integrated quality function deployment approach | |
CN109063105A (en) | File memory method, device, computer equipment and storage medium | |
CN109918630A (en) | Document creation method, device, computer equipment and storage medium | |
US20220414470A1 (en) | Multi-Task Attention Based Recurrent Neural Networks for Efficient Representation Learning | |
CN109376219A (en) | Matching process, device, electronic equipment and the storage medium of text attributes field | |
Deng et al. | Prediction of aqueous solubility of compounds based on neural network | |
CN110427564A (en) | Deep learning is integrated into broad sense additivity melange effect (GAME) frame | |
CN112529151A (en) | Data processing method and device | |
CN117009650A (en) | Recommendation method and device | |
CN106997360A (en) | The treating method and apparatus of user behavior data | |
CN108876422A (en) | For the method, apparatus of information popularization, electronic equipment and computer-readable medium |
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 |