CN107609060A - Resource recommendation method and device - Google Patents
Resource recommendation method and device Download PDFInfo
- Publication number
- CN107609060A CN107609060A CN201710750386.1A CN201710750386A CN107609060A CN 107609060 A CN107609060 A CN 107609060A CN 201710750386 A CN201710750386 A CN 201710750386A CN 107609060 A CN107609060 A CN 107609060A
- Authority
- CN
- China
- Prior art keywords
- resource
- customer resources
- training sample
- sale
- classification results
- 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
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention, which provides a kind of resource recommendation method and device, this method, to be included：Sale resource training sample and customer resources training sample input resource classification model are trained respectively, obtain sale resource classification results and customer resources classification results；Sale resource classification results and customer resources classification results input similarity model are trained, export some matching results as recommendation.On the one hand resource recommendation method and device provided by the invention model training by machine learning and carry out intelligent classification to sale resource and customer resources, and intelligent Matching is carried out to classification results, solve the problems, such as cold start-up, realize and more accurately recommend compared to existing scheme；On the other hand by excavating the intelligent classification of customer resources the largely data message on client or potential customers, complete to create data assets while CRM precisely recommends.
Description
Technical field
The application is related to customer relation management technical field, and in particular to a kind of resource recommendation method and device.
Background technology
With the continuous expansion of big data, mobile Internet epoch data message scale, people are gradually from absence of information
Epoch have entered into information overload (information overload) epoch.In this epoch, customer relation management
The species and number rapid growth of customer resources, client in (Customer relationship management, abbreviation CRM)
Or the colony of potential customers is very huge, it could be that client matches required service that sale, which requires a great deal of time, visitor
Family does not know how to find and excavate own resource advantage again.Therefore, how effectively to manage, excavate the CRM customer relationships of complexity
Resource, intelligent classification matching is carried out to related resource, be finally reached resource most reasonable distribution and utilize, be that Customer mining is maximally effective
Marketing activity strategy turns into current enterprise question of common concern with enterprise income.
CRM resource recommendation methods mainly have following two at present：
One kind is content-based recommendation method：It is based primarily upon commodity transaction total value (the Gross Merchandise of client
Volume, abbreviation GMV), brand recognition analyzed, for example client GMV is high, brand recognition is high, can be preferentially to VIP gold medals
Similar client is recommended in sale；
Another kind is the recommendation method based on collaborative filtering：It is based primarily upon the historical data of CRM resources, off-line calculation client
Similarity, sale resource similarity.
Following defect be present in above two method：On the one hand, cold start-up be present, can not be to new customer resources and pin
Resource is sold precisely to be recommended；On the other hand, do not account for customer resources and sale resource be characterized in continually changing,
Client feedback or behavior can not be tracked in time to assess recommendation effect, lead to not the matching for optimizing resource, related advantages resource
It can not match in time in client's hand, enterprise can not accurately provide the user service high-qualityly.
The content of the invention
In view of drawbacks described above of the prior art or deficiency, it is expected that cold start-up can be solved the problems, such as by providing one kind, to new visitor
The resource recommendation method and device that family resource and sale resource are precisely recommended；And further it is expected that by tracking feedback optimized
The matching of recommendation.
In a first aspect, the present invention provides a kind of resource recommendation method, including：
Sale resource training sample and customer resources training sample input resource classification model are trained respectively, obtained
Sale resource classification results and customer resources classification results；
Sale resource classification results and customer resources classification results input similarity model are trained, export some
With result as recommendation.
Second aspect, the present invention provide a kind of resource recommendation device, including taxon and matching unit.
Taxon is configured to that sale resource training sample and customer resources training sample are inputted into resource classification respectively
Model is trained, and obtains sale resource classification results and customer resources classification results；
Matching unit is configured to enter sale resource classification results and customer resources classification results input similarity model
Row training, some matching results are exported as recommendation.
The third aspect, the present invention also provide a kind of equipment, including one or more processors and memory, wherein memory
Comprising by instruction that the one or more processors perform the one or more processors being caused to perform according to of the invention each
The resource recommendation method that embodiment provides.
Fourth aspect, the present invention also provide a kind of computer-readable recording medium for being stored with computer program, the calculating
Machine program makes computer perform the resource recommendation method provided according to various embodiments of the present invention.
On the one hand the resource recommendation method and device that many embodiments of the present invention provide pass through machine learning modeling training pair
Sale resource and customer resources carry out intelligent classification, and carry out intelligent Matching to classification results, solve the problems, such as cold start-up, real
Show and more accurately recommended compared to existing scheme；On the other hand by excavating a large amount of passes to the intelligent classification of customer resources
In the data message of client or potential customers, complete to create data assets while CRM precisely recommends；
After the resource recommendation method and device that some embodiments of the invention provide further match recommendation resource by collection
Feedback data and other conventional changes in characteristic, persistently carry out model training with optimize recommend, realize and hold
It is continuous to provide the user optimization, accurately recommend.
Brief description of the drawings
By reading the detailed description made to non-limiting example made with reference to the following drawings, the application's is other
Feature, objects and advantages will become more apparent upon：
Fig. 1 is a kind of flow chart for resource recommendation method that one embodiment of the invention provides.
Fig. 2 is a kind of flow chart for resource recommendation method that another embodiment of the present invention provides.
Fig. 3 is the flow chart of step S30 in one embodiment of the present invention.
Fig. 4 be method shown in Fig. 3 a kind of preferred embodiment in step S34 flow chart.
Fig. 5 is the schematic diagram of propagated forward process in method shown in Fig. 3 and Fig. 4.
Fig. 6 is the schematic diagram that process is reversely adjusted in method shown in Fig. 3 and Fig. 4.
Fig. 7 is a kind of structural representation for resource recommendation device that one embodiment of the invention provides.
Fig. 8 is a kind of structural representation for resource recommendation device that another embodiment of the present invention provides.
Fig. 9 is the structural representation of taxon in one embodiment of the present invention.
Figure 10 is a kind of structural representation for equipment that one embodiment of the invention provides.
Embodiment
The application is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining related invention, rather than the restriction to the invention.It also should be noted that in order to
It is easy to describe, the part related to invention is illustrate only in accompanying drawing.
It should be noted that in the case where not conflicting, the feature in embodiment and embodiment in the application can phase
Mutually combination.Describe the application in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
Fig. 1 is a kind of flow chart for resource recommendation method that one embodiment of the invention provides.
As shown in figure 1, in the present embodiment, the present invention provides a kind of resource recommendation method, including：
S10：Sale resource and the characteristic of customer resources are gathered respectively；
S20：According to the characteristic generation sale resource training sample and customer resources training sample gathered；
S30：Sale resource training sample and customer resources training sample input resource classification model are trained respectively,
Obtain sale resource classification results and customer resources classification results；
S40：Sale resource classification results and customer resources classification results input similarity model are trained, if output
Dry matching result is as recommendation.
Specifically, in step slo, on the one hand, gather the characteristic of sale resource various dimensions.Institute in the present embodiment
The characteristic of the sale resource of collection includes the data of following several types：Sale interest characteristics, it is good at industry line feature, society
Hand over relationship characteristic, ascribed characteristics of population feature, performance feature, geographic location feature etc..
Wherein, sale interest characteristics is used for evaluation factor little as a weight in subsequent step S40 matching,
Specifically include exploitation client, safeguard existing customer, client's after-sale service etc..
Industry line feature is good to be used to pass through the client that is developed in the sector line of the sale and resource quantity, caused stream
Whether the sale of the index evaluations such as water, gross profit is good at the sector line, and it is straight to specifically include single group class data, shops's class data, advertisement
The class that is open to traffic data etc., cover main flow classification industry line.
Social networks feature and ascribed characteristics of population feature belong to individualized feature, for covering personal attribute's difference of the sale
Property and hand in customer resources some classification assessment factors.
Performance feature is used to assess the comprehensive performance ability of the sale, single group in specifically including current exploitation or safeguarding,
Flowing water caused by the existing business of each service line such as shops, through train, gross profit, coverage rate, turnover rate, supplement with money, consume, reimbursement
Deng performance indicators.
On the other hand, core feature, label characteristics and some individualized features of customer resources are chosen, are shielded simultaneously
Or reduce noise effect caused by all kinds of extraneous factors.The characteristic of the customer resources gathered in the present embodiment includes following
The data of several types：Client GMV features, brand recognition feature, shops's label characteristics, category Cover Characteristics, geographical position
Feature etc..
Wherein, client GMV features, which are used to being used as a factor by observing client GMV height, judges client's contribution degree,
Specifically include achievement flowing water, achievement gross profit, checking flowing water, checking gross profit, recharge amount, spending amount, refund amount etc..
Brand recognition feature is used for comprehensive assessment client's brand recognition, while also serve as client's influence power one sentences
Disconnected factor, specifically include trade company's public praise, user's evaluation, whether investigate feedback etc. under chain brand, advertisement coverage rate, line.
Shops's label characteristics are used for the factor of merit of comprehensive assessment client, specifically include whether high-quality shops, shops's rank etc..
Category Cover Characteristics are used for the influence power of comprehensive assessment client, specifically include and whether cover core category, specific product
Coverage rate of class etc..
Above-mentioned sale resource characteristic and customer resources characteristic can be obtained by a system channel collection,
It can be gathered and obtained by multiple different system channels.
The sale resource characteristic and customer resources characteristic gathered in the present embodiment is specifically listed above,
In more different embodiments, the different types of sale resource characteristic of collection can be configured according to the actual requirements and customer resources is special
Levy data, or the characteristic gathered configures different assessment purposes.
Preferably, the characteristic of the customer resources gathered in step S10 can further include customer resources and exist
Feedback data after the sale resource that matching step S40 is recommended；Accordingly, the characteristic of sale resource can also enter one
Step includes sale resource matching step S40 and recommends the feedback data after customer resources.
In step S20, for the step S10 sale resource characteristics gathered and customer resources characteristic, choosing
The characteristic of 5 different time segment limits is taken to generate some sale resource training samples and some customer resources training samples use
In step S30 model training, so as to reduce the influence that data specificity is brought as far as possible；Some sale are generated in the same way
Resource verifies sample and some customer resources checking sample, for the checking in training process.When characteristic and generated
When sample size is less, it also can only generate training sample and verify sample without generating.
In step s 30, entered respectively according to the step S20 sale resource training samples generated and customer resources training sample
Resource classification result and customer resources classification results are sold in row model training, respectively output.In the present embodiment, resource classification mould
Type specifically enters using by Nonlinear Multi perceptron (MLP) with reference to the method reversely adjusted after the output of Softmax activation primitives
Row training, can be described in detail by the method shown in Fig. 3-Fig. 6 below., can be according to different demands in more embodiments
Resource classification model is configured to other machine learning classification models commonly used in the art, can also configure other differences in a model
Activation primitive.
In step s 40, the present embodiment specifically uses cosine similarity model, can be according to difference in more embodiments
Similarity model is configured to the similarity model of other different similarity algorithms of configuration commonly used in the art by demand.
Using cosine similarity model training and the detailed process recommended according to output result it is as follows：
Carry out the initial configuration of cosine similarity model：Batch size, the initial learning rate of network are set, and configuration uses
RMSProp optimization methods etc.；
The sale resource classification results and customer resources classification results obtained according to step S30 training, sale interest is special
Sign, geographic location feature, it is good at the sales performances such as industry line feature, social networks feature, ascribed characteristics of population feature, performance feature point
The sales performance hidden layer of cosine similarity model is not mapped to；It is and client GMV features, geographic location feature, brand is well-known
The client characteristics such as degree feature, shops's label characteristics, category Cover Characteristics are respectively mapped to the client characteristics of cosine similarity model
Hidden layer；
Sales performance is connected and is added to form the final expression of sales performance entirely respectively, and, client characteristics are distinguished
Connect and be added the final expression to form client characteristics entirely；
Recurrence loss (regression_cost) is configured into loss function to be trained, regression error cost is calculated and makees
For output, by training result assessment models, the best model wheel number of effect is found out；
After being finally completed training, cosine similarity is exported, returns to some customer resources of matching degree highest to corresponding
Sale resource, it is finally completed the intelligent recommendation matching of CRM resources.
Fig. 2 is a kind of flow chart for resource recommendation method that another embodiment of the present invention provides.As shown in Fig. 2 another
In embodiment, method of the resource recommendation method provided by the invention compared to Fig. 1, not including the step S10 in method shown in Fig. 1
And S20.
Specifically, in method shown in Fig. 2, when the system as data source channel has been extracted and is stored available for generating
During the characteristic of training sample and checking sample, resource recommendation method provided by the invention need to only include above-mentioned steps S20-
S40, the characteristic without carrying out step S10 again gather；
When the system as data source channel has further generated the training being trained available for resource classification model
When sample and checking sample, resource recommendation method provided by the invention need to only include above-mentioned steps S30-S40, without carrying out again
Step S10 characteristic collection and S20 sample generate.
It will be explained in detail below defeated with reference to Softmax activation primitives by Nonlinear Multi perceptron (MLP) in step S30
Go out the training method reversely adjusted afterwards.
Fig. 3 is the flow chart of step S30 in one embodiment of the present invention.As shown in figure 3, in a preferred embodiment,
Following steps are performed to sale resource training sample and customer resources training sample respectively in step s 30：
S31：Training sample is spliced into vector, inputs multilayer perceptron；
S32：The output result of multilayer perceptron is inputted to activation primitive；
S33：Whether within a predetermined range to judge the error of the output result of activation primitive and corresponding expected result：
It is no, then perform step S34：Reversely adjust the weight of each node of multilayer perceptron according to error, return to step S32 after
It is continuous to be trained；
It is then to perform step S35：Output category result.
Specifically, in step s 32, the initial weight of each node of multilayer perceptron is random arrangement, subsequently through
The weight of step S34 reverse each node of adjustment progressively adjusted optimization.In more embodiments, it can also use and rule of thumb match somebody with somebody
Put the different collocation methods such as the initial weight of each node of multilayer perceptron.In step S33, corresponding expected result can be checking
The output of collection or known expected result.In step S34, carried out instead using back-propagation algorithm in the present embodiment
To adjustment, in more embodiments, the other gradient descent algorithms for reversely adjusting in this area can be configured according to the actual requirements.
Fig. 4 be method shown in Fig. 3 a kind of preferred embodiment in step S34 flow chart.It is as shown in figure 4, excellent one
Select in embodiment, step S34 includes：
S341：The gradient of each weight is calculated according to error and back-propagation algorithm；
S342：Each weight is adjusted according to each gradient and back-propagation algorithm.
The reverse method of adjustment shown in above-mentioned Fig. 3-4 is described in detail below by way of a specific example.Fig. 5 and figure
6 be respectively the schematic diagram of propagated forward process and reverse adjustment process in method shown in Fig. 3 and Fig. 4.
As shown in figure 5, the input layer of multilayer perceptron has two nodes A1 and A2 in addition to bias node.Passed in forward direction
During broadcasting, node A1 and A2 distinguish receiving step S31 and splice achievement flowing water (10500000) and refund amount in vector
(465345).Bias node, node A1, node A2 export to hidden layer initial weight difference random arrangement be w1, w2, w3.
Then have, hide node layer B1 output VB1For：
VB1=fB1(1*w1+10500000*w2+465345*w3)；
Hide node layer B2 output VB2For：
VB2=fB2(1*w1+10500000*w2+465345*w3)；
Wherein, fB1() and fB2() is respectively node B1 and B2 activation primitive.
Output node layer C1 and C2 output V is calculated in the same mannerC1And VC2, then the V by outputC1And VC2Input
Softmax activation primitives, obtain the output result of activation primitive：
P (C1)=exp (VC1)/(exp(VC1)+exp(VC2))=0.4；
P (C2)=exp (VC2)/(exp(VC1)+exp(VC2))=0.6.
Known expected results are (1,0), the preset range of error be no more than ± 0.25, during this propagated forward,
The error of activation primitive output result (0.4,0.6) is (- 0.6,0.6), more than preset range, it is necessary to be carried out according to error reverse
Adjustment.
As shown in fig. 6, the gradient of each weight is calculated according to the above-mentioned error calculated and pre-configured back-propagation algorithm,
Further according to weight corresponding to the gradient adjustment calculated.Specific calculation formula is it will be appreciated by those skilled in the art that herein not
Repeat again.
After adjusting every weight, continue propagated forward and the circulation reversely adjusted, until the mistake of activation primitive output result
Difference is no more than preset range, then output category result.
Fig. 7 is a kind of structural representation for resource recommendation device that one embodiment of the invention provides.Device shown in Fig. 7 can
The corresponding method performed shown in Fig. 1.
As shown in fig. 7, in the present embodiment, the present invention provides a kind of resource recommendation device 10, including collecting unit 11, sample
This generation unit 12, taxon 13 and matching unit 14.
Collecting unit 11 is configured to gather sale resource and the characteristic of customer resources respectively.
Sample generation unit 12 is configured to according to the characteristic generation sale resource training sample gathered and client
Resource training sample.
Taxon 13 is configured to sale resource training sample and customer resources training sample input resource point respectively
Class model is trained, and obtains sale resource classification results and customer resources classification results.
Matching unit 14 is configured to sale resource classification results and customer resources classification results input similarity model
It is trained, exports some matching results as recommendation.
The model training method that taxon 13 and matching unit 14 perform respectively refers to the method shown in Fig. 1, this
Place repeats no more.
Preferably, the characteristic for the customer resources that collecting unit 11 is gathered can further include customer resources and exist
Feedback data after the sale resource that matching step S40 is recommended；Accordingly, the sale resource that collecting unit 11 is gathered
Characteristic can also further comprise that sale resource matching step S40 recommends the feedback data after customer resources.
On the one hand the various embodiments described above model training by machine learning and carry out intelligence point to sale resource and customer resources
Class, and intelligent Matching is carried out to classification results, solve the problems, such as cold start-up, realize more accurate compared to existing scheme
Recommend；On the other hand by excavating the largely data message on client or potential customers to the intelligent classification of customer resources,
Complete to create data assets while CRM precisely recommends；And further match the feedback coefficient after recommending resource by gathering
According to this and the characteristic in other conventional changes, persistently carry out model training to optimize to recommend, realize and be continuously user
Optimization is provided, accurately recommended.
Fig. 8 is a kind of structural representation for resource recommendation device that another embodiment of the present invention provides.Device shown in Fig. 8
The method performed shown in Fig. 2 can be corresponded to.
As shown in figure 8, in another embodiment, compared to the device shown in Fig. 7, the resource recommendation device 10 shown in Fig. 8
Taxon 13 and matching unit 14 can be only included, sample generation unit 12 can also be further comprised.
Fig. 9 is the structural representation of taxon in one embodiment of the present invention.Device shown in Fig. 9, which can correspond to, to be performed
Method shown in Fig. 3-6.
As shown in figure 9, in a preferred embodiment, it is single that taxon 13 includes vectorization subelement 131, propagated forward
Member 132, reversely judgment sub-unit 133, adjustment subelement 134 and output subelement 135.
Wherein, vectorization subelement 131 is configured to training sample being spliced into vector, inputs multilayer perceptron；Forward direction
Subelement 132 is propagated to be configured to input the output result of multilayer perceptron to activation primitive；The configuration of judgment sub-unit 133 is used
In the output result and the error of corresponding expected result for judging activation primitive whether within a predetermined range；Reversely adjustment subelement
134 be configured to judged result be error not within the predefined range when each node of multilayer perceptron is reversely adjusted according to error
Weight continues to train for propagated forward subelement 132；It is that error exists that output subelement 135, which is configured in judged result,
Output category result when in preset range.Above-mentioned training sample is sale resource training sample or customer resources training sample.
In a preferred embodiment, reversely adjustment subelement 134 is further configured to according to error and backpropagation calculation
Method calculates the gradient of each weight, and, each weight is adjusted according to each gradient and back-propagation algorithm.
In a preferred embodiment, propagated forward subelement 132 is further configured to random arrangement multilayer perceptron
The initial weight of each node.
Figure 10 is a kind of structural representation for equipment that one embodiment of the invention provides.
As shown in Figure 10, as on the other hand, present invention also provides a kind of equipment 1000, including one or more centers
Processing unit (CPU) 1001, its can according to the program being stored in read-only storage (ROM) 1002 or from storage part
1008 programs being loaded into random access storage device (RAM) 1003 and perform various appropriate actions and processing.In RAM1003
In, also it is stored with equipment 1000 and operates required various programs and data.CPU1001, ROM1002 and RAM1003 pass through total
Line 1004 is connected with each other.Input/output (I/O) interface 1005 is also connected to bus 1004.
I/O interfaces 1005 are connected to lower component：Importation 1006 including keyboard, mouse etc.；Including such as negative electrode
The output par, c 1007 of ray tube (CRT), liquid crystal display (LCD) etc. and loudspeaker etc.；Storage part including hard disk etc.
1008；And the communications portion 1009 of the NIC including LAN card, modem etc..Communications portion 1009 passes through
Communication process is performed by the network of such as internet.Driver 1010 is also according to needing to be connected to I/O interfaces 1005.It is detachable to be situated between
Matter 1011, such as disk, CD, magneto-optic disk, semiconductor memory etc., it is arranged on as needed on driver 1010, so as to
Storage part 1008 is mounted into as needed in the computer program read from it.
Especially, in accordance with an embodiment of the present disclosure, the resource recommendation method of any of the above-described embodiment description can be implemented
For computer software programs.For example, embodiment of the disclosure includes a kind of computer program product, it includes being tangibly embodied in
Computer program on machine readable media, the computer program include the program code for being used for performing resource recommendation method.
In such embodiments, the computer program can be downloaded and installed by communications portion 1009 from network, and/or from
Detachable media 1011 is mounted.
As another aspect, present invention also provides a kind of computer-readable recording medium, the computer-readable storage medium
Matter can be the computer-readable recording medium included in the device of above-described embodiment；Can also be individualism, it is unassembled
Enter the computer-readable recording medium in equipment.Computer-readable recording medium storage has one or more than one program, should
Program is used for performing the resource recommendation method for being described in the application by one or more than one processor.
Flow chart and block diagram in accompanying drawing, it is illustrated that according to the system of various embodiments of the invention, method and computer journey
Architectural framework in the cards, function and the operation of sequence product.At this point, each square frame in flow chart or block diagram can generation
The part of one module of table, program segment or code, the part of the module, program segment or code include one or more use
In the executable instruction of logic function as defined in realization.It should also be noted that marked at some as in the realization replaced in square frame
The function of note can also be with different from the order marked in accompanying drawing generation.For example, two square frames succeedingly represented are actually
It can perform substantially in parallel, they can also be performed in the opposite order sometimes, and this is depending on involved function.Also
It is noted that the combination of each square frame and block diagram in block diagram and/or flow chart and/or the square frame in flow chart, Ke Yitong
Function as defined in execution or the special hardware based system of operation are crossed to realize, or can be by specialized hardware with calculating
The combination of machine instruction is realized.
Being described in unit or module involved in the embodiment of the present application can be realized by way of software, can also
Realized by way of hardware.Described unit or module can also be set within a processor, for example, each unit can
With the software program being provided in computer or intelligent movable equipment or the hardware unit being separately configured.Wherein, this
The title of a little units or module does not form the restriction to the unit or module in itself under certain conditions.
Above description is only the preferred embodiment of the application and the explanation to institute's application technology principle.People in the art
Member should be appreciated that invention scope involved in the application, however it is not limited to the technology that the particular combination of above-mentioned technical characteristic forms
Scheme, while should also cover in the case where not departing from the application design, appointed by above-mentioned technical characteristic or its equivalent feature
Other technical schemes that meaning is combined and formed.Such as features described above has similar functions with (but not limited to) disclosed herein
The technical characteristic technical scheme being replaced mutually and formed.
Claims (16)
- A kind of 1. resource recommendation method, it is characterised in that including：Sale resource training sample and customer resources training sample input resource classification model are trained respectively, sold Resource classification result and customer resources classification results；The sale resource classification results and customer resources classification results input similarity model are trained, export some With result as recommendation.
- 2. according to the method for claim 1, it is characterised in that described respectively by sale resource training sample and customer resources Training sample input resource classification model is trained, and obtaining sale resource classification results and customer resources classification results includes：Training sample is spliced into vector, inputs multilayer perceptron；The output result of the multilayer perceptron is inputted to activation primitive；Whether within a predetermined range to judge the error of the output result of the activation primitive and corresponding expected result：It is no, then the weight of each node of the multilayer perceptron is reversely adjusted according to the error to continue to train；It is, then output category result；Wherein, the training sample is the sale resource training sample or the customer resources training sample.
- 3. according to the method for claim 2, it is characterised in that it is described according to the error reversely adjust each weight with Continuing training includes：The gradient of each weight is calculated according to the error and back-propagation algorithm；Each weight is adjusted according to each gradient and the back-propagation algorithm.
- 4. according to the method for claim 2, it is characterised in that it is described that training sample is spliced into vector, input multilayer sense Know that device also includes：The initial weight of each node of multilayer perceptron described in random arrangement.
- 5. according to the method for claim 2, it is characterised in that the activation primitive is Softmax functions.
- 6. according to the method for claim 1, it is characterised in that the similarity model is cosine similarity model.
- 7. according to the method described in claim any one of 1-6, it is characterised in that also include：Sale resource and the characteristic of customer resources are gathered respectively；According to the characteristic generation sale resource training sample and customer resources training sample gathered.
- 8. according to the method for claim 7, it is characterised in that the characteristic of the customer resources includes matching and recommended Feedback data after sale resource；And/orThe characteristic of the sale resource includes matching and recommends the feedback data after customer resources.
- A kind of 9. resource recommendation device, it is characterised in that including：Taxon, it is configured to that sale resource training sample and customer resources training sample are inputted into resource classification model respectively It is trained, obtains sale resource classification results and customer resources classification results；Matching unit, it is configured to enter the sale resource classification results and customer resources classification results input similarity model Row training, some matching results are exported as recommendation.
- 10. device according to claim 9, it is characterised in that the taxon includes：Vectorization subelement, it is configured to training sample being spliced into vector, inputs multilayer perceptron；Propagated forward subelement, it is configured to input the output result of the multilayer perceptron to activation primitive；Judgment sub-unit, it is configured to judge the output result of the activation primitive with the error of corresponding expected result whether pre- Determine in scope；Reversely adjustment subelement, be configured to judged result be error not within the predefined range when reversely adjusted according to the error The weight of whole each node of the multilayer perceptron continues to train for the propagated forward subelement；Export subelement, be configured to judged result be error within a predetermined range when output category result；Wherein, the training sample is the sale resource training sample or the customer resources training sample.
- 11. device according to claim 10, it is characterised in that the reversely adjustment subelement is further configured to root The gradient of each weight is calculated according to the error and back-propagation algorithm, and, according to each gradient and the reversely biography Broadcast algorithm and adjust each weight.
- 12. device according to claim 10, it is characterised in that the propagated forward subelement be further configured to Machine configures the initial weight of each node of the multilayer perceptron.
- 13. according to the device described in claim any one of 9-12, it is characterised in that also include：Collecting unit, it is configured to gather sale resource and the characteristic of customer resources respectively；Sample generation unit, it is configured to according to the characteristic generation sale resource training sample and customer resources instruction gathered Practice sample.
- 14. device according to claim 13, it is characterised in that the collecting unit be further configured to it is following at least One：Feedback data of the customer resources after sale resource is recommended in matching, sale resource is after customer resources is recommended in matching Feedback data.
- 15. a kind of equipment, it is characterised in that the equipment includes：One or more processors；Memory, for storing one or more programs,When one or more of programs are by one or more of computing devices so that one or more of processors Perform the method as any one of claim 1-8.
- 16. a kind of computer-readable recording medium for being stored with computer program, it is characterised in that the program is executed by processor Methods of the Shi Shixian as any one of claim 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710750386.1A CN107609060A (en) | 2017-08-28 | 2017-08-28 | Resource recommendation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710750386.1A CN107609060A (en) | 2017-08-28 | 2017-08-28 | Resource recommendation method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107609060A true CN107609060A (en) | 2018-01-19 |
Family
ID=61056297
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710750386.1A Pending CN107609060A (en) | 2017-08-28 | 2017-08-28 | Resource recommendation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107609060A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108230122A (en) * | 2018-03-21 | 2018-06-29 | 北京贝塔智投科技有限公司 | A kind of interactive financing system and method |
| CN108268660A (en) * | 2018-02-08 | 2018-07-10 | 深圳市口袋网络科技有限公司 | A kind of customer data recommends method, server and storage medium |
| CN108415992A (en) * | 2018-02-12 | 2018-08-17 | 百度在线网络技术（北京）有限公司 | Resource recommendation method, device and computer equipment |
| CN109255646A (en) * | 2018-07-27 | 2019-01-22 | 国政通科技有限公司 | Deep learning is carried out using big data to provide method, the system of value-added service |
| CN109460512A (en) * | 2018-10-25 | 2019-03-12 | 腾讯科技（北京）有限公司 | Recommendation information processing method, device, equipment and storage medium |
| WO2019223145A1 (en) * | 2018-05-23 | 2019-11-28 | 平安科技（深圳）有限公司 | Electronic device, promotion list recommendation method and system, and computer-readable storage medium |
| CN111898006A (en) * | 2020-07-30 | 2020-11-06 | 拉扎斯网络科技（上海）有限公司 | Resource data processing method and device |
| CN112232612A (en) * | 2019-06-28 | 2021-01-15 | 百度在线网络技术（北京）有限公司 | Intelligent client data distribution method and device |
| CN113298087A (en) * | 2021-04-29 | 2021-08-24 | 上海淇玥信息技术有限公司 | Method, system, device and medium for cold start of picture classification model |
| CN113919869A (en) * | 2021-09-28 | 2022-01-11 | 上海画龙信息科技有限公司 | Equity distribution method and device based on sales increment model and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105488662A (en) * | 2016-01-07 | 2016-04-13 | 北京歌利沃夫企业管理有限公司 | Bi-directional recommendation-based online recruitment system |
| US20170140262A1 (en) * | 2012-03-09 | 2017-05-18 | Nara Logics, Inc. | Systems and methods for providing recommendations based on collaborative and/or content-based nodal interrelationships |
| CN106933996A (en) * | 2017-02-28 | 2017-07-07 | 广州大学 | A kind of recommendation method of use depth characteristic matching |
-
2017
- 2017-08-28 CN CN201710750386.1A patent/CN107609060A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170140262A1 (en) * | 2012-03-09 | 2017-05-18 | Nara Logics, Inc. | Systems and methods for providing recommendations based on collaborative and/or content-based nodal interrelationships |
| CN105488662A (en) * | 2016-01-07 | 2016-04-13 | 北京歌利沃夫企业管理有限公司 | Bi-directional recommendation-based online recruitment system |
| CN106933996A (en) * | 2017-02-28 | 2017-07-07 | 广州大学 | A kind of recommendation method of use depth characteristic matching |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108268660A (en) * | 2018-02-08 | 2018-07-10 | 深圳市口袋网络科技有限公司 | A kind of customer data recommends method, server and storage medium |
| CN108415992A (en) * | 2018-02-12 | 2018-08-17 | 百度在线网络技术（北京）有限公司 | Resource recommendation method, device and computer equipment |
| CN108415992B (en) * | 2018-02-12 | 2022-03-04 | 百度在线网络技术（北京）有限公司 | Resource recommendation method and device and computer equipment |
| CN108230122A (en) * | 2018-03-21 | 2018-06-29 | 北京贝塔智投科技有限公司 | A kind of interactive financing system and method |
| WO2019223145A1 (en) * | 2018-05-23 | 2019-11-28 | 平安科技（深圳）有限公司 | Electronic device, promotion list recommendation method and system, and computer-readable storage medium |
| CN109255646A (en) * | 2018-07-27 | 2019-01-22 | 国政通科技有限公司 | Deep learning is carried out using big data to provide method, the system of value-added service |
| CN109460512A (en) * | 2018-10-25 | 2019-03-12 | 腾讯科技（北京）有限公司 | Recommendation information processing method, device, equipment and storage medium |
| CN109460512B (en) * | 2018-10-25 | 2022-04-22 | 腾讯科技（北京）有限公司 | Recommendation information processing method, device, equipment and storage medium |
| CN112232612A (en) * | 2019-06-28 | 2021-01-15 | 百度在线网络技术（北京）有限公司 | Intelligent client data distribution method and device |
| CN111898006A (en) * | 2020-07-30 | 2020-11-06 | 拉扎斯网络科技（上海）有限公司 | Resource data processing method and device |
| CN113298087A (en) * | 2021-04-29 | 2021-08-24 | 上海淇玥信息技术有限公司 | Method, system, device and medium for cold start of picture classification model |
| CN113919869A (en) * | 2021-09-28 | 2022-01-11 | 上海画龙信息科技有限公司 | Equity distribution method and device based on sales increment model and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107609060A (en) | Resource recommendation method and device | |
| Ngai et al. | Application of data mining techniques in customer relationship management: A literature review and classification | |
| Van den Bulte et al. | Bias and systematic change in the parameter estimates of macro-level diffusion models | |
| Bauer et al. | Recommender systems based on quantitative implicit customer feedback | |
| CN107403345A (en) | Best-selling product Forecasting Methodology and system, storage medium and electric terminal | |
| Li et al. | AI-driven customer relationship management for sustainable enterprise performance | |
| Seymen et al. | Customer churn prediction using deep learning | |
| Čavoški et al. | Agent-based modelling and simulation in the analysis of customer behaviour on B2C e-commerce sites | |
| CN115619571A (en) | Financing planning method, system and device | |
| Allaymoun et al. | Data visualization and statistical graphics in big data analysis by Google Data Studio–Sales Case Study | |
| Raza et al. | Performance analysis of deep approaches on airbnb sentiment reviews | |
| Gao | Big Data analysis on E-commerce platform | |
| Kumar et al. | Big Mart Sales Prediction Using Machine Learning | |
| Cao | Research on the impact of artificial intelligence-based e-commerce personalization on traditional accounting methods | |
| Singhal et al. | A prediction model for benefitting e-commerce through usage of regional data: A new framework | |
| CN111815351A (en) | Cooperative filtering and association rule-based clothing recommendation method | |
| CN115587170A (en) | Personalized problem item recommendation method, system and storage medium | |
| Soni et al. | Big data analytics for market prediction via consumer insight | |
| Khaneghah et al. | Artemis time: A mathematical model to calculate maximum acceptable waiting time in B2C e-commerce | |
| Yuan | Decision support system for evaluating English media analytics and the role of visualization in online business development using CRITIC and TOPSIS approaches | |
| Singh | Decision Making and Predictive Analysis for Real Time Data | |
| Chhabra et al. | Data Science and Knowledge Discovery through Data Mining Paradigms | |
| Zeng et al. | A knowledge discovery and data mining process model in E-marketing | |
| Marković et al. | Analysis of Interactions of Key Stakeholders on B2C e-Markets-Agent Based Modelling and Simulation Approach | |
| Istrat et al. | Creating a decision-making model using association rules |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180119 |
|
| RJ01 | Rejection of invention patent application after publication |