CN110348928A - Information-pushing method, device and computer readable storage medium - Google Patents
Information-pushing method, device and computer readable storage medium Download PDFInfo
- Publication number
- CN110348928A CN110348928A CN201810304124.7A CN201810304124A CN110348928A CN 110348928 A CN110348928 A CN 110348928A CN 201810304124 A CN201810304124 A CN 201810304124A CN 110348928 A CN110348928 A CN 110348928A
- Authority
- CN
- China
- Prior art keywords
- target
- order
- user
- data
- time
- 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
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000003860 storage Methods 0.000 title claims abstract description 16
- 230000003542 behavioural effect Effects 0.000 claims abstract description 68
- 238000013135 deep learning Methods 0.000 claims abstract description 31
- 238000013528 artificial neural network Methods 0.000 claims abstract description 26
- 230000006399 behavior Effects 0.000 claims description 24
- 239000000284 extract Substances 0.000 claims description 6
- 238000013075 data extraction Methods 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims 1
- 210000005036 nerve Anatomy 0.000 claims 1
- 210000004218 nerve net Anatomy 0.000 claims 1
- 238000006243 chemical reaction Methods 0.000 abstract description 10
- 238000005516 engineering process Methods 0.000 abstract description 6
- 238000010586 diagram Methods 0.000 description 20
- 238000004590 computer program Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 4
- 241001269238 Data Species 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000002452 interceptive effect Effects 0.000 description 3
- 239000000463 material Substances 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000005266 casting Methods 0.000 description 2
- 230000001186 cumulative effect Effects 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000006116 polymerization reaction Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000003032 molecular docking Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002688 persistence Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
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
- 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]
- G06Q30/0631—Item recommendations
-
- 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]
- G06Q30/0633—Lists, e.g. purchase orders, compilation or processing
Abstract
Present disclose provides a kind of information-pushing method, device and computer readable storage mediums, are related to field of computer technology.Information-pushing method therein includes: the behavioral data and time of the act data for extracting user about the target cancelled an order after placing an order；Behavioral data and time of the act data are handled using deep learning neural network, to predict whether user will buy back target；In the case where predicting that user does not buy back target, the relevant information of target is pushed to user.The disclosure can accurately be predicted to cancel an order after placing an order in real time does not buy back the user of target, improves the accuracy of information push, so that improving user while saving information push cost buys back conversion ratio.
Description
Technical field
This disclosure relates to field of computer technology, in particular to a kind of information-pushing method, device and computer-readable deposit
Storage media.
Background technique
Currently, there are a large number of users in electric business platform under after complete order actively or eliminates order automatically.By with
It is analyzed toward data exploration, there are still a large number of users to have the behaviors such as browsing, concern plus purchase, but last non-shape after cancelling an order
At transaction, a large amount of high potential users is caused to be lost.
Based on such phenomenon, the user that how to predict high purchase intention and will be lost, in a manner of real-time intelligent into
The information push that row retrieves formula buys back conversion ratio to improve user, is current urgent problem to be solved.
Summary of the invention
The technical issues of disclosure solves is how accurately to predict to cancel an order after placing an order in real time not buy back the use of target
The accuracy of information push is improved at family, improves user while saving information push cost and buys back conversion ratio.
According to the one aspect of the embodiment of the present disclosure, a kind of information-pushing method is provided, comprising: extract user about under
The behavioral data and time of the act data for the target cancelled an order after list；Using deep learning neural network to behavioral data with
And time of the act data are handled, to predict whether user will buy back target；In the case where predicting that user does not buy back target,
The relevant information of target is pushed to user.
In some embodiments, deep learning neural network meets preset condition in behavioral data and time of the act data
In the case where, prediction user does not buy back target.
In some embodiments, behavioral data includes browsing data and order data of the user about target.
In some embodiments, browsing data include that the following contents at least one of is worked as: whether browsing objective, whether clear
Look at same category target, the total degree of browsing objective, browsing with category target total degree, whether click the commodity details page of target
Master map, whether click target commodity details page comment, whether have brush select difference comment behavior, whether have brush choose comment behavior, whether
Click the big figure of comment, whether click specifications and models, whether click inventory bookmark, whether browse commodity question and answer area, whether
It puts question in commodity question and answer area.
In some embodiments, order data includes that the following contents at least one of is worked as: whether target is carried out plus is purchased,
Whether to target subtract purchase, whether target paid close attention to, whether once placed an order to target, whether once cancelled an order to target,
Whether once target was successfully bought, whether once placed an order to same category target, whether once cancelled an order, whether once to same category target
Said target is successfully bought.
In some embodiments, time of the act data include that the following contents at least one of is worked as: the same day browses to for the first time
The number of browsing objective, the same day are browsed to for the first time in the different time intervals for the period that places an order in different time intervals during placing an order
Browse with category target number, place an order cancel an order during different time intervals in browsing objective number, place an order to
In different time intervals during cancelling an order browsing with category target number, browse to for the first time the time interval to place an order, under
It is single to the time interval cancelled an order, cancel an order after browsing objective for the first time time interval.
In some embodiments, information-pushing method further include: using cancel an order after placing an order and buy back target user,
It cancels an order after placing an order and does not buy back behavioral data and time of the act data of the user about target of target, to deep learning mind
It is trained through network, deep learning neural network is enabled to predict whether user will buy back target.
In some embodiments, behavioral data and time of the act of the user about the target cancelled an order after placing an order are extracted
Data include: using Message Processing streaming frame to user about target related data carry out batch processing；It will locate in batches
Reason result is polymerize in service layer, obtains behavioral data and time of the act data of the user about target.
According to the other side of the embodiment of the present disclosure, a kind of information push-delivery apparatus is provided, comprising: data extract mould
Block is configured as extracting behavioral data and time of the act data of the user about the target cancelled an order after placing an order；User is pre-
Module is surveyed, is configured as handling behavioral data and time of the act data using deep learning neural network, with prediction
Whether user will buy back target；Info push module is configured as pushing away in the case where predicting that user does not buy back target to user
Send the relevant information of target.
In some embodiments, deep learning neural network meets preset condition in behavioral data and time of the act data
In the case where, prediction user does not buy back target.
In some embodiments, behavioral data includes browsing data and order data of the user about target.
In some embodiments, browsing data include that the following contents at least one of is worked as: whether browsing objective, whether clear
Look at same category target, the total degree of browsing objective, browsing with category target total degree, whether click the commodity details page of target
Master map, whether click target commodity details page comment, whether have brush select difference comment behavior, whether have brush choose comment behavior, whether
Click the big figure of comment, whether click specifications and models, whether click inventory bookmark, whether browse commodity question and answer area, whether
It puts question in commodity question and answer area.
In some embodiments, order data includes that the following contents at least one of is worked as: whether target is carried out plus is purchased,
Whether to target subtract purchase, whether target paid close attention to, whether once placed an order to target, whether once cancelled an order to target,
Whether once target was successfully bought, whether once placed an order to same category target, whether once cancelled an order, whether once to same category target
Said target is successfully bought.
In some embodiments, time of the act data include that the following contents at least one of is worked as: the same day browses to for the first time
The number of browsing objective, the same day are browsed to for the first time in the different time intervals for the period that places an order in different time intervals during placing an order
Browse with category target number, place an order cancel an order during different time intervals in browsing objective number, place an order to
In different time intervals during cancelling an order browsing with category target number, browse to for the first time the time interval to place an order, under
It is single to the time interval cancelled an order, cancel an order after browsing objective for the first time time interval.
In some embodiments, information push-delivery apparatus further includes deep learning neural metwork training module, is configured as benefit
With cancel an order after placing an order and buy back the user of target, place an order after cancel an order and do not buy back behavior of the user about target of target
Data and time of the act data, are trained deep learning neural network, and deep learning neural network is predicted
Whether user will buy back target.
In some embodiments, data extraction module is configured as: using Message Processing streaming frame to user about
The related data of target carries out batch processing；Batch processing result is polymerize in service layer, obtains user about target
Behavioral data and time of the act data.
According to the another aspect of the embodiment of the present disclosure, a kind of information push-delivery apparatus is provided, comprising: memory；And
It is coupled to the processor of memory, processor is used to be based on instruction stored in memory, executes information push side above-mentioned
Method.
According to another aspect of the embodiment of the present disclosure, a kind of computer readable storage medium is provided, wherein computer
Readable storage medium storing program for executing is stored with computer instruction, and instruction is executed by processor information-pushing method above-mentioned.
The disclosure can accurately be predicted to cancel an order after placing an order in real time does not buy back the user of target, improves information push
Accuracy, so that improving user while saving information push cost buys back conversion ratio.
By the detailed description referring to the drawings to the exemplary embodiment of the disclosure, the other feature of the disclosure and its
Advantage will become apparent.
Detailed description of the invention
In order to illustrate more clearly of the embodiment of the present disclosure or technical solution in the prior art, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Disclosed some embodiments without any creative labor, may be used also for those of ordinary skill in the art
To obtain other attached drawings according to these attached drawings.
Fig. 1 shows the flow diagram of the information-pushing method of an embodiment of the present disclosure.
Fig. 2 shows the flow diagrams that the disclosure extracts one embodiment of behavioral data and time of the act data.
Fig. 3 shows the flow chart of data processing schematic diagram of Stream Processing frame.
Fig. 4 shows the flow diagram of Stream Processing frame batch processing job.
Fig. 5 shows the flow diagram of the information-pushing method of the disclosure another embodiment.
Fig. 6 shows the structural schematic diagram of the information push-delivery apparatus of an embodiment of the present disclosure.
Fig. 7 shows the structural schematic diagram of the information push-delivery apparatus of the disclosure another embodiment.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present disclosure, the technical solution in the embodiment of the present disclosure is carried out clear, complete
Site preparation description, it is clear that described embodiment is only disclosure a part of the embodiment, instead of all the embodiments.Below
Description only actually at least one exemplary embodiment be it is illustrative, never as to the disclosure and its application or making
Any restrictions.Based on the embodiment in the disclosure, those of ordinary skill in the art are not making creative work premise
Under all other embodiment obtained, belong to the disclosure protection range.
Inventor attempts the conversion ratio that order is promoted from a kind of new angle.The original that inventor is cancelled based on different users
Because, the different different viewing cancelled after time, cancellation the behaviors such as clicks, judges whether it is more without the repurchase user that intervenes etc.
Dimension carries out deep study and analysis, real-time intelligent intervention is carried out to find the trigger point of retrieving of user, thus with new
Mode convection current appraxia family, which is made, retrieves change, retrieves biggish income in real time with lower cost.
It solves the above problems, there are following technological difficulties:
(1) intervene the timeliness retrieved: after user has loss feature, system needs to calculate in real time by algorithm most suitable
The time point intervened is closed, the timeliness of intervention time is ensured, does not miss best intervention point；
(2) under high-magnitude system stability: hundred million grades of electric business user base number, the data such as order, browsing, cancellation of user
Real-time report need to have the real time data computing capability under high-magnitude high concurrent, under mass data and complicated business logic, mention
For the support of Millisecond real time data and decision-making work；
(3) intervene the accuracy of crowd: being browsed in face of the user of magnanimity, place an order, cancel the behavioral data browsed again, deposited
In the principle of not cost of idleness, if user to be bought back itself cannot do and intervene, so be difficult to accurately to distinguish repurchase user and
Non- repurchase (loss) user intervenes.
Fig. 1 is combined to introduce the information-pushing method of an embodiment of the present disclosure first.
Fig. 1 shows the flow diagram of the information-pushing method of an embodiment of the present disclosure.As shown in Figure 1, this implementation
Information-pushing method in example includes step S102~step S106.
In step s 102, behavioral data and time of the act number of the user about the target cancelled an order after placing an order are extracted
According to.
Behavioral data may include browsing data and order data of the user about target.Wherein, browsing data can be with
At least one of work as including the following contents: whether browsing objective, whether browse same category target, browsing objective total degree,
The total degree with category target is browsed, the commodity details page master map of target whether is clicked, whether clicks the commodity details page of target
Comment, whether have brush select difference comment behavior, whether have brush choose comment behavior, whether click the big figure of comment, whether click specifications and models,
Inventory bookmark whether is clicked, commodity question and answer area whether is browsed, whether is putd question in commodity question and answer area；Order data may include
Whether whether the following contents at least one of is worked as: carrying out to target plus purchase, carry out subtracting purchase to target, close to target
It infuses, whether once placed an order, whether once cancelled an order to target to target, whether once target successfully being bought, whether once to same category
Whether whether target place an order, once cancelled an order, once successfully bought to said target to same category target.
Time of the act data include that the following contents at least one of is worked as: the same day browses to for the first time place an order during difference when
Between be spaced in the number of browsing objective, the same day browse in the different time intervals during placing an order browsing for the first time with category target
Number, place an order cancel an order during different time intervals in browsing objective number, place an order cancel an order during not
With in time interval browse with category target number, browse to for the first time the time interval to place an order, place an order to cancel an order when
Between be spaced, cancel an order after browsing objective for the first time time interval.
In step S104, behavioral data and time of the act data are handled using deep learning neural network,
To predict whether user will buy back target.
Deep learning neural network, can be pre- in the case where behavioral data and time of the act data meet preset condition
It surveys user and does not buy back target.Based on previous a large number of users is browsed, place an order, cancel the behavioral data browsed again and user itself stream
The data exploration for losing the multiple dimensions such as attribute is analyzed and is polymerize, to a set of behavioral data of each three-level category training and time of the act
Model of the data about the category rank for the property retrieved retrieve which user of the decision of system intelligence can, to reach
To the effect of promotion conclusion of the business total amount, while also more suiting the vision being intelligently sold.
For example, A indicates that user buries a behavioral data what the shopping of electric business sample platform generated, include browsing pages information, browsing
Number clicks position, concern commodity plus shopping cart, a series of elements that influences place an order and cancel such as buys, places an order, cancelling；T
It indicates the data sequence of time correlation, mainly by lower single message, message of cancelling an order and buries a behavioral data and generate, including
Click data is browsed before placing an order to arrive the time interval quartile evidence to place an order, place an order to the browsing click data time between cancellation
The quartile at interval according to, cancel after to the behavioral data time interval, behavioral data generation that place an order again period (behavior
The period that data generate is time interval between true two key operations, for example we can will place an order to cancelling an order
The number of minutes of this period is also input to deep learning neural network separately as decision duration characteristics) etc.；R indicates that the user is
No is that can retrieve user, is as positive example sample, other situations to the user's mark to place an order again after cancellation according to pin experience is adopted
As negative example sample.The value of R is that 0 or 1,1 expression user can be intervened, and 0 expression user cannot be intervened.
The quartile evidence of time interval is explained below.Specific time interval is divided into the quartering, analysis system
Count the characteristic at each interval.Such as the user in first 4 hours that places an order will generate 4 characteristic values to the number of clicks of the detailed page of quotient:
Preceding 4-3 hours of total number of clicks, preceding 3-2 hours of total number of clicks, preceding 2-1 hours of total number of clicks and preceding click in 1 hour are total
Number.Discovery after for statistical analysis to a large amount of historical data, user execute key operations (for example place an order, cancel an order) it
In preceding a period of time, relevant behavioral data has specific rule on time dimension, and (closer to lower single time, click is got over
Frequently).For the browsing data before placing an order, the characteristic value for taking the statistical data of its quartile to generate can more really react
The behavioral data of user.
Vector A and T pass through to be counted under line, is carried out Macro or mass analysis to the data of category rank, is passed through machine learning related side
Both method training to the model M of R, in this way we just obtained category rank about action trail and time of the act track to can
The decision model for the property retrieved.When user's shopping, according to the category that it is browsed, corresponding trained M is chosen as meter
Model is calculated, its behavioral data A and time of the act data T is counted in real time as input, can thus obtain this in real time
Whether user is that can retrieve user R.
In step s 106, in the case where predicting that user does not buy back target, the relevant information of target is pushed to user.
After the real-time behavioral data of user flows into computing platform, it is not that output result R, which shows the user under the category,
It is repairable user.For type user can be retrieved, user can be kept here using Different Strategies, it is guided to complete shopping.Example
Such as: passive touching reaches, for this certain customers, if there is the behavior of the browsing detailed page of quotient again in its next behavioral data, that
It can be its active push category discount coupon, to achieve the purpose that keep user here；Active marketing sends short messages and gives this certain customers,
Include the frightened web page address for enjoying valence, user placed an order by corresponding web page address can enjoy it is exclusive preferential.
Above-described embodiment can accurately be predicted to cancel an order after placing an order not buying back the user of target in real time, relative to extensive
The selection information for formula of casting net pushes user, and the present embodiment can make relatively reasonable fining to target user and divide, to not
Different information push strategies is made with user, the accuracy of information push is improved, is mentioned while saving information push cost
High user buys back conversion ratio.At the same time it can also determine the material time point of information push using above-described embodiment, cancellation is ordered
Single user can provide the natural change-over period of the reasonable category rank of comparison, both in view of the complicated heart of user psychology variation
Decision factor is managed, also using having there is experience to predict the following time buying of user.
The data such as order, browsing and the cancellation of user need real-time report, and single peak can under cancelling during the big rush of electric business
Up to 100,000/second, real time data processing peak value is up to 400,000,000/second, to make system delay less than 1 second, involves how to efficiently locate
Reason and storing data.Above-mentioned data source mainly includes two classes, and one kind is the behavioral data that each channel customer end reports, another kind of
It is that user that each business subsystem reports such as places an order and cancel at the behavioral datas.
The behavioral data of client mainly buries a generation by client, and user can trigger various when being applied using electric business
The page of various kinds, which buries, a little reports corresponding behavioral data.Behavioral data can be collected by the software of Kafka message transmission and carry out unification
Processing.The characteristics of this part messages is data volume big (million seconds grade data of peak value), quasi real time (data delay was on 6 seconds left sides
It is right).For real-time so high data, can using Spark Streaming Message Processing streaming frame come to data into
Row processing.Specifically include: behavioral data is accumulative, and all kinds of behavioral datas are summarized according to category and gather each statistical time section
(such as 1 minute dynamics, 5 minutes dynamics, 30 minutes dynamics, 1 hour dynamics, the statistics of all behavioral datas all can with hair
Add up in the time corresponding time dynamics section of raw behavior) in form A and T, this partial data needs to be buffered, and in real time
Variation；Driving model calculating is retrieved as a result, to extract real-time behavioral statistics data A and T during this, in advance by A and T input
The model M of the correspondence category first loaded, obtains result R, and on condition that transaction data has been marked can start to calculate.When
So, if R has been calculated within the specific period, checkout result is directly returned, will not be recalculated.For example,
If user U has browsed commodity S, " whether browse lower single commodity " this dimension will be marked as 1, otherwise always 0,
Its relevant dimension is similar；For such as whether to the commodity placed an order this dimension for checking user history information by inquiry go through
History data are deposited in the buffer in advance；Time dynamics can be accumulated to for browsing lower single this kind of real-time accumulated dimension of commodity total degree
In, every execution one-off, corresponding value can add 1；It is all to place an order, cancel time point also and be saved in the buffer, until U
Browse action has occurred after cancelling an order and outside decision-making period, this movement can trigger decision.It looks into the buffer first
User U is ask to the true and false judgement feature (whether browsing the similar characteristics such as lower single commodity) of category P, secondly inquires user in the buffer
U finally inquires user U to category P's to the historical data feature (whether buying, if cancelled) of category P in the buffer
Real-time accumulated feature, the period determined by the time that key operations generate (browsing time first time, lower single time, cancel in this world,
The time browsed for the first time after cancellation), the behavior cumulative data of corresponding period is asked under quartile characteristic, such as 12 points
Single to 13 points are cancelled 1 hour altogether, then quartile is 12:00-12:15,12:15-12:30,12:30-12:45,12 respectively:
45-13:00 can deposit in caching inquiry same day the 144th 5 minutes for the data of 12:00-12:15, the 145th 5 minutes,
146th accumulative behavioral data summation in five minutes, other three periods are similar).In this way, we will in triggering decision point
All data in relation to A and T deposit inquiry in caching, cumulative have just obtained this feature of user U about category P
Lower forms data and browsing data for directly generating for transaction system etc., lack real-time relative to behavioral data amount
Height, therefore the software for directly docking MQ message transmission carries out real-time consumption.There are three effects for this partial data: time of the act data
Accumulative, this partial data is related to placing an order, cancels etc. the data of material time points, and strong real-time is mainly related to time interval
Behavioral data thus add up generate；Mark calculate beginning (user for only placing an order and cancelling an order is likely to be intervened,
If user, without playing list also not cancel order, just without intervening, accumulative behavioral data will not carry out for that
Call), only (such as behavior interval cannot be too short twice, otherwise after this partial data reaches and has reached certain condition
This kind of user is most likely robot brush list, to filter the user that this part non-human normal operating can come out), behavioral data drives
Dynamic model calculating is possible to be performed；The model calculation is reset, it may by the result R that user behavior data driving generates
It can be reset, for example be marked as repairable user and place an order again, then calculated result R should be erased.
The process of behavioral data and time of the act data is extracted below with reference to Fig. 2 introduction.
Fig. 2 shows the flow diagrams that the disclosure extracts one embodiment of behavioral data and time of the act data.
As shown in Fig. 2, including step S2024~step S2026 in the present embodiment.
In step S2024, using Message Processing streaming frame to user about target related data carry out in batches
Processing.
In step S2026, batch processing result is polymerize in service layer, obtains behavior of the user about target
Data and time of the act data.
Fig. 3 shows the flow chart of data processing schematic diagram of Stream Processing frame, and Fig. 4 shows Stream Processing frame batch processing
The flow diagram of operation.Spark Streaming Stream Processing frame is that streaming computing is resolved into a series of short and small batches
Operation is handled, batch processing engine is Spark computing engines, by the input data of Spark Streaming according to division batch
Time interval (such as 1 second) is divided into sectional data, and every one piece of data is all converted into the elasticity distribution formula data set in Spark
Then RDD will become being directed in Spark to RDD in Spark Streaming to the conversion operation of discrete data DStream
Conversion operation, by RDD by operation become intermediate result save in memory.Such as during behavior data accumulation,
For the data of 1 minute dynamics, in spark, 60/5=12 batch can be divided into, 12 batches successively reach at any time.
First batch data aggregate (inside merges, for example has the same user here to two browsings of same commodity record,
User is so merged into goods browse twice) handled after, we by result there are waiting next batch in memory, under
It is merged after one batch wise polymerization has been handled with the result of a upper batch, until 12 batches have all been handled 12 batches
Polymerization result polymerize, just obtained the data of one minute dynamics.It is after intermediate result, that is, each batch processed as a result, in
Between result by spark once be submitted to caching redis in.Entire streaming computing can be to intermediate result according to the demand of business
It is overlapped, or external equipment is arrived in storage.
Real-time behavioral data is handled using Spark Streaming mainly to have the advantage that
(1) zmodem: above-mentioned business determines that data must have calamity standby, in Spark each RDD be one not
The data set that variable distribution can rerun, records deterministic operation inheritance, as long as input data is fault tolerant
, then the subregion error of any one RDD or the unavailable original input data that is all available with pass through conversion operation weight
Newly calculate；
(2) real-time is good: streaming computing is resolved into multiple Spark operations by Spark Streaming, for each number of segment
According to processing all can the directed acyclic graph by the genetic connection between data set in Spark operation decompose and the task of Spark
Collect scheduling process；By time interval be set as 5 seconds can near real-time intervention, meet business need；
(3) scalability is strong, and handling capacity is high: Spark at present on Stream Processing platform can linear expansion to 1000
Node, can be with the data volume of the delay disposal 6GB/s of several seconds, and the Storm computing engines of throughput ratio prevalence are 2~5 times high；
(4) parallel processing mode: streaming frame determines that its data processing mode is parallel processing.For the time interval phase
Interior data can submit Spark operation to achieve the effect that data segment parallel processing by whole, improve data processing
Speed, it is primary to submit intermediate result caching, it reduces for the interactive operation between the various components such as caching, persistence.
For being related to real-time accumulated batch mode, since current state is related to historic state, traditional technology
Realization is to store the calculated result of present lot, and next batch goes the last state of active inquiry.This system devises intermediate state
The key technology of caching, the aggregate-value of present lot are directly buffered in each task node and (i.e. execute in spark parallel after calculating
The node excutor of data processing), next batch can be calculated directly using current value, avoid frequent group
Interactive operation between part.
The information-pushing method of another embodiment of the disclosure is introduced below with reference to Fig. 5.
Fig. 5 shows the flow diagram of the information-pushing method of the disclosure another embodiment.As shown in figure 5, in Fig. 1
On the basis of illustrated embodiment, the information-pushing method in the present embodiment further includes step S501: using cancelling an order after placing an order
Buy back the user of target again, place an order after cancel an order and do not buy back behavioral data and time of the act of the user about target of target
Data are trained deep learning neural network, and deep learning neural network is enabled to predict whether user will buy back mesh
Mark.
The embodiment of the present disclosure use machine learning algorithm specifically can be logistic regression, support vector machines, GBDT,
XGBOOST and BP neural network model, wherein the effect of BP neural network model is best.
After Tables 1 and 2 shows whether target will be bought back using deep learning neural network prediction user, preferential letter is carried out
Animation effect is retrieved in breath push.As it can be seen from table 1 certain category commodity is once retrieved in activity, intervention group compares control group
24.73% is improved in terms of repurchase rate, favor information utilization rate promotes 169% with daily compared with dynamics moving average,
Utilization rate be substantially improved illustrate deep learning neural network can Accurate Prediction user whether will buy back commodity.
Table 1
Table 2
The information push-delivery apparatus of an embodiment of the present disclosure is introduced below with reference to Fig. 6.
Fig. 6 shows the structural schematic diagram of the information push-delivery apparatus of an embodiment of the present disclosure.As shown in fig. 6, this implementation
Example information push-delivery apparatus 60 include:
Data extraction module 602, be configured as extract user about the target cancelled an order after placing an order behavioral data with
And time of the act data；
User in predicting module 604 is configured as using deep learning neural network to behavioral data and time of the act number
According to being handled, to predict whether user will buy back target；
Info push module 606, is configured as in the case where predicting that user does not buy back target, pushes target to user
Relevant information.
Above-described embodiment can accurately be predicted to cancel an order after placing an order not buying back the user of target in real time, relative to extensive
The selection information for formula of casting net pushes user, and the present embodiment can make relatively reasonable fining to target user and divide, to not
Different information push strategies is made with user, the accuracy of information push is improved, is mentioned while saving information push cost
High user buys back conversion ratio.At the same time it can also determine the material time point of information push using above-described embodiment, cancellation is ordered
Single user can provide the natural change-over period of the reasonable category rank of comparison, both in view of the complicated heart of user psychology variation
Decision factor is managed, also using having there is experience to predict the following time buying of user.
In some embodiments, deep learning neural network meets preset condition in behavioral data and time of the act data
In the case where, prediction user does not buy back target.
In some embodiments, behavioral data includes browsing data and order data of the user about target.
In some embodiments, browsing data include that the following contents at least one of is worked as: whether browsing objective, whether clear
Look at same category target, the total degree of browsing objective, browsing with category target total degree, whether click the commodity details page of target
Master map, whether click target commodity details page comment, whether have brush select difference comment behavior, whether have brush choose comment behavior, whether
Click the big figure of comment, whether click specifications and models, whether click inventory bookmark, whether browse commodity question and answer area, whether
It puts question in commodity question and answer area.
In some embodiments, order data includes that the following contents at least one of is worked as: whether target is carried out plus is purchased,
Whether to target subtract purchase, whether target paid close attention to, whether once placed an order to target, whether once cancelled an order to target,
Whether once target was successfully bought, whether once placed an order to same category target, whether once cancelled an order, whether once to same category target
Said target is successfully bought.
In some embodiments, time of the act data include that the following contents at least one of is worked as: the same day browses to for the first time
The number of browsing objective, the same day are browsed to for the first time in the different time intervals for the period that places an order in different time intervals during placing an order
Browse with category target number, place an order cancel an order during different time intervals in browsing objective number, place an order to
In different time intervals during cancelling an order browsing with category target number, browse to for the first time the time interval to place an order, under
It is single to the time interval cancelled an order, cancel an order after browsing objective for the first time time interval.
In some embodiments, information push-delivery apparatus 60 further includes deep learning neural metwork training module 601, is configured
For using cancel an order after placing an order and buy back the user of target, place an order after cancel an order the user for not buying back target about target
Behavioral data and time of the act data, are trained deep learning neural network, enable deep learning neural network
Whether prediction user will buy back target.
In some embodiments, data extraction module 602 is configured as: using Message Processing streaming frame to user about
Target related data carry out batch processing；Batch processing result is polymerize in service layer, obtains user about mesh
Target behavioral data and time of the act data.
For being related to real-time accumulated batch mode, since current state is related to historic state, traditional technology
Realization is to store the calculated result of present lot, and next batch goes the last state of active inquiry.This system devises intermediate state
The key technology of caching, the aggregate-value of present lot are directly buffered in each task node after calculating, and next batch can be straight
It connects using current value and is calculated, avoid the interactive operation between frequent component.
Fig. 7 shows the structural schematic diagram of the information push-delivery apparatus of the disclosure another embodiment.As shown in fig. 7, the reality
The information push-delivery apparatus 70 for applying example includes: memory 710 and the processor 720 for being coupled to the memory 710, processor 720
It is configured as executing the information-pushing method in any one aforementioned embodiment based on the instruction being stored in memory 710.
Wherein, memory 710 is such as may include system storage, fixed non-volatile memory medium.System storage
Device is for example stored with operating system, application program, Boot loader (Boot Loader) and other programs etc..
Information push-delivery apparatus 70 can also include input/output interface 730, network interface 740, memory interface 770 etc..This
It can for example be connected by bus 760 between a little interfaces 730,740,750 and memory 710 and processor 720.Wherein, defeated
Enter output interface 730 and provides connecting interface for input-output equipment such as display, mouse, keyboard, touch screens.Network interface 740
Connecting interface is provided for various networked devices.The external storages such as memory interface 940 is SD card, USB flash disk provide connecting interface.
The disclosure further includes a kind of computer readable storage medium, is stored thereon with computer instruction, and the instruction is processed
Device realizes the information-pushing method in any one aforementioned embodiment when executing.
It should be understood by those skilled in the art that, embodiment of the disclosure can provide as method, system or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the disclosure
Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the disclosure, which can be used in one or more,
The calculating implemented in non-transient storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) can be used
The form of machine program product.
The disclosure is referring to method, the process of equipment (system) and computer program product according to the embodiment of the present disclosure
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions can also be loaded into computer or other programmable data processing devices, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
The foregoing is merely the preferred embodiments of the disclosure, not to limit the disclosure, all spirit in the disclosure and
Within principle, any modification, equivalent replacement, improvement and so on be should be included within the protection scope of the disclosure.
Claims (18)
1. a kind of information-pushing method, comprising:
Extract behavioral data and time of the act data of the user about the target cancelled an order after placing an order；
The behavioral data and time of the act data are handled using deep learning neural network, to predict the user
Whether the target will be bought back；
In the case where predicting that the user does not buy back the target, Xiang Suoshu user pushes the relevant information of the target.
2. information-pushing method as described in claim 1, wherein the deep learning neural network the behavioral data with
And in the case that time of the act data meet preset condition, predict that the user does not buy back the target.
3. information-pushing method as described in claim 1, wherein the behavioral data includes the user about the target
Browsing data and order data.
4. information-pushing method as claimed in claim 3, wherein the browsing data include at least one in the following contents
Kind:
Whether browse the target, whether browse same category target, the total degree of the browsing target, browsing with category target
Total degree, the commodity details page master map for whether clicking the target, whether click the target commodity details page comment, whether
There is brush that difference is selected to comment behavior, whether there is brush to choose to comment behavior, whether click the big figure of comment, whether click specifications and models, whether click
Whether whether inventory bookmark browse commodity question and answer area, put question in commodity question and answer area.
5. information-pushing method as claimed in claim 3, wherein the order data includes at least one in the following contents
Kind:
Whether the target is carried out plus is purchased, whether to the target subtract purchase, whether the target is paid close attention to, whether
Once it placed an order, whether once cancelled an order to the target, whether once the target successfully bought, whether once to same to the target
Whether whether category target place an order, once cancelled an order, once successfully bought to said target to same category target.
6. information-pushing method as claimed in claim 3, wherein the time of the act data include in the following contents extremely
Few one kind:
The same day browses to the number of the interior browsing target of different time intervals for the period that places an order for the first time, the same day browses to down for the first time
The number with category target is browsed in different time intervals during list, in the different time intervals during cancelling an order that place an order
Browse the number of the target, number, head of the browsing with category target in the different time intervals during cancelling an order that place an order
It is secondary browse to the time interval to place an order, place an order to the time interval cancelled an order, cancel an order to after cancelling an order browses for the first time
The time interval of the target.
7. information-pushing method as described in claim 1, wherein the information-pushing method further include:
Using cancel an order after placing an order and buy back the user of the target, place an order after cancel an order and do not buy back the user of the target
About the behavioral data and time of the act data of the target, the deep learning neural network is trained, so that institute
Stating deep learning neural network can predict whether the user will buy back the target.
8. information-pushing method as described in claim 1, wherein the extraction user is about the target cancelled an order after placing an order
Behavioral data and time of the act data include:
Batch processing is carried out about the related data of the target to the user using Message Processing streaming frame；
Batch processing result is polymerize in service layer, obtains behavioral data and row of the user about the target
For time data.
9. a kind of information push-delivery apparatus, comprising:
Data extraction module, when being configured as extracting behavioral data and behavior of the user about the target cancelled an order after placing an order
Between data；
User in predicting module, be configured as using deep learning neural network to the behavioral data and time of the act data into
Row processing, to predict whether the user will buy back the target；
Info push module is configured as in the case where predicting that the user does not buy back the target, Xiang Suoshu user's push
The relevant information of the target.
10. information push-delivery apparatus as claimed in claim 9, wherein the deep learning neural network is in the behavioral data
And in the case that time of the act data meet preset condition, predict that the user does not buy back the target.
11. information push-delivery apparatus as claimed in claim 9, wherein the behavioral data includes the user about the mesh
Target browses data and order data.
12. information push-delivery apparatus as claimed in claim 11, wherein the browsing data include in the following contents at least
It is a kind of:
Whether browse the target, whether browse same category target, the total degree of the browsing target, browsing with category target
Total degree, the commodity details page master map for whether clicking the target, whether click the target commodity details page comment, whether
There is brush that difference is selected to comment behavior, whether there is brush to choose to comment behavior, whether click the big figure of comment, whether click specifications and models, whether click
Whether whether inventory bookmark browse commodity question and answer area, put question in commodity question and answer area.
13. information push-delivery apparatus as claimed in claim 11, wherein the order data include in the following contents at least
It is a kind of:
Whether the target is carried out plus is purchased, whether to the target subtract purchase, whether the target is paid close attention to, whether
Once it placed an order, whether once cancelled an order to the target, whether once the target successfully bought, whether once to same to the target
Whether whether category target place an order, once cancelled an order, once successfully bought to said target to same category target.
14. information push-delivery apparatus as claimed in claim 11, wherein the time of the act data include in the following contents
It is at least one:
The same day browses to the number of the interior browsing target of different time intervals for the period that places an order for the first time, the same day browses to down for the first time
The number with category target is browsed in different time intervals during list, in the different time intervals during cancelling an order that place an order
Browse the number of the target, number, head of the browsing with category target in the different time intervals during cancelling an order that place an order
It is secondary browse to the time interval to place an order, place an order to the time interval cancelled an order, cancel an order to after cancelling an order browses for the first time
The time interval of the target.
15. information push-delivery apparatus as claimed in claim 9, wherein the information push-delivery apparatus further includes deep learning nerve
Network training module, be configured as using cancel an order after placing an order and buy back the user of the target, place an order after cancel an order not
Behavioral data and time of the act data of the user for buying back the target about the target, to the deep learning nerve net
Network is trained, and the deep learning neural network is enabled to predict whether the user will buy back the target.
16. information push-delivery apparatus as claimed in claim 9, wherein the data extraction module is configured as:
Batch processing is carried out about the related data of the target to the user using Message Processing streaming frame；
Batch processing result is polymerize in service layer, obtains behavioral data and row of the user about the target
For time data.
17. a kind of information push-delivery apparatus, comprising:
Memory；And
It is coupled to the processor of the memory, the processor is used for the instruction based on storage in the memory, executes
Such as information-pushing method described in any item of the claim 1 to 8.
18. a kind of computer readable storage medium, wherein the computer-readable recording medium storage has computer instruction, institute
It states and realizes such as information-pushing method described in any item of the claim 1 to 8 when instruction is executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810304124.7A CN110348928A (en) | 2018-04-08 | 2018-04-08 | Information-pushing method, device and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810304124.7A CN110348928A (en) | 2018-04-08 | 2018-04-08 | Information-pushing method, device and computer readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110348928A true CN110348928A (en) | 2019-10-18 |
Family
ID=68172955
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810304124.7A Pending CN110348928A (en) | 2018-04-08 | 2018-04-08 | Information-pushing method, device and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110348928A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112084447A (en) * | 2020-08-17 | 2020-12-15 | 北京字节跳动网络技术有限公司 | Data distribution method, device, medium and electronic equipment |
| CN112101611A (en) * | 2020-07-31 | 2020-12-18 | 重庆锐云科技有限公司 | Real estate customer buyback time prediction method, server and storage medium |
| CN112784147A (en) * | 2019-11-04 | 2021-05-11 | 阿里巴巴集团控股有限公司 | Information processing method, device, equipment and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017016119A1 (en) * | 2015-07-29 | 2017-02-02 | 百度在线网络技术（北京）有限公司 | Recommendation method, apparatus and device based on e-commerce platform, and storage medium |
| CN107437203A (en) * | 2017-08-07 | 2017-12-05 | 北京京东尚科信息技术有限公司 | Information-pushing method, device, electronic installation and computer-readable medium |
| CN107481093A (en) * | 2017-07-21 | 2017-12-15 | 北京京东尚科信息技术有限公司 | Personalized shop Forecasting Methodology and device |
-
2018
- 2018-04-08 CN CN201810304124.7A patent/CN110348928A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017016119A1 (en) * | 2015-07-29 | 2017-02-02 | 百度在线网络技术（北京）有限公司 | Recommendation method, apparatus and device based on e-commerce platform, and storage medium |
| CN107481093A (en) * | 2017-07-21 | 2017-12-15 | 北京京东尚科信息技术有限公司 | Personalized shop Forecasting Methodology and device |
| CN107437203A (en) * | 2017-08-07 | 2017-12-05 | 北京京东尚科信息技术有限公司 | Information-pushing method, device, electronic installation and computer-readable medium |
Non-Patent Citations (1)
| Title |
|---|
| 张鹏程 等: "基于消费者购物记录的商品推荐去重方案", 软件工程, vol. 21, no. 3, 5 March 2018 (2018-03-05), pages 16 - 19 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112784147A (en) * | 2019-11-04 | 2021-05-11 | 阿里巴巴集团控股有限公司 | Information processing method, device, equipment and system |
| CN112101611A (en) * | 2020-07-31 | 2020-12-18 | 重庆锐云科技有限公司 | Real estate customer buyback time prediction method, server and storage medium |
| CN112101611B (en) * | 2020-07-31 | 2022-11-18 | 重庆锐云科技有限公司 | Real estate customer buyback time prediction method, server and storage medium |
| CN112084447A (en) * | 2020-08-17 | 2020-12-15 | 北京字节跳动网络技术有限公司 | Data distribution method, device, medium and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11526485B2 (en) | System and method for improved data consistency in data systems including dependent algorithms | |
| Golfarelli et al. | Designing what-if analysis: towards a methodology | |
| US8909644B2 (en) | Real-time adaptive binning | |
| JP4717945B2 (en) | Business analysis program and business analysis device | |
| US20150371244A1 (en) | Forecasting information technology workload demand | |
| US20200057918A1 (en) | Systems and methods for training artificial intelligence to predict utilization of resources | |
| US9123052B2 (en) | Marketing model determination system | |
| CN104035926B (en) | A kind of dispensing of internet information and system | |
| CN106796527A (en) | The system and method for the selection based on price and performance optimization cloud service | |
| US20150006292A1 (en) | Promotion scheduling management | |
| CN105869019A (en) | Method and apparatus for predicting goods price | |
| CN110348928A (en) | Information-pushing method, device and computer readable storage medium | |
| US11151012B2 (en) | Predictive reserved instance for hyperscaler management | |
| Antomarioni et al. | Data-driven decision support system for managing item allocation in an ASRS: A framework development and a case study | |
| CN110109901A (en) | The method and apparatus for screening target object | |
| Cakmak et al. | Accurate Prediction of Advertisement Clicks based on Impression and Click-Through Rate using Extreme Gradient Boosting. | |
| CN111242356A (en) | Wealth trend prediction method, device, equipment and storage medium | |
| US20140129269A1 (en) | Forecasting Business Entity Characteristics Based on Planning Infrastructure | |
| US20060047562A1 (en) | Method and apparatus for planning marketing scenarios | |
| CN103744938A (en) | Method and device for generating user page corresponding to target system | |
| US20230342699A1 (en) | Systems and methods for modeling and analysis of infrastructure services provided by cloud services provider systems | |
| CN113692598A (en) | System and method for automatic training and prediction of garment usage models | |
| Tan et al. | The impact of feature selection: A data‐mining application in direct marketing | |
| US20210166177A1 (en) | Adjusting a master build plan based on events using machine learning | |
| CN112686705A (en) | Method and device for predicting sales effect data and electronic equipment |
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 |