CN112381607A - Network commodity ordering method, device, equipment and medium - Google Patents
Network commodity ordering method, device, equipment and medium Download PDFInfo
- Publication number
- CN112381607A CN112381607A CN202011264017.XA CN202011264017A CN112381607A CN 112381607 A CN112381607 A CN 112381607A CN 202011264017 A CN202011264017 A CN 202011264017A CN 112381607 A CN112381607 A CN 112381607A
- Authority
- CN
- China
- Prior art keywords
- commodity
- click
- sequence
- rate
- conversion rate
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000006243 chemical reaction Methods 0.000 claims abstract description 178
- 230000006399 behavior Effects 0.000 claims abstract description 43
- 238000012163 sequencing technique Methods 0.000 claims description 17
- 238000004590 computer program Methods 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 4
- 238000004891 communication Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
Classifications
-
- 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]
- G06Q30/0641—Shopping interfaces
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a method, a device, equipment and a medium for sorting network commodities. The method comprises the following steps: acquiring user behavior data and commodity exposure data; the user behavior data comprises a purchase intention sequence and a historical click sequence; inputting user behavior data and commodity exposure data into a multi-target estimation model constructed based on a multi-task learning framework to obtain corresponding commodity conversion rate; the multi-target estimation model comprises a commodity click rate learning task and a commodity conversion rate learning task; sharing the embedded layer parameters and partial attention parameters between the commodity click rate learning task and the commodity conversion rate learning task; inputting the historical click sequence and the commodity exposure data into a click rate estimation model to obtain the corresponding commodity click rate; the click conversion probability of the commodities is determined based on the commodity conversion rate and the commodity click rate, the commodities are sorted based on the click conversion probability, the commodities can be sorted according to the comprehensive quality of the commodities, and then the commodity conversion rate and the total amount of finished goods are improved.
Description
Technical Field
The invention relates to the field of online shopping, in particular to a method, a device, equipment and a medium for ordering online commodities.
Background
Currently, the Click Rate (i.e., CTR, Click Through Rate) of the goods in the cyber mall is not completely proportional to the Conversion Rate (i.e., CVR, Conversion Rate), which indicates the probability that the exposed goods are clicked and the Conversion Rate indicates the probability that the clicked goods are purchased. For example, the cover design of a certain commodity is graceful, the commodity click rate is high, but the actual quality of the commodity does not accord with the commodity description, so that the user cannot buy the commodity after clicking; or the cover of a certain commodity is not attractive enough and the probability of being clicked by the user is low, but the performances of other indexes such as the quality, the evaluation and the historical sales volume of the commodity are good, and the probability of purchasing the commodity by the user is high after the user clicks and enters the commodity detail page and observes a series of historical performances of the commodity. Therefore, how to prioritize high-quality commodities to maximize the amount of trades is a major current problem.
In the prior art, the conversion rate of a commodity is predicted through a model, but the conversion rate prediction model predicts the probability of the commodity being converted on the assumption that the commodity is clicked, namely, the traditional conversion rate prediction model usually takes click data as a training set, wherein the clicked commodity but not converted is a negative example, and the clicked and converted commodity is a positive example, so that the problem of sample deviation exists, and the generalization capability of the model is reduced. In the prior art, the Click and Conversion probability (i.e. CTCVR, Click Through & Conversion Rate) of the commodity is directly predicted Through the ESMM network structure, but in practical application, when online sorting is performed according to the output CTCVR score, the Click Rate drops greatly, the Click Rate of the commodity is reduced, and further the Conversion Rate and the total volume of the commodity are affected.
Disclosure of Invention
In view of the above, the present invention provides a method, an apparatus, a device and a medium for sorting network commodities, which can sort the commodities according to their comprehensive qualities, thereby improving commodity conversion rate and total volume. The specific scheme is as follows:
in a first aspect, the present application discloses a method for ordering network commodities, comprising:
acquiring user behavior data and commodity exposure data; the user behavior data comprises a purchase intention sequence and a historical click sequence;
inputting the user behavior data and the commodity exposure data into a multi-target estimation model constructed based on a multi-task learning framework to obtain corresponding commodity conversion rate; the multi-target estimation model comprises a commodity click rate learning task and a commodity conversion rate learning task; sharing an embedded layer parameter and a part of attention parameters between the commodity click rate learning task and the commodity conversion rate learning task;
inputting the historical click sequence and the commodity exposure data into a click rate estimation model to obtain a corresponding commodity click rate;
determining the click conversion probability of the commodities based on the commodity conversion rate and the commodity click rate, and sequencing the commodities according to the click conversion probability.
Optionally, the acquiring the user behavior data includes:
constructing an operation sequence based on the purchase adding operation and the collection operation of the user and the click operation again of the history footprints and/or the favorites and/or the shopping carts;
sequencing the operation sequence according to the sequence of each operation from near to far from the current time to obtain the purchase intention sequence;
and constructing the historical click sequence based on the click operation of the user.
Optionally, the inputting the user behavior data and the commodity exposure data into a multi-target estimation model constructed based on a multi-task learning framework includes:
inputting the commodity exposure data into the commodity click rate learning task and the commodity conversion rate learning task;
configuring non-shared parameters for the purchase intention sequence, and inputting the purchase intention sequence into the commodity click rate learning task and the commodity conversion rate learning task;
and configuring sharing parameters for the historical click sequence, and inputting the historical click sequence into the commodity click rate learning task and the commodity conversion rate learning task so that the commodity click rate learning task and the commodity conversion rate learning task share the attention parameters of the historical click sequence.
Optionally, the process of constructing the multi-target estimation model includes:
and constructing a double-task learning model sharing the embedded layer parameters and the hidden layer parameters based on the ESMM network structure to obtain the multi-target estimation model.
Optionally, the inputting the user behavior data and the commodity exposure data into a multi-target estimation model constructed based on a multi-task learning framework includes:
acquiring commodity characteristics of commodities; the commodity characteristics comprise any one or more of commodity categories, commodity prices, commodity sales amounts and commodity praise amounts;
the commodity features correspond to commodities in the historical click sequence to obtain a historical click sequence containing the commodity features;
and inputting the historical click sequence containing the commodity characteristics, the purchase intention sequence and the commodity exposure data into the multi-target estimation model.
Optionally, the determining the click conversion probability of the commodity based on the commodity conversion rate and the commodity click rate includes:
carrying out constraint operation on the commodity conversion rate by using a preset value range control function to obtain a constrained commodity conversion rate;
and multiplying the commodity conversion rate after constraint with the commodity click rate to obtain the click conversion probability.
In a second aspect, the present application discloses a network commodity sorting apparatus, including:
the data acquisition module is used for acquiring user behavior data and commodity exposure data; the user behavior data comprises a purchase intention sequence and a historical click sequence;
the conversion rate determining module is used for inputting the user behavior data and the commodity exposure data into a multi-target estimation model constructed based on a multi-task learning framework to obtain corresponding commodity conversion rate; the multi-target estimation model comprises a commodity click rate learning task and a commodity conversion rate learning task; sharing an embedded layer parameter and a part of attention parameters between the commodity click rate learning task and the commodity conversion rate learning task;
the click rate determining module is used for inputting the historical click sequence and the commodity exposure data into a click rate estimation model to obtain a corresponding commodity click rate;
and the click conversion probability determining module is used for determining the click conversion probability of the commodities based on the commodity conversion rate and the commodity click rate and sequencing the commodities according to the click conversion probability.
Optionally, the conversion rate determining module further includes:
a characteristic acquisition unit for acquiring commodity characteristics of a commodity; the commodity characteristics comprise any one or more of commodity categories, commodity prices, commodity sales amounts and commodity praise amounts;
the historical click sequence construction unit is used for corresponding the commodity characteristics with commodities in the historical click sequence to obtain a historical click sequence containing the commodity characteristics;
and the data input unit is used for inputting the historical click sequence containing the commodity characteristics, the purchase intention sequence and the commodity exposure data into the multi-target estimation model.
In a third aspect, the present application discloses an electronic device, comprising:
a memory for storing a computer program;
and the processor is used for executing the computer program to realize the network commodity sequencing method.
In a fourth aspect, the present application discloses a computer readable storage medium for storing a computer program; wherein the computer program when executed by the processor implements the aforementioned network commodity ordering method.
The application discloses a network commodity sequencing method, which comprises the following steps: the method comprises the steps of obtaining user behavior data and commodity exposure data; the user behavior data comprises a purchase intention sequence and a historical click sequence; inputting the user behavior data and the commodity exposure data into a multi-target estimation model constructed based on a multi-task learning framework to obtain corresponding commodity conversion rate; the multi-target estimation model comprises a commodity click rate learning task and a commodity conversion rate learning task; sharing an embedded layer parameter and a part of attention parameters between the commodity click rate learning task and the commodity conversion rate learning task; inputting the historical click sequence and the commodity exposure data into a click rate estimation model to obtain a corresponding commodity click rate; determining the click conversion probability of the commodities based on the commodity conversion rate and the commodity click rate, and sequencing the commodities according to the click conversion probability.
It can be seen that a purchase intention sequence, a history click sequence and commodity exposure data are input into a multi-target pre-estimation model comprising a commodity click rate learning task and a commodity conversion rate learning task, embedded layer parameters and partial attention parameters are shared between the commodity click rate learning task and the commodity conversion rate learning task, so that all samples, namely commodity exposure data are pre-estimated, more accurate commodity conversion rate can be obtained through co-learning between the commodity click rate learning task and the commodity conversion rate learning task, finally, the click conversion probability of commodities is determined according to the commodity conversion rate obtained based on the multi-target pre-estimation model and the commodity click rate obtained based on the click rate pre-estimation model, and the commodities are sorted according to the click conversion probability. Therefore, the optimal sorting sequence of the commodities can be determined, the commodity conversion rate is improved while the commodity click rate is guaranteed, and the total amount of the finished products is further improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
Fig. 1 is a flowchart of a network commodity sorting method provided in the present application;
FIG. 2 is a schematic diagram of a partial network architecture for a single task provided herein;
FIG. 3 is a flowchart of a specific network commodity ordering method provided by the present application;
fig. 4 is a schematic structural diagram of a network commodity sorting apparatus provided in the present application;
fig. 5 is a block diagram of an electronic device provided in the present application.
Detailed Description
In the prior art, the conversion rate of a commodity is predicted through a model, but the conversion rate prediction model predicts the probability of the commodity being converted on the assumption that the commodity is clicked, namely, the traditional conversion rate prediction model usually takes click data as a training set, wherein the clicked commodity but not converted is a negative example, and the clicked and converted commodity is a positive example, so that the problem of sample deviation exists, and the generalization capability of the model is reduced. In order to overcome the technical problems, the application provides a network commodity sequencing method utilizing multi-model combination and multi-task learning, which can sequence commodities according to the comprehensive quality of the commodities, and further improve the commodity conversion rate and the total amount of finished goods.
The embodiment of the application discloses a network commodity ordering method, and as shown in fig. 1, the method can comprise the following steps:
step S11: acquiring user behavior data and commodity exposure data; the user behavior data includes a sequence of purchasing intentions and a sequence of historical clicks.
In this embodiment, first, user behavior data and commodity exposure data are obtained, where the user behavior data is determined according to user behavior and includes a purchase intention sequence and a history click sequence.
In this embodiment, the acquiring user behavior data may include: constructing an operation sequence based on the purchase adding operation and the collection operation of the user and the click operation again of the history footprints and/or the favorites and/or the shopping carts; sequencing the operation sequence according to the sequence of each operation from near to far from the current time to obtain the purchase intention sequence; and constructing the historical click sequence based on the click operation of the user. The purchase intention sequence is sorted according to the sequence of each operation from near to far from the current time, the operation behavior closer to the current time is in front of the relative position in the sequence, and correspondingly, the operation behavior farther from the current time is in back of the relative position in the sequence. It will be appreciated that the user's collection, click and buy operations for an item, and then click again from the history footprint and/or favorites and/or shopping cart, may indicate that the user's intent to purchase the item is strong. Therefore, model learning and prediction are carried out by taking the operation behavior path as a purchase intention sequence, and more effective information related to deal can be introduced.
Step S12: inputting the user behavior data and the commodity exposure data into a multi-target estimation model constructed based on a multi-task learning framework to obtain corresponding commodity conversion rate; the multi-target estimation model comprises a commodity click rate learning task and a commodity conversion rate learning task; and sharing an embedded layer parameter and a part of attention parameters between the commodity click rate learning task and the commodity conversion rate learning task.
In the embodiment, after the user behavior data and the commodity exposure data are obtained, the data are input into a multi-target estimation model constructed based on a multi-task learning framework, and then the corresponding commodity conversion rate of the commodity is obtained through model prediction; the multi-target estimation model comprises a commodity click rate learning task and a commodity conversion rate learning task, the commodity click rate learning is used for predicting the click rate of commodities, the commodity conversion rate learning task is used for predicting the conversion rate of the commodities, and embedded layer (namely embedding layer) parameters and partial attention parameters are shared between the commodity click rate learning task and the commodity conversion rate learning task.
It will be appreciated that the user's sequence of buying intentions and the user's sequence of historical clicks are input into the model simultaneously for training, wherein both the user's sequence of historical clicks and the sequence of buying intentions are fed to both tasks, and both tasks use conventional attention mechanism structures to host these sequences. However, the attention parameter of the historical click sequence is shared by two tasks, and the attention parameter of the purchase intention sequence is not shared between the two tasks and is learned by the two tasks. And finally, obtaining the final conversion rate from the commodity conversion rate learning task by fusing the attention parameter output results of the historical click sequence and the deal intention sequence, an MLP full-link layer, dimension reduction, a nonlinear activation function and the like. It can be understood that the commodity conversion rate learning task can assist training through parameters learned by the commodity click rate learning task, and between the commodity click rate learning task and the commodity conversion rate learning task, the learning capability of the commodity conversion rate learning task can be improved by sharing the embedded layer parameters and the attention parameters of the historical click sequence, so that the commodity conversion rate more suitable for practical application is obtained.
In this embodiment, the inputting the user behavior data and the commodity exposure data into a multi-target estimation model constructed based on a multi-task learning framework may include: inputting the commodity exposure data into the commodity click rate learning task and the commodity conversion rate learning task; configuring non-shared parameters for the purchase intention sequence, and inputting the purchase intention sequence into the commodity click rate learning task and the commodity conversion rate learning task; and configuring sharing parameters for the historical click sequence, and inputting the historical click sequence into the commodity click rate learning task and the commodity conversion rate learning task so that the commodity click rate learning task and the commodity conversion rate learning task share the attention parameters of the historical click sequence. It is to be appreciated that the attention parameter of the historical click sequence may be shared between the item click rate learning task and the item conversion rate learning task, but not the purchase intent sequence, by configuring the non-sharing parameter for the purchase intent sequence and configuring the sharing parameter for the historical click sequence.
In this embodiment, the process of constructing the multi-target estimation model may include: and constructing a double-task learning model sharing the embedded layer parameters and the hidden layer parameters based on the ESMM network structure to obtain the multi-target estimation model. It can be understood that the original embedded layer sharing can be kept on the basis of the ESMM network structure, and meanwhile, the attention parameter sharing of the historical click sequence is newly added, so that the double-task learning model is obtained; wherein, a partial network structure of each task is as shown in fig. 2, the two tasks share the parameters of the embedding layer and the partial parameters in the hidden layer, that is, the parameters of the embedding layer and the attention parameters of the historical click sequence, and the parameters of the upper MLP full-link layer are separated and not shared.
Step S13: and inputting the historical click sequence and the commodity exposure data into a click rate estimation model to obtain the corresponding commodity click rate.
In the embodiment, the historical click sequence and the commodity exposure data are simultaneously input into a click rate estimation model, and the commodity click rate is obtained through the click rate estimation model; the click rate estimation model can be a Wide & deep FM model.
Step S14: determining the click conversion probability of the commodities based on the commodity conversion rate and the commodity click rate, and sequencing the commodities according to the click conversion probability.
In this embodiment, the click conversion probability of the commodity is determined according to the obtained commodity conversion rate and the commodity click rate, then the commodities are sorted according to the click conversion probability, and the commodities with high click conversion rate are sorted preferentially.
As can be seen from the above, in this embodiment, a purchase intention sequence, a history click sequence, and commodity exposure data are input into a multi-target pre-estimation model including a commodity click rate learning task and a commodity conversion rate learning task, and an embedded layer parameter and a part of attention parameters are shared between the commodity click rate learning task and the commodity conversion rate learning task, so that all samples, namely commodity exposure data are used for pre-estimation, and through co-learning between the commodity click rate learning task and the commodity conversion rate learning task, a relatively accurate commodity conversion rate can be obtained, and finally, a click conversion probability of a commodity is determined according to a commodity conversion rate obtained based on the multi-target pre-estimation model and a commodity click rate obtained based on the click rate pre-estimation model, and commodities are sorted according to the click conversion probability. Therefore, the optimal sorting sequence of the commodities can be determined, the commodity conversion rate is improved while the commodity click rate is guaranteed, and the total amount of the finished products is further improved.
The embodiment of the application discloses a specific network commodity ordering method, and as shown in fig. 3, the method may include the following steps:
step S21: acquiring user behavior data and commodity exposure data; the user behavior data includes a sequence of purchasing intentions and a sequence of historical clicks.
Step S22: acquiring commodity characteristics of commodities; the commodity characteristics comprise any one or more of commodity categories, commodity prices, commodity sales amounts and commodity praise amounts; and corresponding the commodity characteristics with the commodities in the historical click sequence to obtain the historical click sequence containing the commodity characteristics.
In the embodiment, the commodity characteristics of the commodities can be acquired simultaneously; and then, the commodity characteristics are corresponding to the commodities in the historical click sequence to obtain the historical click sequence containing the commodity characteristics. Specifically, the commodity features may be partitioned into 8 buckets, for example, the commodity price is partitioned into 8 sections, and price sections corresponding to different commodities are marked; wherein, the boundary of the sub-barrels can be set manually; after bucket separation, encoding each type of characteristics, and constructing a special imbedding matrix of each type of characteristics to serve as supplementary information of the historical click sequence; it will be appreciated that these item characteristics are also presented in a sequence and correspond one-to-one to each item in the historical click sequence, i.e., each item in the sequence contains categories, prices, sales, and likes. Specifically, these coded signature sequences can be concatenated with the embedding expression of the historical click sequences in the last dimension.
Therefore, by using the commodity characteristics as supplementary information of the historical click sequence, hidden information of the user, such as category preference, price preference, evaluation star rating or evaluation quantity of whether to pay attention to the commodity, commodity sales and popular preference degree, can be predicted according to the user behavior. Together with the historical click sequences, the information can help the model to learn commodity expression, and further enhance the capability of the commodity conversion rate task side.
Step S23: inputting the historical click sequence containing the commodity characteristics, the purchase intention sequence and the commodity exposure data into a multi-target pre-estimation model to obtain a corresponding commodity conversion rate; the multi-target estimation model comprises a commodity click rate learning task and a commodity conversion rate learning task; and sharing an embedded layer parameter and a part of attention parameters between the commodity click rate learning task and the commodity conversion rate learning task.
In the embodiment, the obtained historical click sequence, purchase intention sequence and commodity exposure data containing the commodity characteristics are input into a multi-target estimation model, and then the corresponding commodity conversion rate of the commodity is obtained through model prediction; the multi-target estimation model comprises a commodity click rate learning task and a commodity conversion rate learning task, the commodity click rate learning is used for predicting the click rate of commodities, the commodity conversion rate learning task is used for predicting the conversion rate of the commodities, and the embedded layer parameters and part of attention parameters are shared between the commodity click rate learning task and the commodity conversion rate learning task.
Step S24: and inputting the historical click sequence and the commodity exposure data into a click rate estimation model to obtain the corresponding commodity click rate.
Step S25: and carrying out constraint operation on the commodity conversion rate by using a preset value range control function to obtain the commodity conversion rate after constraint.
In this embodiment, after the commodity conversion rate is obtained, a preset value range control function is used to perform constraint operation on the commodity conversion rate to obtain a constrained commodity conversion rate; specifically, the value range constraint formula of the commodity conversion rate is as follows:
SigmoidCVR＝1/(alpha+e^(-beta*CVR))；
wherein SigmoidCVR is the commodity conversion rate after constraint, alpha can be 0.5, and beta can be 1.1. It can be understood that the influence of the commodity conversion rate on the whole value range can be reduced by carrying out value range constraint on the commodity conversion rate output by the multi-target estimation model.
Step S26: and multiplying the commodity conversion rate after constraint with the commodity click rate to obtain the click conversion probability, and sequencing the commodities according to the click conversion probability.
In this embodiment, after the constrained commodity conversion rate is obtained, the constrained commodity conversion rate is multiplied by the commodity click rate to obtain a click conversion probability CTCVR ═ CTR ═ sigmoid cvr of the commodity, and then the commodities are sorted according to the click conversion probability, and the commodities with high click conversion rate are sorted preferentially.
For the specific processes of step S21 and step S24, reference may be made to the corresponding contents disclosed in the foregoing embodiments, and details are not repeated here.
As can be seen from the above, in this embodiment, a historical click sequence including commodity features is obtained by obtaining the commodity features of the commodity and corresponding the commodity features to the commodities in the historical click sequence; inputting a historical click sequence containing commodity characteristics, a purchase intention sequence and commodity exposure data into a multi-target estimation model to obtain a corresponding commodity conversion rate; and performing constraint operation on the commodity conversion rate by using a preset value range control function to obtain the constrained commodity conversion rate, finally multiplying the constrained commodity conversion rate by the commodity click rate to obtain the click conversion probability, and sequencing the commodities according to the click conversion probability. By using the commodity characteristics as supplementary information of the historical click sequence, the model can be helped to learn commodity expression, and the capability of the commodity conversion rate task side is further enhanced. And by carrying out constraint operation on the obtained commodity conversion rate, the influence of the commodity conversion rate on the whole value range can be reduced, so that the determined click conversion probability is more in line with the actual application scene.
Correspondingly, an embodiment of the present application further discloses a network commodity sorting device, as shown in fig. 4, the device includes:
the data acquisition module 11 is used for acquiring user behavior data and commodity exposure data; the user behavior data comprises a purchase intention sequence and a historical click sequence;
the conversion rate determining module 12 is configured to input the user behavior data and the commodity exposure data into a multi-target estimation model constructed based on a multi-task learning framework, so as to obtain a corresponding commodity conversion rate; the multi-target estimation model comprises a commodity click rate learning task and a commodity conversion rate learning task; sharing an embedded layer parameter and a part of attention parameters between the commodity click rate learning task and the commodity conversion rate learning task;
the click rate determining module 13 is configured to input the historical click sequence and the commodity exposure data into a click rate estimation model to obtain a corresponding commodity click rate;
and the click conversion probability determining module 14 is configured to determine the click conversion probability of the commodity based on the commodity conversion rate and the commodity click rate, and sort the commodities according to the click conversion probability.
As can be seen from the above, in this embodiment, a purchase intention sequence, a history click sequence, and commodity exposure data are input into a multi-target pre-estimation model including a commodity click rate learning task and a commodity conversion rate learning task, and an embedded layer parameter and a part of attention parameters are shared between the commodity click rate learning task and the commodity conversion rate learning task, so that all samples, namely commodity exposure data are used for pre-estimation, and through co-learning between the commodity click rate learning task and the commodity conversion rate learning task, a relatively accurate commodity conversion rate can be obtained, and finally, a click conversion probability of a commodity is determined according to a commodity conversion rate obtained based on the multi-target pre-estimation model and a commodity click rate obtained based on the click rate pre-estimation model, and commodities are sorted according to the click conversion probability. Therefore, the optimal sorting sequence of the commodities can be determined, the commodity conversion rate is improved while the commodity click rate is guaranteed, and the total amount of the finished products is further improved.
In some specific embodiments, the data obtaining module 11 may specifically include:
the purchase intention sequence building unit is used for building an operation sequence based on the purchase adding operation and the collection operation of the user and the re-clicking operation of the history footprints and/or the favorites and/or the shopping carts; sequencing the operation sequence according to the sequence of each operation from near to far from the current time to obtain the purchase intention sequence;
and the historical click sequence construction unit is used for constructing the historical click sequence based on the click operation of the user.
In some embodiments, the conversion rate determining module 12 may specifically include:
a characteristic acquisition unit for acquiring commodity characteristics of a commodity; the commodity characteristics comprise any one or more of commodity categories, commodity prices, commodity sales amounts and commodity praise amounts;
the historical click sequence construction unit is used for corresponding the commodity characteristics with commodities in the historical click sequence to obtain a historical click sequence containing the commodity characteristics;
and the data input unit is used for inputting the historical click sequence containing the commodity characteristics, the purchase intention sequence and the commodity exposure data into the multi-target estimation model.
In some specific embodiments, the click rate determining module 13 may further include:
the constrained commodity conversion rate determining unit is used for carrying out constraint operation on the commodity conversion rate by using a preset value range control function to obtain the constrained commodity conversion rate;
and the click conversion probability determining unit is used for multiplying the commodity click rate by the commodity conversion rate after constraint to obtain the click conversion probability.
Further, the embodiment of the present application also discloses an electronic device, which is shown in fig. 5, and the content in the drawing cannot be considered as any limitation to the application scope.
Fig. 5 is a schematic structural diagram of an electronic device 20 according to an embodiment of the present disclosure. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input output interface 25, and a communication bus 26. The memory 22 is configured to store a computer program, and the computer program is loaded and executed by the processor 21 to implement relevant steps in the network commodity ordering method disclosed in any one of the foregoing embodiments.
In this embodiment, the power supply 23 is configured to provide a working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and an external device, and a communication protocol followed by the communication interface is any communication protocol applicable to the technical solution of the present application, and is not specifically limited herein; the input/output interface 25 is configured to obtain external input data or output data to the outside, and a specific interface type thereof may be selected according to specific application requirements, which is not specifically limited herein.
In addition, the memory 22 is used as a carrier for resource storage, and may be a read-only memory, a random access memory, a magnetic disk or an optical disk, etc., where the stored resources include an operating system 221, a computer program 222, data 223 including user behavior data and commodity exposure data, etc., and the storage manner may be a transient storage or a permanent storage.
The operating system 221 is used for managing and controlling each hardware device and the computer program 222 on the electronic device 20, so as to realize the operation and processing of the mass data 223 in the memory 22 by the processor 21, and may be Windows Server, Netware, Unix, Linux, and the like. The computer program 222 may further include a computer program that can be used to perform other specific tasks in addition to the computer program that can be used to perform the network commodity sorting method performed by the electronic device 20 disclosed in any of the foregoing embodiments.
Further, an embodiment of the present application further discloses a computer storage medium, where computer-executable instructions are stored in the computer storage medium, and when the computer-executable instructions are loaded and executed by a processor, the steps of the network commodity ordering method disclosed in any of the foregoing embodiments are implemented.
The embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same or similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in Random Access Memory (RAM), memory, Read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The network commodity ordering method, device, equipment and medium provided by the invention are introduced in detail, a specific example is applied in the text to explain the principle and the implementation mode of the invention, and the description of the embodiment is only used for helping to understand the method and the core idea of the invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
Claims (10)
1. A network commodity ordering method is characterized by comprising the following steps:
acquiring user behavior data and commodity exposure data; the user behavior data comprises a purchase intention sequence and a historical click sequence;
inputting the user behavior data and the commodity exposure data into a multi-target estimation model constructed based on a multi-task learning framework to obtain corresponding commodity conversion rate; the multi-target estimation model comprises a commodity click rate learning task and a commodity conversion rate learning task; sharing an embedded layer parameter and a part of attention parameters between the commodity click rate learning task and the commodity conversion rate learning task;
inputting the historical click sequence and the commodity exposure data into a click rate estimation model to obtain a corresponding commodity click rate;
determining the click conversion probability of the commodities based on the commodity conversion rate and the commodity click rate, and sequencing the commodities according to the click conversion probability.
2. The method for sorting the network commodities according to the claim 1, wherein the obtaining the user behavior data includes:
constructing an operation sequence based on the purchase adding operation and the collection operation of the user and the click operation again of the history footprints and/or the favorites and/or the shopping carts;
sequencing the operation sequence according to the sequence of each operation from near to far from the current time to obtain the purchase intention sequence;
and constructing the historical click sequence based on the click operation of the user.
3. The method for sorting online commodities, according to claim 1, wherein said inputting said user behavior data and said commodity exposure data into a multi-objective estimation model constructed based on a multi-task learning framework comprises:
inputting the commodity exposure data into the commodity click rate learning task and the commodity conversion rate learning task;
configuring non-shared parameters for the purchase intention sequence, and inputting the purchase intention sequence into the commodity click rate learning task and the commodity conversion rate learning task;
and configuring sharing parameters for the historical click sequence, and inputting the historical click sequence into the commodity click rate learning task and the commodity conversion rate learning task so that the commodity click rate learning task and the commodity conversion rate learning task share the attention parameters of the historical click sequence.
4. The method for sequencing the network commodities according to claim 1, wherein the process for constructing the multi-target estimation model comprises the following steps:
and constructing a double-task learning model sharing the embedded layer parameters and the hidden layer parameters based on the ESMM network structure to obtain the multi-target estimation model.
5. The method for sorting online commodities, according to claim 1, wherein said inputting said user behavior data and said commodity exposure data into a multi-objective estimation model constructed based on a multi-task learning framework comprises:
acquiring commodity characteristics of commodities; the commodity characteristics comprise any one or more of commodity categories, commodity prices, commodity sales amounts and commodity praise amounts;
the commodity features correspond to commodities in the historical click sequence to obtain a historical click sequence containing the commodity features;
and inputting the historical click sequence containing the commodity characteristics, the purchase intention sequence and the commodity exposure data into the multi-target estimation model.
6. The network commodity ordering method according to any one of claims 1 to 5, wherein said determining a click conversion probability of a commodity based on said commodity conversion rate and said commodity click rate comprises:
carrying out constraint operation on the commodity conversion rate by using a preset value range control function to obtain a constrained commodity conversion rate;
and multiplying the commodity conversion rate after constraint with the commodity click rate to obtain the click conversion probability.
7. A network commodity sorting apparatus, comprising:
the data acquisition module is used for acquiring user behavior data and commodity exposure data; the user behavior data comprises a purchase intention sequence and a historical click sequence;
the conversion rate determining module is used for inputting the user behavior data and the commodity exposure data into a multi-target estimation model constructed based on a multi-task learning framework to obtain corresponding commodity conversion rate; the multi-target estimation model comprises a commodity click rate learning task and a commodity conversion rate learning task; sharing an embedded layer parameter and a part of attention parameters between the commodity click rate learning task and the commodity conversion rate learning task;
the click rate determining module is used for inputting the historical click sequence and the commodity exposure data into a click rate estimation model to obtain a corresponding commodity click rate;
and the click conversion probability determining module is used for determining the click conversion probability of the commodities based on the commodity conversion rate and the commodity click rate and sequencing the commodities according to the click conversion probability.
8. The network commodity sequencing device of claim 7, wherein the conversion rate determining module further comprises:
a characteristic acquisition unit for acquiring commodity characteristics of a commodity; the commodity characteristics comprise any one or more of commodity categories, commodity prices, commodity sales amounts and commodity praise amounts;
the historical click sequence construction unit is used for corresponding the commodity characteristics with commodities in the historical click sequence to obtain a historical click sequence containing the commodity characteristics;
and the data input unit is used for inputting the historical click sequence containing the commodity characteristics, the purchase intention sequence and the commodity exposure data into the multi-target estimation model.
9. An electronic device, comprising:
a memory for storing a computer program;
a processor for executing the computer program to implement the network commodity ordering method of any one of claims 1 to 6.
10. A computer-readable storage medium for storing a computer program; wherein the computer program when executed by the processor implements the network commodity ordering method of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011264017.XA CN112381607B (en) | 2020-11-12 | 2020-11-12 | Network commodity ordering method, device, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011264017.XA CN112381607B (en) | 2020-11-12 | 2020-11-12 | Network commodity ordering method, device, equipment and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112381607A true CN112381607A (en) | 2021-02-19 |
| CN112381607B CN112381607B (en) | 2023-11-24 |
Family
ID=74583480
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011264017.XA Active CN112381607B (en) | 2020-11-12 | 2020-11-12 | Network commodity ordering method, device, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112381607B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112200627A (en) * | 2020-10-09 | 2021-01-08 | 四川长虹电器股份有限公司 | Method for improving shopping cart order conversion rate based on improved LFU strategy |
| CN113159905A (en) * | 2021-05-20 | 2021-07-23 | 深圳马六甲网络科技有限公司 | Commodity recommendation method, commodity recommendation device, commodity recommendation equipment and storage medium for new user |
| CN113312555A (en) * | 2021-06-15 | 2021-08-27 | 北京百度网讯科技有限公司 | Conversion rate prediction method, conversion rate prediction device, electronic device, and storage medium |
| CN113506154A (en) * | 2021-07-16 | 2021-10-15 | 杭州时趣信息技术有限公司 | Commodity recommendation method and system, electronic equipment and related components |
| CN115860870A (en) * | 2022-12-16 | 2023-03-28 | 深圳市云积分科技有限公司 | Commodity recommendation method, system and device and readable medium |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106779985A (en) * | 2017-02-24 | 2017-05-31 | 武汉奇米网络科技有限公司 | A kind of method and system of personalized commercial sequence |
| CN110008399A (en) * | 2019-01-30 | 2019-07-12 | 阿里巴巴集团控股有限公司 | A kind of training method and device, a kind of recommended method and device of recommended models |
| CN110033314A (en) * | 2019-03-18 | 2019-07-19 | 北京品友互动信息技术股份公司 | Advertisement data processing method and device |
| WO2019169977A1 (en) * | 2018-03-07 | 2019-09-12 | 阿里巴巴集团控股有限公司 | Information conversion rate prediction method and apparatus, and information recommendation method and apparatus |
| CN110336944A (en) * | 2019-07-05 | 2019-10-15 | 杭州时趣信息技术有限公司 | Automatically snap and analyze the method, apparatus, equipment and system of image of clothing |
| CN110415032A (en) * | 2019-07-24 | 2019-11-05 | 深圳乐信软件技术有限公司 | A kind of exposure conversion ratio predictor method, device, equipment and storage medium |
| CN110569427A (en) * | 2019-08-07 | 2019-12-13 | 智者四海（北京）技术有限公司 | Multi-target sequencing model training and user behavior prediction method and device |
| CN110796513A (en) * | 2019-09-25 | 2020-02-14 | 北京三快在线科技有限公司 | Multitask learning method and device, electronic equipment and storage medium |
| CN110838043A (en) * | 2019-11-05 | 2020-02-25 | 智者四海（北京）技术有限公司 | Commodity recommendation method and device |
| CN110880124A (en) * | 2019-09-29 | 2020-03-13 | 清华大学 | Conversion rate evaluation method and device |
| CN111027895A (en) * | 2019-05-16 | 2020-04-17 | 珠海随变科技有限公司 | Stock prediction and behavior data collection method, apparatus, device and medium for commodity |
| CN111079015A (en) * | 2019-12-17 | 2020-04-28 | 腾讯科技（深圳）有限公司 | Recommendation method and device, computer equipment and storage medium |
| CN111159241A (en) * | 2019-12-20 | 2020-05-15 | 深圳前海微众银行股份有限公司 | Click conversion estimation method and device |
| WO2020107762A1 (en) * | 2018-11-27 | 2020-06-04 | 深圳前海微众银行股份有限公司 | Ctr estimation method and device, and computer readable storage medium |
-
2020
- 2020-11-12 CN CN202011264017.XA patent/CN112381607B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106779985A (en) * | 2017-02-24 | 2017-05-31 | 武汉奇米网络科技有限公司 | A kind of method and system of personalized commercial sequence |
| WO2019169977A1 (en) * | 2018-03-07 | 2019-09-12 | 阿里巴巴集团控股有限公司 | Information conversion rate prediction method and apparatus, and information recommendation method and apparatus |
| WO2020107762A1 (en) * | 2018-11-27 | 2020-06-04 | 深圳前海微众银行股份有限公司 | Ctr estimation method and device, and computer readable storage medium |
| CN110008399A (en) * | 2019-01-30 | 2019-07-12 | 阿里巴巴集团控股有限公司 | A kind of training method and device, a kind of recommended method and device of recommended models |
| CN110033314A (en) * | 2019-03-18 | 2019-07-19 | 北京品友互动信息技术股份公司 | Advertisement data processing method and device |
| CN111027895A (en) * | 2019-05-16 | 2020-04-17 | 珠海随变科技有限公司 | Stock prediction and behavior data collection method, apparatus, device and medium for commodity |
| CN110336944A (en) * | 2019-07-05 | 2019-10-15 | 杭州时趣信息技术有限公司 | Automatically snap and analyze the method, apparatus, equipment and system of image of clothing |
| CN110415032A (en) * | 2019-07-24 | 2019-11-05 | 深圳乐信软件技术有限公司 | A kind of exposure conversion ratio predictor method, device, equipment and storage medium |
| CN110569427A (en) * | 2019-08-07 | 2019-12-13 | 智者四海（北京）技术有限公司 | Multi-target sequencing model training and user behavior prediction method and device |
| CN110796513A (en) * | 2019-09-25 | 2020-02-14 | 北京三快在线科技有限公司 | Multitask learning method and device, electronic equipment and storage medium |
| CN110880124A (en) * | 2019-09-29 | 2020-03-13 | 清华大学 | Conversion rate evaluation method and device |
| CN110838043A (en) * | 2019-11-05 | 2020-02-25 | 智者四海（北京）技术有限公司 | Commodity recommendation method and device |
| CN111079015A (en) * | 2019-12-17 | 2020-04-28 | 腾讯科技（深圳）有限公司 | Recommendation method and device, computer equipment and storage medium |
| CN111159241A (en) * | 2019-12-20 | 2020-05-15 | 深圳前海微众银行股份有限公司 | Click conversion estimation method and device |
Non-Patent Citations (4)
| Title |
|---|
| XIAO MA，等: "Entire Space Multi-Task Model: An Effective Approach for Estimating Post-Click Conversion Rate", SHORT RESEARCH PAPERS II, pages 1137 - 1140 * |
| 杨妍婷;韩斌;: "基于增强型FNN的广告点击率预测模型", 南京理工大学学报, no. 01, pages 37 - 43 * |
| 秦川;祝恒书;庄福振;郭庆宇;张琦;张乐;王超;陈恩红;熊辉;: "基于知识图谱的推荐系统研究综述", 中国科学:信息科学, no. 07, pages 937 - 956 * |
| 邓秀勤;谢伟欢;刘富春;张翼飞;樊娟;: "基于特征工程的广告点击转化率预测模型", 数据采集与处理, no. 05, pages 56 - 63 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112200627A (en) * | 2020-10-09 | 2021-01-08 | 四川长虹电器股份有限公司 | Method for improving shopping cart order conversion rate based on improved LFU strategy |
| CN112200627B (en) * | 2020-10-09 | 2022-03-25 | 四川长虹电器股份有限公司 | Method for improving shopping cart order conversion rate based on improved LFU strategy |
| CN113159905A (en) * | 2021-05-20 | 2021-07-23 | 深圳马六甲网络科技有限公司 | Commodity recommendation method, commodity recommendation device, commodity recommendation equipment and storage medium for new user |
| CN113312555A (en) * | 2021-06-15 | 2021-08-27 | 北京百度网讯科技有限公司 | Conversion rate prediction method, conversion rate prediction device, electronic device, and storage medium |
| CN113312555B (en) * | 2021-06-15 | 2023-10-03 | 北京百度网讯科技有限公司 | Conversion rate prediction method, conversion rate prediction device, electronic equipment and storage medium |
| CN113506154A (en) * | 2021-07-16 | 2021-10-15 | 杭州时趣信息技术有限公司 | Commodity recommendation method and system, electronic equipment and related components |
| CN115860870A (en) * | 2022-12-16 | 2023-03-28 | 深圳市云积分科技有限公司 | Commodity recommendation method, system and device and readable medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112381607B (en) | 2023-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112381607A (en) | Network commodity ordering method, device, equipment and medium | |
| AU2016303436B2 (en) | Method and system for applying probabilistic topic models to content in a tax environment to improve user satisfaction with a question and answer customer support system | |
| CN109829116B (en) | Content recommendation method and device, server and computer readable storage medium | |
| EP3399491A1 (en) | Data processing method and device | |
| CN104246755A (en) | Method and system to provide video-based search results | |
| CN111798280B (en) | Multimedia information recommendation method, device and equipment and storage medium | |
| CN114117216A (en) | Recommendation probability prediction method and device, computer storage medium and electronic equipment | |
| CN106445954A (en) | Business object display method and apparatus | |
| EP4283511A1 (en) | Information processing method and device, server and user equipment | |
| CN112258268A (en) | Method, apparatus, device and medium for determining recommendation model and determining price of item | |
| CN111680213B (en) | Information recommendation method, data processing method and device | |
| WO2024051707A1 (en) | Recommendation model training method and apparatus, and resource recommendation method and apparatus | |
| KR102425758B1 (en) | System for recommending company based machine learning | |
| KR102104162B1 (en) | Generating network resource | |
| Zhou et al. | Maximizing the profits of cloud service providers via dynamic virtual resource renting approach | |
| CN113360816A (en) | Click rate prediction method and device | |
| CN110990719A (en) | Information pushing method and device, readable storage medium and electronic equipment | |
| KR102372432B1 (en) | Method, device and system for providing recommended content using click and exposure information | |
| CN113128597B (en) | Method and device for extracting user behavior characteristics and classifying and predicting user behavior characteristics | |
| CN116266323A (en) | Intelligent charging method and device in cloud scene | |
| CN110659919A (en) | Data matching method, device, medium and electronic equipment | |
| CN113761352A (en) | Information pushing method, device, equipment and storage medium | |
| CN111178987A (en) | Method and device for training user behavior prediction model | |
| CN112328769A (en) | Automatic customer service response method, device and computer readable storage medium | |
| CN110880141A (en) | Intelligent deep double-tower model matching algorithm and device |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |