CN115880027A - Electronic commerce website commodity seasonal prediction model creation method - Google Patents
Electronic commerce website commodity seasonal prediction model creation method Download PDFInfo
- Publication number
- CN115880027A CN115880027A CN202211590459.2A CN202211590459A CN115880027A CN 115880027 A CN115880027 A CN 115880027A CN 202211590459 A CN202211590459 A CN 202211590459A CN 115880027 A CN115880027 A CN 115880027A
- Authority
- CN
- China
- Prior art keywords
- commodity
- seasonal
- prediction model
- training
- training sample
- 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
Images
Classifications
-
- 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
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a method for creating a seasonal prediction model of an electronic commerce website commodity, which is used for improving the accuracy of seasonal prediction of a new commodity. The method comprises the following steps: acquiring a training sample set; constructing an initial neural network prediction model; selecting training samples from a training sample set and inputting the training samples into an initial neural network prediction model; extracting the label characteristics and commodity characteristics of the training sample through a commodity seasonal discrimination layer; sequentially inputting the label characteristics and the commodity characteristics into the N label residual error layers and the characteristic fusion layer for residual error extraction to generate a target residual error; inputting the target residual error into a softmax function layer for analysis to obtain season discrimination probability data; calculating a loss value according to the seasonal discrimination probability data, the target prediction value and a preset loss function of the initial neural network prediction model to generate loss value change data; judging whether the loss value of the loss value change data in a preset interval is converged to 0 or not; if so, determining that the training of the initial neural network prediction model is finished.
Description
Technical Field
The embodiment of the application relates to the field of prediction models, in particular to a method for creating a seasonal prediction model of an electronic commerce website commodity.
Background
With the continuous development of economy, various commodities emerge endlessly, and the popularity of the commodities depends on the aspects of the market, the future trend of the market is judged earlier, and the seasonal judgment is particularly important.
Existing commodities can be classified into new commodities that change over existing commodities, new commodities that are improved over existing commodities, and brand new commodities. The brand new commodity does not appear in the prior market, so that the season is not well sold for prediction and judgment. The new goods changed on the existing goods and the new goods improved on the existing goods have higher similarity with the original goods, and the historical data of the original goods can be selected for prediction. For clothing commodities, the existing clothing commodities are rich in types, and each large merchant can only design styles, but is difficult to completely reform a certain type of clothing commodities. For example: the original commodity is a jacket of a certain style, the color of the jacket is single white, the new style pushed out by a merchant is black and white zebra stripes, the new commodity is created by changing the pattern, the merchant exits the jacket, a new structure is added to cuffs, namely, the original commodity is updated, a new product is created, the similarity of the commodity relative to the original commodity is higher, the seasonality of the new commodity can be calculated by referring to the sales history data of the original commodity, and the sales condition of the new commodity in which season is better can be predicted.
The existing prediction method is to determine an original commodity closest to a new commodity through artificial analysis, then acquire historical sales data of the original commodity, calculate according to the trend of the historical sales data, and perform systematic analysis on the trend of the historical sales data through a mathematical model to obtain a seasonal prediction result. However, the current new goods usually merge the features of a plurality of original goods, for example, a certain coat refers to the overall architecture of a first type of salable coat, refers to the collar design and cuff design of a second type of salable coat, and refers to the pocket design of a third type of coat, and at this time, the historical sales data of the three types of coats need to be merged for prediction, but the historical sales data of the first type of coat cannot be accessed. The method is used for not determining the similarity between a new commodity and an original commodity, generally, only the similarity between the new commodity and the original commodities can be determined manually, and then the seasonal prediction model is combined for calculation, but due to the fact that human factors exist in the predicted result, the predicted result is greatly discounted.
Disclosure of Invention
The application discloses a method for creating a seasonal prediction model of an e-commerce website commodity, which is used for improving the accuracy of seasonal prediction of a new commodity
The application provides a method for creating a seasonal prediction model of an electronic commerce website commodity, which comprises the following steps:
acquiring a training sample set, wherein the training sample set comprises at least two commodity images fused with commodity seasonal labels and classification labels, and training samples in the training sample set are provided with corresponding target predicted values;
constructing an initial neural network prediction model, wherein the initial neural network prediction model comprises a commodity seasonal discrimination layer and a softmax function layer, and the commodity seasonal discrimination layer comprises a feature extraction layer, N label residual error layers and a feature fusion layer;
selecting training samples from a training sample set and inputting the training samples into an initial neural network prediction model;
extracting the label characteristics and commodity characteristics of the training sample through a commodity seasonal discrimination layer;
sequentially inputting the label characteristics and the commodity characteristics into the N label residual error layers and the characteristic fusion layer for residual error extraction to generate a target residual error;
inputting the target residual error into a softmax function layer for analysis to obtain season discrimination probability data, wherein the season discrimination probability data comprises season discrimination probability values of at least four seasons;
calculating a loss value according to the seasonal discrimination probability data, the target prediction value and a preset loss function of the initial neural network prediction model to generate loss value change data, wherein the loss value change data is statistical data of the loss value generated by each training;
judging whether the loss value of the loss value change data in a preset interval is converged to 0 or not;
and if the loss value of the loss value change data in the preset interval is converged to 0, determining that the initial neural network prediction model is the target commodity seasonal prediction model.
Optionally, obtaining a training sample set includes:
acquiring a training sample generator and a commodity image set, wherein the commodity image set comprises images of at least two different commodities;
acquiring historical sales data and commodity classification type data of commodities corresponding to the commodity image set;
generating a commodity seasonal label according to historical sales data;
generating N type labels subjected to N-level classification for the commodity classification type labels according to the commodity classification type data, wherein N is an integer greater than 2;
inputting the commodity image set, the corresponding commodity seasonal labels and the N type labels subjected to N-level classification into a training sample generator to generate a training sample set, fusing the corresponding commodity seasonal labels and the type labels into training samples in the training sample set, and setting corresponding target predicted values for the training samples in the training sample set.
Optionally, inputting the commodity image set, the commodity seasonal labels corresponding to the commodity image set, and the N type labels subjected to N-level classification into a training sample generator, and generating a training sample set, including:
carrying out convolution operation on the commodity image set to generate a sampling feature set;
performing convolution operation on the seasonal labels of the commodities and the N type labels subjected to N-level classification to generate seasonal label characteristics and type characteristic labels;
and carrying out channel fusion on the sampling feature set, the seasonal tag features and the type feature tags according to the weight of the training sample generator, and carrying out reduction output through an output module Conv _ out to generate a training sample set.
Optionally, the tag feature and the commodity feature are sequentially input into the N tag residual error layers and the feature fusion layer to perform residual error extraction, so as to generate a target residual error, where the method includes:
inputting the label characteristics and the commodity characteristics into a first label residual error layer to perform residual error calculation, and generating a first residual error of a training sample;
channel superposition is carried out on the first residual error and a first class label in the N types of labels which are classified in the N levels, and a first fusion characteristic is generated;
inputting the first fusion characteristic into a first label residual error layer to generate a second residual error of the training sample;
channel superposition is carried out on the second residual error and a second class label in the N class-classified type labels to generate a second fusion characteristic;
inputting the (N-1) th residual error into an Nth label residual error layer to generate an Nth residual error of the training sample;
and sequentially fusing the first residual to the Nth residual through the residual fusion layer to generate a target residual.
Optionally, the initial neural network prediction model further includes a regional channel attention module SKConv;
after the label characteristics and the commodity characteristics are sequentially input into the N label residual error layers and the characteristic fusion layer for residual error extraction, and a target residual error is generated, the target residual error is input into the softmax function layer for analysis to obtain seasonal discrimination probability data, and before the seasonal discrimination probability data contains seasonal discrimination probability values of at least four seasons, the electronic commerce website commodity seasonal prediction model establishing method further comprises the following steps:
attention is distributed to different size regions of the target residual error through convolution kernels of different size receptive fields in a region channel attention module SKConv, and different feature channels of the target residual error are screened through the attention distribution.
Optionally, the initial neural network prediction model further includes an attention channel pooling module ACD;
after distributing attention to different size regions of the target residual error through convolution kernels of different size receptive fields in the region channel attention module SKConv and performing screening conversion on different feature channels of the target residual error through the distributed attention, the method for creating the commodity seasonal prediction model of the e-commerce website further comprises the following steps:
attention is distributed to each channel of the target residual error through an attention channel pooling module ACD, and the channels with the later attention ranking are abandoned.
Optionally, the initial neural network prediction model further includes a channel shuffling module CSA;
the target residual is channel shuffled by a channel shuffle module CSA.
Optionally, after determining whether the loss value of the loss value variation data in the preset interval converges to 0, the method for creating the seasonal prediction model of the e-commerce website commodity further includes:
if the loss value of the loss value change data in the preset interval is not converged to 0, judging whether the training times of the training samples reach the standard or not;
and if the training times of the training samples reach the standard, updating the weight of the initial neural network prediction model according to a small batch gradient descent method, and reselecting the training samples from the training sample set to input the training samples into the initial neural network prediction model for training.
Optionally, after determining whether the training times of the training samples reach the standard, the method for creating the seasonal prediction model of the commodity of the e-commerce website further includes:
and if the training times of the training samples do not reach the standard, updating the weight of the initial neural network prediction model according to a small batch gradient descent method, and inputting the training samples into the initial neural network prediction model again for training.
Optionally, after determining whether the training times of the training samples reach the standard, the method for creating the commodity seasonal prediction model of the e-commerce website further includes:
and updating the weight of the training sample generator by a small batch gradient descent method according to the loss value, and updating the training samples in the training sample set by the training sample generator.
According to the technical scheme, the embodiment of the application has the following advantages:
according to the method, a training sample set is obtained firstly, the training sample set comprises at least two commodity images fused with commodity seasonal labels and classification labels, corresponding target predicted values are set in training samples in the training sample set, and loss values of subsequent models are calculated. The training sample is image data generated by appearance images of original commodities and combining the seasonal sales condition of the commodities and corresponding classification labels. Next, an initial neural network prediction model needs to be constructed, wherein the initial neural network prediction model comprises a commodity seasonal discrimination layer and a softmax function layer, and the commodity seasonal discrimination layer comprises a feature extraction layer, N tag residual error layers and a feature fusion layer. Firstly, a training sample is selected from a training sample set and input into an initial neural network prediction model, wherein the training samples can be input one by one or simultaneously input in batches. And extracting the label characteristics and the commodity characteristics of the training sample through a commodity seasonal discrimination layer. And then, sequentially inputting the label characteristics and the commodity characteristics into the N label residual error layers and the characteristic fusion layer for residual error extraction, and generating a target residual error. And inputting the target residual error into the softmax function layer for analysis to obtain season discrimination probability data, wherein the season discrimination probability data comprises season discrimination probability values of at least four seasons. And calculating a loss value according to the seasonal discrimination probability data, the target prediction value and a preset loss function of the initial neural network prediction model to generate loss value change data, wherein the loss value change data is statistical data of the loss value generated by each training. And judging whether the loss value of the loss value change data in the preset interval is converged to 0 or not. And if the loss value of the loss value change data in the preset interval is converged to 0, determining that the initial neural network prediction model is the target commodity seasonal prediction model. The invention learns the characteristics of the training samples by taking the convolutional neural network model as a basic framework of the initial neural network prediction model, and the corresponding classification labels and seasonal labels exist in the training samples, so that the network model can analyze the seasonal labels and the classification labels while identifying the commodities to obtain the probability value of the corresponding commodities belonging to a certain seasonal commodity, the old commodities similar to the new commodities are not required to be determined by human factors, the similarity between the new commodities and the original commodities can be identified by the seasonal prediction model of the target commodities, the corresponding seasonal discrimination probability data is given, and the accuracy of seasonal prediction of the new commodities is improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.
FIG. 1 is a schematic diagram of an embodiment of a method for creating a seasonal prediction model of an e-commerce website commodity according to the present application;
FIG. 2-1 is a schematic diagram of a first stage of the method for creating a seasonal prediction model of an e-commerce website commodity according to an embodiment of the present application;
FIG. 2-2 is a schematic diagram of a second stage of the method for creating a seasonal prediction model of merchandise in an e-commerce website according to an embodiment of the present application;
FIGS. 2-3 are schematic diagrams illustrating an embodiment of a third stage of the method for creating a seasonal prediction model of an item on an e-commerce website according to the present application;
FIG. 3 is a flow diagram illustrating an embodiment of an initial neural network prediction model network layer in an embodiment of the present application;
FIG. 4 is a schematic structural diagram of another embodiment of an initial neural network prediction model network layer in the embodiment of the present application;
fig. 5 is a schematic structural diagram of another embodiment of the initial neural network prediction model network layer in the embodiment of the present application.
Detailed Description
In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular system structures, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. It will be apparent, however, to one skilled in the art that the present application may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.
It will be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It should also be understood that the term "and/or" as used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.
As used in this specification and the appended claims, the term "if" may be interpreted contextually as "when", "upon" or "in response to" determining "or" in response to detecting ". Similarly, the phrase "if it is determined" or "if a [ described condition or event ] is detected" may be interpreted contextually to mean "upon determining" or "in response to determining" or "upon detecting [ described condition or event ]" or "in response to detecting [ described condition or event ]".
Furthermore, in the description of the present application and the appended claims, the terms "first," "second," "third," and the like are used for distinguishing between descriptions and not necessarily for describing or implying relative importance.
Reference throughout this specification to "one embodiment" or "some embodiments," or the like, means that a particular feature, structure, or characteristic described in connection with the embodiment is included in one or more embodiments of the present application. Thus, appearances of the phrases "in one embodiment," "in some embodiments," "in other embodiments," or the like, in various places throughout this specification are not necessarily all referring to the same embodiment, but rather "one or more but not all embodiments" unless specifically stated otherwise. The terms "comprising," "including," "having," and variations thereof mean "including, but not limited to," unless otherwise specifically stated.
In the prior art, the existing prediction method is to determine an original commodity closest to a new commodity through artificial analysis, then obtain historical sales data of the original commodity, calculate according to the trend and the trend of the historical sales data, and perform systematic analysis on the trend and the trend of the historical sales data through a mathematical model to obtain a seasonal prediction result. However, the current new goods usually merge the features of a plurality of original goods, for example, a certain coat refers to the overall architecture of a first type of salable coat, refers to the collar design and cuff design of a second type of salable coat, and refers to the pocket design of a third type of coat, and at this time, the historical sales data of the three types of coats need to be merged for prediction, but the historical sales data of the first type of coat cannot be accessed. The method is used for determining the similarity between a new commodity and an original commodity, and generally, the similarity between the new commodity and a plurality of original commodities can only be determined artificially, and then the seasonal prediction model is used for calculating, but due to the existence of artificial factors in the predicted result, the predicted result is greatly discounted, and in conclusion, the existing seasonal prediction model has low accuracy of carrying out seasonal prediction on the new product according to historical sales data of a plurality of new commodities.
Based on the method, the application discloses a method for creating a seasonal prediction model of the commodities of the electronic commerce website, which is used for improving the accuracy of seasonal prediction of new commodities.
The technical solutions in the present application will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present application, and it is obvious that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The method of the present application may be applied to a server, a device, a terminal, or other devices with logic processing capability, and the present application is not limited thereto. For convenience of description, the following description will be given taking the execution body as an example.
Referring to fig. 1, the present application provides an embodiment of a method for creating a seasonal prediction model of an e-commerce website commodity, including:
101. acquiring a training sample set, wherein the training sample set comprises at least two commodity images fused with commodity seasonal labels and classification labels, and training samples in the training sample set are provided with corresponding target predicted values;
the terminal first obtains a set of training samples. The method for obtaining the training sample mainly comprises the steps of shooting at least two images of an original commodity, obtaining historical sales data of the commodity, extracting seasonal data of the commodity and classification data of the commodity from the historical sales data, converting the seasonal data and the classification data into digital parameters, and integrating the digital parameters into the commodity image.
For example: a certain popular coat is sold well in autumn and is classified as a coat. And (3) carrying out channel fusion on the two parameters and the commodity image to form a training sample, wherein the commodity seasonal label is 1, and the classification label of the coat is 10.
The rules for specific article season labeling and classification labeling can be more complex, by way of simple example only, and in fact the labels are a set of channel parameters that can generate image data.
102. Constructing an initial neural network prediction model, wherein the initial neural network prediction model comprises a commodity seasonal discrimination layer and a softmax function layer, and the commodity seasonal discrimination layer comprises a feature extraction layer, N label residual error layers and a feature fusion layer;
the terminal constructs an initial neural network prediction model, the initial neural network prediction model comprises a commodity seasonal discrimination layer and a softmax function layer, and the commodity seasonal discrimination layer comprises a feature extraction layer, N label residual error layers and a feature fusion layer.
Wherein, the characteristic extraction layer is formed by two groups of convolution layers-normalization layers which are arranged in series.
103. Selecting training samples from the training sample set and inputting the training samples into an initial neural network prediction model;
and the terminal selects the training samples from the training sample set and inputs the training samples into the initial neural network prediction model. A particular terminal may randomly extract a certain number of samples from a training sample set while training. In this embodiment, a small batch of 32 training initial neural network prediction models are used, and a training effect is achieved through multiple iterations. In this embodiment, the number of iterations is about 25000.
104. Extracting the label characteristics and commodity characteristics of the training sample through a commodity seasonal discrimination layer;
after the terminal selects the training samples from the training sample set and inputs the training samples into the initial neural network prediction model, the label features and the commodity features of the training samples are extracted through the feature extraction layer in the commodity seasonal discrimination layer, namely commodity information of the commodity images and seasonal information and classification information blended into the training samples are read from the training samples.
105. Sequentially inputting the label characteristics and the commodity characteristics into the N label residual error layers and the characteristic fusion layer for residual error extraction to generate a target residual error;
and the terminal sequentially inputs the label characteristics and the commodity characteristics into the N label residual error layers and the characteristic fusion layer for residual error extraction to generate a target residual error, so that the classification of the commodity information can be fully learned in the target residual error, and the classification of each commodity image can be accurately identified by the whole prediction model.
106. Inputting the target residual error into a softmax function layer for analysis to obtain season discrimination probability data, wherein the season discrimination probability data comprises season discrimination probability values of at least four seasons;
and the terminal inputs the target residual error into the softmax function layer for analysis to obtain season discrimination probability data, wherein the season discrimination probability data comprises season discrimination probability values of at least four seasons.
Specifically, the terminal carries out classification calculation on the target residual errors through a global average pooling layer and a softmax layer, and season discrimination probability values of the training samples belonging to all seasons are generated.
The specific calculation method is as follows:
wherein e x Is an exponential function, y i Representing the first input neuron in the output layer, the operation of denominator represents that n output neurons are in total in the output layer, and the index sum of the input neurons in all the output layers is calculated, yi represents the output of the ith neuron, softmax (y) i ) Probability data is discriminated for seasons.
And the terminal determines the seasonal prediction of the training sample according to the seasonal discrimination probability value in the seasonal discrimination probability data.
107. Calculating a loss value according to the seasonal discrimination probability data, the target prediction value and a preset loss function of the initial neural network prediction model to generate loss value change data, wherein the loss value change data is statistical data of the loss value generated by each training;
and the terminal calculates a loss value according to the seasonal discrimination probability data, the target prediction value and a preset loss function of the initial neural network prediction model to generate loss value change data, wherein the loss value change data is statistical data of the loss value generated by each training.
The classifier is composed of a global average pooling layer and a softmax layer, the classification is a multi-classification problem, so a loss function adopted in the training process of the initial neural network prediction model is a cross entropy loss function, and the expression is as follows:
wherein Loss is a Loss value, i.e. a predicted value for each quarter, the parameter meaning of C is the number of total samples, y i Target prediction value, P, for the real season of the ith training sample i Namely seasonal discrimination probability data softmax (y) output by the initial neural network prediction model i ). The following illustrates the manner in which this cross entropy loss function is used:
in the digital recognition task, if the digital sample is the number "5", then the true distribution should be: <xnotran> [0,0,0,0,0,1,0,0,0,0], : </xnotran> [0.1,0, 0.7,0,0.1, 0] should be 10, then the loss function is calculated as:
Loss＝-[0*log(0.1)]*3-[0*log(0)]*6-1*log(0.7)≈0.3567
if the distribution of the network output is: [0.2,0.3,0.1, 0], then the loss function is calculated as:
Loss＝-[0*log(0.1)]*2-[0*log(0.2)]-[0*log(0)]*5-[0*log(0.3)]*2-log(0.7)≈1.204
compared with the two cases, the loss value calculated by the cross entropy function of the loss of the first distribution is obviously lower than the loss value calculated by the cross entropy function of the second distribution, and the first distribution is closer to the real distribution.
The loss value change data in this embodiment is recorded data of each loss value, and the standard of convergence is that the loss value reaches within a preset threshold value for more than a predetermined number of times. For example, inputting a training sample, wherein the real and popular season is autumn, the target prediction value is eighty percent, and after calculating the loss value, if the initial neural network prediction models of the last 10000 times reach the area which tends to 0, the initial neural network prediction models are converged, namely the training is finished.
108. Judging whether the loss value of the loss value change data in a preset interval is converged to 0 or not;
109. and if the loss value of the loss value change data in the preset interval is converged to 0, determining that the initial neural network prediction model is the target commodity seasonal prediction model.
And the terminal judges whether the loss value of the loss value change data in the preset interval is converged to 0, and when the loss value change data in the preset interval and the magnitude and the trend of all the loss values are converged to 0, the initial neural network prediction model training can be determined to be finished.
In this embodiment, a training sample set is first obtained, where the training sample set includes at least two commodity images with commodity seasonal labels and classification labels fused together, and training samples in the training sample set are all provided with corresponding target predicted values, and loss values of subsequent models are calculated. The training sample is image data generated by appearance images of original commodities and combining the seasonal sales condition of the commodities and corresponding classification labels. Next, an initial neural network prediction model needs to be constructed, wherein the initial neural network prediction model comprises a commodity seasonal discrimination layer and a softmax function layer, and the commodity seasonal discrimination layer comprises a feature extraction layer, N tag residual error layers and a feature fusion layer. Firstly, a training sample is selected from a training sample set and input into an initial neural network prediction model, wherein the training samples can be input one by one or simultaneously input in batches. And extracting the label characteristics and the commodity characteristics of the training sample through a commodity seasonal discrimination layer. And then, sequentially inputting the label characteristics and the commodity characteristics into the N label residual error layers and the characteristic fusion layer for residual error extraction, and generating a target residual error. And inputting the target residual error into the softmax function layer for analysis to obtain season discrimination probability data, wherein the season discrimination probability data comprises season discrimination probability values of at least four seasons. And calculating a loss value according to the seasonal discrimination probability data, the target predicted value and a preset loss function of the initial neural network prediction model to generate loss value change data, wherein the loss value change data is statistical data of the loss value generated by each training. And judging whether the loss value of the loss value change data in the preset interval is converged to 0 or not. And if the loss value of the loss value change data in the preset interval is converged to 0, determining that the initial neural network prediction model is the target commodity seasonal prediction model. The method learns the characteristics of the training samples by taking the convolutional neural network model as a basic framework of the initial neural network prediction model, and the corresponding classification labels and seasonal labels exist in the training samples, so that the network model can analyze the seasonal labels and the classification labels while identifying the commodities to obtain the probability value of the corresponding commodity belonging to a certain seasonal commodity, the old commodity similar to the new commodity is not required to be determined by human factors, the similarity between the new commodity and the original commodity can be identified by the target commodity seasonal prediction model, the corresponding seasonal discrimination probability data is given, and the accuracy of seasonal prediction on the new commodity is improved.
Referring to fig. 2, the present application provides an embodiment of a method for creating a seasonal prediction model of an e-commerce website commodity, including:
201. acquiring a training sample generator and a commodity image set, wherein the commodity image set comprises images of at least two different commodities;
202. acquiring historical sales data and commodity classification type data of commodities corresponding to the commodity image set;
203. generating a commodity seasonal label according to historical sales data;
204. generating N type labels subjected to N-level classification for the commodity classification type labels according to the commodity classification type data, wherein N is an integer greater than 2;
205. carrying out convolution operation on the commodity image set to generate a sampling feature set;
206. performing convolution operation on the seasonal labels of the commodities and the N type labels subjected to N-level classification to generate seasonal label characteristics and type characteristic labels;
207. performing channel fusion on the sampling feature set, the seasonal label features and the type feature labels according to the weight of the training sample generator, and performing reduction output through an output module Conv _ out to generate a training sample set, wherein corresponding commodity seasonal labels and type labels are fused into training samples in the training sample set, and the training samples in the training sample set are provided with corresponding target predicted values;
the terminal firstly acquires a training sample generator and a commodity image set, wherein the commodity image set comprises images of at least two different commodities, such as different types of commodity images of coats, shirts, coats, trousers and the like. And commodity images of different styles are classified in the same mode. The terminal obtains historical sales data and commodity classification type data of commodities corresponding to the commodity image set, and generates commodity seasonal labels according to the historical sales data.
For example: if the current season of a jacket belongs to the first period and the second period of autumn, corresponding season labels are generated for the jacket.
And generating N type labels subjected to N-level classification for the commodity classification type labels according to the commodity classification type data, wherein N is an integer greater than 2.
For example: a style jacket belongs to four categories of men's clothing, coats, jackets and the like, and 5 types of labels are produced for the commodity.
The terminal conducts convolution operation on the commodity image set to generate a sampling feature set, conducts convolution operation on the commodity seasonal labels and the N type labels which are classified in the N levels, and generates seasonal label features and type feature labels. And then, the terminal performs channel fusion on the sampling feature set, the seasonal label features and the type feature labels according to the weight of the training sample generator, and performs reduction output through an output module Conv _ out to generate a training sample set, wherein the training samples in the training sample set are fused with corresponding commodity seasonal labels and type labels, and the training samples in the training sample set are all provided with corresponding target predicted values.
208. Constructing an initial neural network prediction model, wherein the initial neural network prediction model comprises a commodity seasonal discrimination layer and a softmax function layer, and the commodity seasonal discrimination layer comprises a feature extraction layer, N label residual error layers and a feature fusion layer;
209. selecting training samples from a training sample set and inputting the training samples into an initial neural network prediction model;
210. extracting the label characteristics and commodity characteristics of the training sample through a commodity seasonal discrimination layer;
in this embodiment, steps 208 to 210 are similar to steps 102 to 104, and are not described herein again.
211. Inputting the label characteristics and the commodity characteristics into a first label residual error layer to perform residual error calculation, and generating a first residual error of a training sample;
212. channel superposition is carried out on the first residual error and a first class label in the N class-classified type labels to generate a first fusion characteristic;
213. inputting the first fusion characteristic into a first label residual error layer to generate a second residual error of the training sample;
214. channel superposition is carried out on the second residual error and a second class label in the N class labels after N-level classification, and a second fusion feature is generated;
215. inputting the (N-1) th residual error into an Nth label residual error layer to generate an Nth residual error of the training sample;
216. sequentially fusing the first residual error to the Nth residual error through a residual error fusion layer to generate a target residual error;
the terminal inputs the label characteristics and the commodity characteristics into a first label residual error layer to perform residual error calculation, generates a first residual error of a training sample, performs channel superposition on the first residual error and a first class label in N types of labels classified by N levels to generate a first fusion characteristic, inputs the first fusion characteristic into the first label residual error layer to generate a second residual error of the training sample, and performs channel superposition on the second residual error and a second class label in the N types of labels classified by N levels to generate a second fusion characteristic. And inputting the (N-1) th residual into the (N) th label residual layer to generate an (N) th residual of the training sample. And sequentially fusing the first residual to the Nth residual through the residual fusion layer to generate a target residual.
In this embodiment, the sub-category type labels are strengthened in the first to nth fusion features, and the sub-category type labels are continuously fused, so that the final target residual is fully learned about the commodity and the classification type corresponding to the commodity.
217. Distributing attention to different size regions of the target residual error through convolution cores of different size receptive fields in a region channel attention module SKConv, and screening different characteristic channels of the target residual error through the distributed attention;
the terminal distributes attention to different regions of the target residual error through convolution cores of different-size receptive fields in the region channel attention module SKConv, and screens different characteristic channels of the target residual error through the distribution attention, and the purpose is to screen out channels with the region channel attention lower than a preset value, so that the network model can pay more attention to useful channels and learn the useful channels.
218. Distributing attention to each channel of the target residual errors through an attention channel pooling module ACD, and discarding the channels with the later attention ranking;
the terminal distributes attention to each channel of the target residual errors through an attention channel pooling module ACD, and abandons the channels with later attention ranking, so as to further abandon the channels with the attention lower than a preset value, so that the network model can pay more attention to the useful channels and learn the useful channels.
219. Performing channel shuffling on the target residual error through a channel shuffling module CSA;
and the terminal performs channel shuffling on the target residual error through a channel shuffling module CSA, and fully fuses the characteristics in the target residual error again.
220. Inputting the target residual error into a softmax function layer for analysis to obtain season discrimination probability data, wherein the season discrimination probability data comprises season discrimination probability values of at least four seasons;
221. calculating a loss value according to the seasonal discrimination probability data, the target prediction value and a preset loss function of the initial neural network prediction model to generate loss value change data, wherein the loss value change data is statistical data of the loss value generated by each training;
222. judging whether the loss value of the loss value change data in a preset interval is converged to 0 or not;
223. if the loss value of the loss value variation data in the preset interval is converged to 0, determining that the initial neural network prediction model is a target commodity seasonal prediction model;
in this embodiment, steps 220 to 223 are similar to steps 106 to 109, and are not described herein.
224. If the loss value of the loss value change data in the preset interval is not converged to 0, judging whether the training times of the training samples reach the standard or not;
225. if the training times of the training samples reach the standard, updating the weight of the initial neural network prediction model according to a small batch gradient descent method, and reselecting the training samples from the training sample set to input the training samples into the initial neural network prediction model for training;
226. if the training times of the training samples do not reach the standard, updating the weight of the initial neural network prediction model according to a small batch gradient descent method, and inputting the training samples into the initial neural network prediction model again for training;
in this embodiment, the terminal determines whether the loss value of the loss value change data in the preset interval converges to 0, if the loss value of the loss value change data in the preset interval does not converge to 0, determines whether the training frequency of the training sample reaches the standard, if the training frequency of the training sample reaches the standard, updates the weight of the initial neural network prediction model according to a small batch gradient descent method, reselects the training sample from the training sample set to input the training sample into the initial neural network prediction model for training, and if the training frequency of the training sample does not reach the standard, updates the weight of the initial neural network prediction model according to the small batch gradient descent method, and inputs the training sample into the initial neural network prediction model again for training.
For the weight update of the initial neural network prediction model, a small batch stochastic gradient descent method is taken as an example for updating the convolutional neural network model, wherein the formula of the gradient update mode of batch training is as follows:
n is the batch size, η is the learning rate, w t Is the current weight, W t+1 In order to update the weight values,
updating the subfunction for the weight value, wherein x is a preset value.
Using inverse gradient derivation, referring to fig. 3, fig. 3 is a schematic diagram of an initial neural network prediction model network layer.
On the left side is the first layer, also the input layer, which contains two neurons a and b. In the middle is a second layer, also the hidden layer, which contains two neurons c and d. The third layer, also the output layer, on the right, contains e and f, marked on each line
Is a layer ofThe weight of the connections between layers.
W represents a weight matrix, Z represents an input matrix, A represents an output matrix, and Y represents a standard answer.
L represents the number of layers of the convolutional neural network model.
The forward propagation method is to transmit the signal of the input layer to the hidden layer, taking hidden layer node c as an example, and looking backward (in the direction of the input layer) on node c, it can be seen that there are two arrows pointing to node c, so the information of nodes a and b will be transmitted to node c, and each arrow has a certain weight, so for node c, the input signal is:
similarly, the input signal of the node d is:
since the terminal is good at doing tasks with loops, it can be represented by a matrix multiplication:
Z 2 ＝W 1 *A 1 +B 2
therefore, the output of the hidden layer node after nonlinear transformation is represented as follows:
A 2 ＝sigmoid(Z 2 )
similarly, the input signal of the output layer is represented as the output of the weight matrix multiplied by the above layer:
Z 3 ＝W 2 *A 2 +B 3
similarly, the final output of the output layer node after nonlinear mapping is represented as:
A 3 ＝sigmoid(Z 3 )
the input signal gets the output of each layer with the help of the weight matrix, and finally reaches the output layer. Therefore, the weight matrix plays a role of a transportation soldier in the process of forward signal propagation and plays a role of starting and starting.
Referring to fig. 4, fig. 4 is a schematic diagram of an initial neural network prediction model network layer. The backward propagation method, since gradient descent requires explicit errors in each layer to update the parameters, is followed by how to propagate the errors of the output layer back to the hidden layer.
Wherein, the errors of the nodes of the output layer and the hidden layer are shown in the figure, the error of the output layer is known, and then the error analysis is carried out on the first node c of the hidden layer. Or on node c, except this time looking forward (in the direction of the output layer), it can be seen that the two blue thick arrows pointing to node c start from node e and node f, so the error for node c must be related to nodes e and f of the output layer. The node e of the output layer has arrows pointing to the nodes c and d of the hidden layer respectively, so that the error of the hidden node e cannot be owned by the hidden node c, but the error of the node f is subject to the principle of distribution according to the labor (distribution according to the weight), and similarly, the error of the node f is subject to the principle, so that the error of the node c of the hidden layer is:
wherein e is o1 And e o2 For the output layer back propagation coefficient, the error for the hidden layer node d is, similarly:
wherein e is h1 And e h2 For the hidden layer back propagation coefficients, to reduce the workload, we can write the form of matrix multiplication:
the matrix is relatively complicated, can be simplified to a forward propagation form, and does not destroy the proportion of the forward propagation form, so that the denominator part can be omitted, and the matrix is formed again as follows:
the weight matrix is actually the transpose of the weight matrix w in forward propagation, so the form is abbreviated as follows:
E h ＝W T *E o
the output layer errors are passed to the hidden layer with the help of the transposed weight matrix, so that we can update the weight matrix connected to the hidden layer with indirect errors. It can be seen that the weight matrix also acts as a transportation soldier in the back propagation process, but this time the output error of the transport, not the input signal.
Referring to fig. 5, fig. 5 is a schematic diagram of an initial neural network prediction model network layer. Chain derivation is then required, the forward propagation of the input information and the backward propagation of the output error are described above, and the parameters are then updated according to the calculated error.
First w to the hidden layer 11 Updating parameters, before updating let us deduce from back to front until w is foreseen 11 The calculation is as follows:
thus error pair w 11 The derivation is as follows:
the following formula is derived (all values are known):
similarly, error is for w 12 The partial derivatives of (A) are as follows:
likewise, derived by w 12 Evaluation formula of (c):
similarly, the error is biased for the offset as follows:
similarly, the error is biased for the offset as follows:
then w to the input layer 11 Updating parameters, and before updating, deriving the parameters from back to front until predicting w of the first layer 11 So far:
the error is therefore the partial derivative of w11 for the input layer as follows:
the derivation is as follows:
similarly, the other three parameters of the input layer can be used to calculate their respective partial derivatives according to the same method, which is not described herein. In the case where the partial derivative of each parameter is definite, the gradient descent formula is substituted by:
so far, the task of updating each layer of parameters by using the chain rule has been completed.
After the weight of the initial neural network prediction model is updated, one part of the initial neural network prediction model is reserved, so that when problems such as generalization and overfitting occur in the subsequent training process, the originally stored initial neural network prediction model can be used.
227. And updating the weight of the training sample generator by a small batch gradient descent method according to the loss value, and updating the training samples in the training sample set by the training sample generator.
The terminal updates the weight of the training sample generator through a small batch gradient descent method according to the loss value, updates the training samples in the training sample set through the training sample generator, continuously updates the commodity image through the generator repeatedly, and generates new training samples which are more suitable for the initial neural network prediction model, wherein the updating method is in the step 226, which is not repeated here.
In this embodiment, a training sample generator and a commodity image set are first obtained, where the commodity image set includes images of at least two different commodities, and historical sales data and commodity classification type data of commodities corresponding to the commodity image set are obtained. And generating a commodity season label according to the historical sales data. And generating N type labels subjected to N-level classification for the commodity classification type labels according to the commodity classification type data, wherein N is an integer greater than 2. And carrying out convolution operation on the commodity image set to generate a sampling feature set. And performing convolution operation on the seasonal labels of the commodities and the N type labels subjected to N-level classification to generate seasonal label characteristics and type characteristic labels. And carrying out channel fusion on the sampling feature set, the seasonal label features and the type feature labels according to the weight of the training sample generator, carrying out reduction output through an output module Conv _ out, and generating a training sample set, wherein corresponding commodity seasonal labels and type labels are fused into training samples in the training sample set, and the training samples in the training sample set are provided with corresponding target predicted values. Next, an initial neural network prediction model needs to be constructed, wherein the initial neural network prediction model comprises a commodity seasonal discrimination layer and a softmax function layer, and the commodity seasonal discrimination layer comprises a feature extraction layer, N tag residual error layers and a feature fusion layer. Firstly, a training sample is selected from a training sample set and input into an initial neural network prediction model, wherein the training samples can be input one by one or simultaneously input in batches. And extracting the label characteristics and the commodity characteristics of the training sample through a commodity seasonal discrimination layer. And then inputting the label characteristics and the commodity characteristics into a first label residual error layer for residual error calculation to generate a first residual error of a training sample, performing channel superposition on the first residual error and a first class label in N types of labels classified by N levels to generate a first fusion characteristic, inputting the first fusion characteristic into the first label residual error layer to generate a second residual error of the training sample, performing channel superposition on the second residual error and a second class label in the N types of labels classified by N levels to generate a second fusion characteristic, inputting the N-1 th residual error into an Nth label residual error layer to generate an Nth residual error of the training sample, and sequentially fusing the first residual error to the Nth residual error through the residual error fusion layer to generate a target residual error.
The terminal distributes attention to different regions of a target residual error through convolution cores of different magnitude receptive fields in a region channel attention module SKConv, screens different characteristic channels of the target residual error through the distribution attention, distributes attention to each channel of the target residual error through an attention channel pooling module ACD, discards the channels with the later attention ranking, merges the channels with the target residual error through a channel shuffling module CSA, reserves the channels with high attention, and discards the channels with the attention lower than a preset value, so that a network model can pay more attention to useful channels and learn the useful channels.
And inputting the target residual error into the softmax function layer for analysis to obtain season discrimination probability data, wherein the season discrimination probability data comprises season discrimination probability values of at least four seasons. And calculating a loss value according to the seasonal discrimination probability data, the target prediction value and a preset loss function of the initial neural network prediction model to generate loss value change data, wherein the loss value change data is statistical data of the loss value generated by each training. And judging whether the loss value of the loss value change data in the preset interval is converged to 0 or not. And if the loss value of the loss value change data in the preset interval is converged to 0, determining that the initial neural network prediction model is the target commodity seasonal prediction model.
And if the loss value of the loss value change data in the preset interval is not converged to 0, judging whether the training times of the training samples reach the standard or not. And if the training times of the training samples reach the standard, updating the weight of the initial neural network prediction model according to a small batch gradient descent method, and reselecting the training samples from the training sample set to input the training samples into the initial neural network prediction model for training. And if the training times of the training samples do not reach the standard, updating the weight of the initial neural network prediction model according to a small batch gradient descent method, and inputting the training samples into the initial neural network prediction model again for training. And updating the weight of the training sample generator by a small batch gradient descent method according to the loss value, and updating the training samples in the training sample set by the training sample generator.
The method learns the characteristics of the training samples by taking the convolutional neural network model as a basic framework of the initial neural network prediction model, and the corresponding classification labels and seasonal labels exist in the training samples, so that the network model can analyze the seasonal labels and the classification labels while identifying the commodities to obtain the probability value of the corresponding commodity belonging to a certain seasonal commodity, the old commodity similar to the new commodity is not required to be determined by human factors, the similarity between the new commodity and the original commodity can be identified by the target commodity seasonal prediction model, the corresponding seasonal discrimination probability data is given, and the accuracy of seasonal prediction on the new commodity is improved.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one type of logical functional division, and other divisions may be realized in practice, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application, which are essential or part of the technical solutions contributing to the prior art, or all or part of the technical solutions, may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and the like.
Claims (10)
1. A method for creating a seasonal prediction model of an e-commerce website commodity is characterized by comprising the following steps:
acquiring a training sample set, wherein the training sample set comprises at least two commodity images fused with commodity seasonal labels and classification labels, and training samples in the training sample set are provided with corresponding target predicted values;
constructing an initial neural network prediction model, wherein the initial neural network prediction model comprises a commodity seasonal discrimination layer and a softmax function layer, and the commodity seasonal discrimination layer comprises a feature extraction layer, N label residual error layers and a feature fusion layer;
selecting training samples from the training sample set and inputting the training samples into the initial neural network prediction model;
extracting the label characteristics and commodity characteristics of the training samples through the commodity seasonal judging layer;
inputting the label features and the commodity features into N label residual error layers and feature fusion layers in sequence for residual error extraction to generate target residual errors;
inputting the target residual error into a softmax function layer for analysis to obtain season discrimination probability data, wherein the season discrimination probability data comprises season discrimination probability values of at least four seasons;
calculating a loss value according to the season discrimination probability data, the target prediction value and a preset loss function of the initial neural network prediction model to generate loss value change data, wherein the loss value change data is statistical data of the loss value generated by each training;
judging whether the loss value of the loss value change data in a preset interval is converged to 0 or not;
and if the loss value of the loss value change data in a preset interval is converged to 0, determining that the initial neural network prediction model is a target commodity seasonal prediction model.
2. The method as claimed in claim 1, wherein the obtaining a training sample set comprises:
acquiring a training sample generator and a commodity image set, wherein the commodity image set comprises images of at least two different commodities;
acquiring historical sales data and commodity classification type data of commodities corresponding to the commodity image set;
generating a commodity seasonal label according to the historical sales data;
generating N type labels subjected to N-level classification for the commodity classification type labels according to the commodity classification type data, wherein N is an integer greater than 2;
inputting the commodity image set, the commodity seasonal labels corresponding to the commodity image set and the N type labels subjected to N-level classification into the training sample generator to generate a training sample set, wherein the corresponding commodity seasonal labels and the type labels are blended into training samples in the training sample set, and the training samples in the training sample set are provided with corresponding target predicted values.
3. The method for creating a seasonal predictive model of an e-commerce website commodity according to claim 2, wherein the step of inputting the commodity image set and the corresponding commodity seasonal labels and N class-N classified type labels into the training sample generator generates a training sample set, comprising:
performing convolution operation on the commodity image set to generate a sampling feature set;
performing convolution operation on the commodity seasonal labels and the N type labels subjected to N-level classification to generate seasonal label characteristics and type characteristic labels;
and performing channel fusion on the sampling feature set, the seasonal tag features and the type feature tags according to the weight of the training sample generator, and performing reduction output through an output module Conv _ out to generate a training sample set.
4. The method for creating the seasonal prediction model of the commodity on the e-commerce website according to claims 1 to 3, wherein the step of sequentially inputting the tag feature and the commodity feature into N tag residual layers and feature fusion layers for residual extraction to generate a target residual comprises:
inputting the label features and the commodity features into a first label residual error layer for residual error calculation to generate a first residual error of the training sample;
channel superposition is carried out on the first residual error and a first class label in the N class-classified type labels to generate a first fusion characteristic;
inputting the first fusion feature into a first label residual error layer to generate a second residual error of the training sample;
channel superposition is carried out on the second residual error and a second class label in the N class-classified type labels to generate a second fusion characteristic;
inputting the N-1 th residual error into an Nth label residual error layer to generate an Nth residual error of the training sample;
and sequentially fusing the first residual to the Nth residual through the residual fusion layer to generate a target residual.
5. The method for creating a seasonal prediction model of an e-commerce website commodity according to claim 1, wherein the initial neural network prediction model further comprises a regional channel attention module SKConv;
after the label features and the commodity features are sequentially input into the N label residual layers and the feature fusion layer for residual extraction, and a target residual is generated, the target residual is input into the softmax function layer for analysis to obtain seasonal discrimination probability data, and before the seasonal discrimination probability data comprises seasonal discrimination probability values of at least four seasons, the establishment method of the commodity seasonal prediction model of the electronic commerce website further comprises the following steps:
distributing attention to the areas with different sizes of the target residual errors through convolution cores with different sizes of receptive fields in the area channel attention module SKConv, and screening different characteristic channels of the target residual errors through the distributed attention.
6. The method of creating a seasonal prediction model of an e-commerce website commodity according to claim 5, wherein the initial neural network prediction model further comprises an attention channel pooling module ACD;
after allocating attention to the different-size regions of the target residual error through the convolution kernel of different-size receptive fields in the region channel attention module SKConv and performing screening conversion on the different feature channels of the target residual error through the allocated attention, the method for creating the seasonal prediction model of the e-commerce website commodity further comprises:
attention is distributed to each channel of the target residual error through an attention channel pooling module ACD, and channels with lower attention ranks are discarded.
7. The method for creating a commodity seasonal prediction model for an e-commerce website of claim 6, wherein the initial neural network prediction model further comprises a channel shuffle module (CSA);
channel shuffling the target residual by the channel shuffle module CSA.
8. The method as claimed in claim 3, wherein after said determining whether the loss value of the loss value variation data in the preset interval converges to 0, the method further comprises:
if the loss value of the loss value change data in a preset interval does not converge to 0, judging whether the training times of the training samples reach the standard or not;
and if the training times of the training samples reach the standard, updating the weight of the initial neural network prediction model according to a small batch gradient descent method, and reselecting the training samples from the training sample set to input the training samples into the initial neural network prediction model for training.
9. The method for creating a seasonal prediction model of an e-commerce website commodity according to claim 8, wherein after determining whether the number of times of training of the training samples reaches the standard, the method for creating a seasonal prediction model of an e-commerce website commodity further comprises:
and if the training times of the training samples do not reach the standard, updating the weight of the initial neural network prediction model according to a small batch gradient descent method, and inputting the training samples into the initial neural network prediction model again for training.
10. The method of claim 8, wherein after determining whether the training times of the training samples meet the criteria, the method further comprises:
and updating the weight of the training sample generator by a small batch gradient descent method according to the loss value, and updating the training samples in the training sample set by the training sample generator.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211590459.2A CN115880027A (en) | 2022-12-12 | 2022-12-12 | Electronic commerce website commodity seasonal prediction model creation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211590459.2A CN115880027A (en) | 2022-12-12 | 2022-12-12 | Electronic commerce website commodity seasonal prediction model creation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115880027A true CN115880027A (en) | 2023-03-31 |
Family
ID=85767078
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211590459.2A Pending CN115880027A (en) | 2022-12-12 | 2022-12-12 | Electronic commerce website commodity seasonal prediction model creation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115880027A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116542456A (en) * | 2023-04-25 | 2023-08-04 | 秒优大数据科技(杭州)有限公司 | Intelligent order sending method, device and equipment |
| CN116911908A (en) * | 2023-07-25 | 2023-10-20 | 维妮科技(深圳)有限公司 | Sales data prediction method and system based on artificial intelligence |
-
2022
- 2022-12-12 CN CN202211590459.2A patent/CN115880027A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116542456A (en) * | 2023-04-25 | 2023-08-04 | 秒优大数据科技(杭州)有限公司 | Intelligent order sending method, device and equipment |
| CN116542456B (en) * | 2023-04-25 | 2023-12-01 | 秒优大数据科技(杭州)有限公司 | Intelligent order sending method, device and equipment |
| CN116911908A (en) * | 2023-07-25 | 2023-10-20 | 维妮科技(深圳)有限公司 | Sales data prediction method and system based on artificial intelligence |
| CN116911908B (en) * | 2023-07-25 | 2024-02-27 | 维妮科技(深圳)有限公司 | Sales data prediction method and system based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111523047B (en) | Multi-relation collaborative filtering algorithm based on graph neural network | |
| CN110929697B (en) | Neural network target identification method and system based on residual error structure | |
| US11809985B2 (en) | Algorithmic apparel recommendation | |
| CN109033107B (en) | Image retrieval method and apparatus, computer device, and storage medium | |
| CN115880027A (en) | Electronic commerce website commodity seasonal prediction model creation method | |
| CN110909754B (en) | Attribute generation countermeasure network and matching clothing generation method based on same | |
| CN110263257B (en) | Deep learning based recommendation method for processing multi-source heterogeneous data | |
| CN110955826B (en) | Recommendation system based on improved cyclic neural network unit | |
| CN110503531A (en) | The dynamic social activity scene recommended method of timing perception | |
| CN108960304B (en) | Deep learning detection method for network transaction fraud behaviors | |
| CN112784929B (en) | Small sample image classification method and device based on double-element group expansion | |
| CN113761359B (en) | Data packet recommendation method, device, electronic equipment and storage medium | |
| CN112381179A (en) | Heterogeneous graph classification method based on double-layer attention mechanism | |
| CN112115358A (en) | Personalized recommendation method using multi-hop path features in knowledge graph | |
| CN111400525A (en) | Intelligent fashionable garment matching and recommending method based on visual combination relation learning | |
| CN106600347B (en) | Method for constructing sequence prediction model based on multi-view data and cyclic network | |
| CN113297936A (en) | Volleyball group behavior identification method based on local graph convolution network | |
| CN116152570A (en) | Commodity image classification method based on hierarchical convolutional neural network | |
| Wang et al. | Towards predicting the likeability of fashion images | |
| CN116932862A (en) | Cold start object recommendation method, cold start object recommendation device, computer equipment and storage medium | |
| CN114936890A (en) | Counter-fact fairness recommendation method based on inverse tendency weighting method | |
| CN114519600A (en) | Graph neural network CTR estimation algorithm fusing adjacent node variances | |
| CN113780027A (en) | Multi-label object identification method, device and equipment based on augmented graph convolution | |
| Chen et al. | Hierarchical sequence representation with graph network | |
| CN117786234B (en) | Multimode resource recommendation method based on two-stage comparison learning |
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 |