Sales forecasting method and device, commodity processing method and device

Abstract

The present specification provides a sales forecasting method and device, wherein the sales forecasting method comprises: obtaining a time series of a first traffic feature and a time series of a second traffic feature of an object to be predicted within a preset time interval; obtaining a prediction feature vector based on the time series of the first traffic feature and the time series of the second traffic feature; obtaining a sales forecast value of the object to be predicted based on the prediction feature vector. The sales forecasting method can predict the sales value of the object to be predicted by integrating the traffic characteristics of the object to be predicted itself, the traffic characteristics of its category, the traffic characteristics of its store, and the traffic characteristics of an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted, so that the prediction information is more comprehensive and the prediction is more accurate.

Description

Technical Field

[0001] The present invention relates to the field of information processing technology, and in particular to a sales forecasting method. The present invention also relates to a sales forecasting device, a commodity processing method, a commodity processing device, a supply chain demand plan configuration method, a supply chain demand plan configuration device, a computing device, and a computer-readable storage medium. Background Art

[0002] Sales forecasting is an important demand of e-commerce and the core of product supply chain management and platform traffic distribution. Sales forecasting of newly listed products can help merchants and platforms make better decisions on traffic support, creating hot products, stocking and replenishing new products after they are launched, which is of great significance to the supply of new products on the platform and enriching the supply of goods.

[0003] Traditional sales forecasting methods based on time series (Arima, Holt-Winters, LSTM, etc.) require the target to be predicted to have a long time series of observations, but new products do not have such characteristics, so it is impossible to accurately predict the sales of new products. Sales forecasting methods based on feature engineering and machine learning (linear regression, GBDT, DNN, etc.) require a large number of product-related artificial features, but machine learning models cannot be well constructed and integrated with some features, resulting in information loss and reduced effectiveness. Therefore, a sales forecasting method with more comprehensive sales forecasting information and more accurate prediction is needed to predict the sales value of new products. Summary of the invention

[0004] In view of this, the embodiment of the present application provides a sales forecasting method. The present application also relates to a sales forecasting device, a commodity processing method, a commodity processing device, a supply chain demand plan configuration method, a supply chain demand plan configuration device, a computing device, and a computer-readable storage medium to solve the technical defects existing in the prior art.

[0005] According to a first aspect of an embodiment of the present application, a sales volume forecasting method is provided, comprising:

[0006] Obtaining a time series of a first traffic feature and a time series of a second traffic feature of the object to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the object to be predicted, a traffic feature of a category to which the object to be predicted belongs, and a traffic feature of a store to which the object to be predicted belongs, and the second traffic feature includes a traffic feature of an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted;

[0007] Obtaining a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0008] The sales volume forecast value of the object to be forecasted is obtained according to the forecast feature vector.

[0009] Optionally, obtaining the prediction feature vector according to the time series of the first traffic feature and the time series of the second traffic feature includes:

[0010] Performing feature extraction on the time series of the first flow feature and the time series of the second flow feature respectively to obtain a first time series vector of the first flow feature and a second time series vector of the second flow feature;

[0011] Performing feature aggregation on the second time series vector to obtain a third time series vector;

[0012] The first time series vector and the third time series vector are subjected to feature fusion to obtain the prediction feature vector.

[0013] Optionally, performing feature extraction on the time series of the first flow feature and the time series of the second flow feature respectively to obtain a first time series vector of the first flow feature and a second time series vector of the second flow feature includes:

[0014] Based on the k time series of each first traffic feature of the object to be predicted within the preset time interval, a first time series matrix with a dimension of [k, n] is generated; based on the k time series of each second traffic feature of the object to be predicted within the preset time interval, a second time series matrix with a dimension of [k, n] is generated, where n is the number of preset unit durations, and both k and n are natural numbers;

[0015] Obtaining the first time series vector according to the first time series matrix; and

[0016] The second time series vector is obtained according to the second time series matrix.

[0017] Optionally, obtaining the first time series vector according to the first time series matrix includes: inputting the generated first time series matrix into a pre-trained convolutional neural network model to obtain a first time series vector with a dimension of [1, n-w+1], wherein the pre-trained convolutional neural network model uses a convolution kernel with a dimension of [k, w] to process the first time series matrix into the first time series vector, and w is a natural number.

[0018] Optionally, obtaining the second time series vector according to the second time series matrix includes:

[0019] The generated second time series matrix is ​​input into a pre-trained convolutional neural network model to obtain a second time series vector with a dimension of [1, n-w+1], wherein the pre-trained convolutional neural network model uses a convolution kernel with a dimension of [k, w] to process the second time series matrix into a second time series vector, where w is a natural number.

[0020] Optionally, the convolutional neural network model is trained by the following steps:

[0021] Receive a time series of a first flow feature of each sample object in a plurality of input sample objects within the preset time interval;

[0022] Generate a first time series matrix based on the time series of each first flow feature of each sample object in the plurality of sample objects within the preset time interval;

[0023] The convolutional neural network model to be trained is trained according to the generated first time series matrix and the first time series vector to obtain a convolutional neural network model, wherein the convolutional neural network model associates the first time series matrix of each sample object in the plurality of sample objects with the first time series vector of the object. Optionally, the number of the second traffic features is m, where m is a natural number, and the second time series vector of the second traffic feature is subjected to feature aggregation to obtain a third time series vector, including:

[0024] Obtain m second time series vectors of the second traffic characteristics;

[0025] Based on the m second time series vectors of the second traffic features, generate a third time series matrix with a dimension of "m, n-w+1";

[0026] The generated third time series matrix is ​​input into a pre-trained graph convolutional neural network model to obtain a third time series vector with a dimension of [1, n-w+1], wherein the pre-trained graph convolutional neural network model associates the generated third time series matrix with the third time series vector.

[0027] Optionally, the graph convolutional neural network model is trained by the following steps:

[0028] For each of the plurality of sample objects, generating a third time series matrix based on a second time series vector of a second flow feature of the sample object;

[0029] The generated third time series matrix is ​​used as a data sample and the third time series vector with a dimension of [1, n-w+1] is used as a label to be input into the graph convolutional neural network model to be trained, so as to obtain a graph convolutional neural network model, wherein the graph convolutional neural network model associates the third time series matrix of the sample object with the third time series vector of the sample object.

[0030] Among them, the nodes of the graph convolutional neural network model are the second time series vector and the third time series vector of each second flow feature of the sample object, and the edges are the relationship between the second time series vector and the third time series vector of each second flow feature of the sample object.

[0031] Optionally, performing feature fusion on the first time series vector of the first traffic feature and the third time series vector to obtain the prediction feature vector includes:

[0032] The first time series vector of the first traffic feature is concatenated with the third time series vector.

[0033] Optionally, obtaining the sales forecast value of the object to be predicted based on the prediction feature vector includes: inputting the prediction feature vector into a pre-trained sales forecast model to obtain the sales forecast value of the object to be predicted, wherein the sales forecast model associates the prediction feature vector with the sales value of the object to be predicted.

[0034] Optionally, the sales forecasting model is trained through the following steps:

[0035] Acquire a time series of a first flow characteristic and a time series of a second flow characteristic of a plurality of sample objects within the preset time interval;

[0036] For each of the multiple sample objects, feature extraction is performed on the time series of the first flow feature and the time series of the second flow feature to obtain a first time series vector of the first flow feature and a second time series vector of the second flow feature;

[0037] For each of the plurality of sample objects, perform feature aggregation on the second time series vector to obtain a third time series vector;

[0038] For each of the multiple sample objects, perform feature fusion on the first time series vector and the third time series vector to obtain the prediction feature vector;

[0039] The predicted feature vector of each sample object among the multiple sample objects is used as a data sample and the sales value of the sample object is used as a label to input the sales prediction model to be trained to obtain a sales prediction model. The sales prediction model associates the predicted feature vector of each sample object among the multiple sample objects with the sales value of the sample object.

[0040] Optionally, the time series of the first traffic feature includes a time series of the number of times an action to generate traffic is executed and / or the number of people who execute it within the preset time interval for one or more of the object to be predicted, the category to which the object to be predicted belongs, and the store to which the object to be predicted belongs, and the time series of the second traffic feature includes a time series of the number of times an action to generate traffic is executed and / or the number of people who execute it within the preset time interval for an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted.

[0041] According to a second aspect of an embodiment of the present application, a sales volume forecasting device is provided, comprising:

[0042] A first acquisition module is configured to acquire a time series of a first traffic feature and a time series of a second traffic feature of the object to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the object to be predicted, a traffic feature of a category to which the object to be predicted belongs, and a traffic feature of a store to which the object to be predicted belongs, and the second traffic feature includes a traffic feature of an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted;

[0043] A first feature vector obtaining module is configured to obtain a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0044] The first obtaining module is configured to obtain the sales volume prediction value of the object to be predicted according to the prediction feature vector.

[0045] According to a third aspect of an embodiment of the present application, a commodity processing method is provided, including:

[0046] Obtaining a time series of a first traffic feature and a time series of a second traffic feature of the commodity to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the commodity to be predicted, a traffic feature of a category to which the commodity to be predicted belongs, and a traffic feature of a store to which the commodity to be predicted belongs, and the second traffic feature includes a traffic feature of a commodity that belongs to the same category as the commodity to be predicted and has the same attributes as the commodity to be predicted;

[0047] Obtaining a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0048] Obtaining a sales forecast value of the commodity to be forecasted according to the forecast feature vector;

[0049] Obtaining the inventory quantity of the commodity to be predicted and the replenishment cycle of the commodity to be predicted;

[0050] The replenishment quantity of the commodity to be predicted is determined according to the sales forecast value, the inventory quantity and the replenishment cycle.

[0051] Optionally, obtaining the inventory quantity of the commodity to be predicted and the replenishment cycle of the commodity to be predicted includes:

[0052] Sending an information collection request for the inventory quantity of the commodity to be predicted and the replenishment cycle of the commodity to be predicted;

[0053] receiving information inputted in response to the information collection request;

[0054] The inventory quantity of the commodity to be predicted and the replenishment cycle of the commodity to be predicted are obtained according to the information.

[0055] According to a fourth aspect of an embodiment of the present application, there is provided a commodity processing device, including:

[0056] A second acquisition module is configured to acquire a time series of a first traffic feature and a time series of a second traffic feature of the commodity to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the commodity to be predicted, a traffic feature of a category to which the commodity to be predicted belongs, and a traffic feature of a store to which the commodity to be predicted belongs, and the second traffic feature includes a traffic feature of a commodity that belongs to the same category as the commodity to be predicted and has the same attributes as the commodity to be predicted;

[0057] A second feature vector obtaining module is configured to obtain a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0058] A second obtaining module is configured to obtain a sales forecast value of the commodity to be predicted according to the prediction feature vector;

[0059] A third acquisition module is configured to acquire the inventory of the commodity to be predicted and the replenishment cycle of the commodity to be predicted;

[0060] The second determination module is configured to determine the replenishment quantity of the to-be-predicted commodity according to the sales forecast value, the inventory quantity and the replenishment cycle.

[0061] According to a fifth aspect of an embodiment of the present application, a supply chain demand planning configuration method is provided, including:

[0062] Obtaining a time series of a first traffic feature and a time series of a second traffic feature of the commodity to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the commodity to be predicted, a traffic feature of a category to which the commodity to be predicted belongs, and a traffic feature of a store to which the commodity to be predicted belongs, and the second traffic feature includes a traffic feature of a commodity that belongs to the same category as the commodity to be predicted and has the same attributes as the commodity to be predicted;

[0063] Obtaining a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0064] Obtaining a sales forecast value of the commodity to be forecasted according to the forecast feature vector;

[0065] According to the sales forecast value, based on a plurality of replenishment modes pre-stored in a replenishment mode library, a replenishment mode that matches the change rule of the sales forecast value is configured for the commodity to be predicted.

[0066] According to a sixth aspect of an embodiment of the present application, a supply chain demand planning configuration device is provided, including:

[0067] a fourth acquisition module, configured to acquire a time series of a first traffic feature and a time series of a second traffic feature of the commodity to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the commodity to be predicted, a traffic feature of a category to which the commodity to be predicted belongs, and a traffic feature of a store to which the commodity to be predicted belongs, and the second traffic feature includes a traffic feature of a commodity that belongs to the same category as the commodity to be predicted and has the same attributes as the commodity to be predicted;

[0068] a third feature vector obtaining module, configured to obtain a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0069] A third obtaining module is configured to obtain a sales forecast value of the commodity to be predicted according to the prediction feature vector;

[0070] The matching module is configured to configure a replenishment mode that matches the change rule of the sales forecast value for the commodity to be predicted based on the sales forecast value and multiple replenishment modes pre-stored in a replenishment mode library.

[0071] According to a seventh aspect of an embodiment of the present application, there is provided a computing device, including:

[0072] Memory and processor;

[0073] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions:

[0074] Obtaining a time series of a first traffic feature and a time series of a second traffic feature of the object to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the object to be predicted, a traffic feature of a category to which the object to be predicted belongs, and a traffic feature of a store to which the object to be predicted belongs, and the second traffic feature includes a traffic feature of an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted;

[0075] Obtaining a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0076] The sales volume forecast value of the object to be forecasted is obtained according to the forecast feature vector.

[0077] According to an eighth aspect of an embodiment of the present application, a computer-readable storage medium is provided, which stores computer-executable instructions, and when the instructions are executed by a processor, the steps of any of the sales forecasting methods are implemented.

[0078] The sales forecasting method provided in the present specification obtains the time series of the first traffic feature and the time series of the second traffic feature of the object to be predicted within a preset time interval; obtains a prediction feature vector based on the time series of the first traffic feature and the time series of the second traffic feature; obtains the sales forecast value of the object to be predicted based on the prediction feature vector. Since the sales value of the object to be predicted is predicted by integrating the traffic characteristics of the object to be predicted itself, the traffic characteristics of the category to which it belongs, the traffic characteristics of the store to which it belongs, and the traffic characteristics of the object belonging to the same category as the object to be predicted and having the same attributes as the object to be predicted, the prediction information is more comprehensive and the prediction is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 A schematic diagram showing an application scenario of a sales forecasting method according to an embodiment of the present application;

[0080] Figure 2 A flow chart of a sales forecasting method according to an embodiment of the present application is shown;

[0081] FIG3( a ) shows a timing diagram of a training process of a sales volume prediction method according to an embodiment of the present application;

[0082] FIG3( b ) shows a timing diagram of the prediction process of the sales volume prediction method according to an embodiment of the present application;

[0083] Figure 4 A flowchart showing a processing procedure of a sales forecasting method according to an embodiment of the present application is shown;

[0084] Figure 5 The CNN model framework used in the feature extraction step in the sales forecasting method according to an embodiment of the present application is shown;

[0085] Figure 6 The GCN model framework used in the aggregation step in the sales forecasting method according to an embodiment of the present application is shown;

[0086] Figure 7 A schematic diagram of a GCN model in a sales volume prediction method according to an embodiment of the present application is shown;

[0087] Figure 8 The DNN model framework used in the sales forecasting method according to an embodiment of the present application is shown;

[0088] Fig. 9 A schematic diagram of the structure of a sales forecasting device according to an embodiment of the present application is shown;

[0089] Fig.10 A structural block diagram of a computing device provided according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0090] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

[0091] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0092] It should be understood that although the terms first, second, etc. may be used in one or more embodiments of this specification to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "when" or "while" or "in response to determining".

[0093] First, the noun terms related to one or more embodiments of this specification are explained.

[0094] Time series: A sequence formed by arranging the values of the same statistical indicator of a commodity in the order of their occurrence in time.

[0095] CPV: <Category, Product Name, Product Value>, <category, attribute, attribute term>, the smallest granularity of e-commerce product attributes, such as <T-shirt, material, cotton>, or <cotton-padded jacket, function, warm>.

[0096] In this specification, a sales volume prediction method is provided. This specification also relates to a sales volume prediction device, a computing device, and a computer-readable storage medium, which will be described in detail one by one in the following embodiments.

[0097] Figure 1 A schematic diagram showing an application scenario of a sales volume prediction method according to an embodiment of the present application is shown.

[0098] See Figure 1 , the application scenario includes: a terminal and a server.

[0099] Specifically, the terminal sends the time series of the first traffic feature and the time series of the second traffic feature of the object to be predicted on the terminal within a preset time interval to the server, where the first traffic feature includes one or more of the traffic feature of the object to be predicted, the traffic feature of the category to which the object to be predicted belongs, and the traffic feature of the store to which the object to be predicted belongs, and the second traffic feature includes the traffic feature of an object that belongs to the same category as the object to be predicted and has one of the same attributes as the object to be predicted.

[0100] The server obtains a prediction feature vector according to the time series of the first traffic feature and the time series of the second traffic feature, and then obtains the sales volume prediction value of the object to be predicted according to the prediction feature vector.

[0101] Figure 2A flow chart of a sales forecasting method provided according to an embodiment of the present application is shown, which specifically includes the following steps 202 to 206. FIG3 (a) shows a timing chart of the training process of the sales forecasting method according to an embodiment of the present application, and FIG3 (b) shows a timing chart of the forecasting process of the sales forecasting method according to an embodiment of the present application. Figure 2 and Figure 3(a) to Figure 3(b) Provide explanation.

[0102] Step 202: Obtain a time series of a first flow characteristic and a time series of a second flow characteristic of an object to be predicted within a preset time interval.

[0103] The first traffic feature includes one or more of the traffic feature of the object to be predicted, the traffic feature of the category to which the object to be predicted belongs, and the traffic feature of the store to which the object to be predicted belongs. The second traffic feature includes the traffic feature of an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted.

[0104] The time series of the first traffic feature includes the time series of the number of executions of traffic-generating actions and / or the number of people who have executed one or more of the object to be predicted, the category to which the object to be predicted belongs, and the store to which the object to be predicted belongs within a preset time interval. The time series of the second traffic feature includes the time series of the number of executions of traffic-generating actions and / or the number of people who have executed an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted within a preset time interval. The traffic-generating actions may include exposure, clicks, add-to-cart, and purchases. Specifically, the object to be predicted may be any item that needs to be predicted, such as a new product, and the preset time interval is the preset observation period of the new product. In the embodiment of the present application, as shown in FIG3(b), the time series of exposure / clicks / add-to-cart / orders of the current new product in the past n days are extracted respectively, for example, n is 60, that is, the time series of the new product and the products in the same category and one of the same attributes as the new product are taken, and if there is no flow sequence of 60 days, such as the new product is released less than 60 days, 0 is added in front. In practical applications, various time series of one or more of the new product itself, the category to which the product belongs, and the store to which the product belongs in the past n days can be extracted.

[0105] The time series of the first traffic feature includes the number of times of action execution and / or the number of people generating traffic within 60 days for one or more of the new product, the category to which the new product belongs, and the store to which the new product belongs. The time series of the second traffic feature includes the number of times of action execution and / or the number of people generating traffic within 60 days for products that belong to the same category as the new product and share one of the attributes of the new product. The actions generating traffic may include exposure, click, add to cart, and purchase.

[0106] The products that belong to the same category as the new product and share one of the attributes of the new product can be determined by <category, attribute, attribute value> (CPV). Of course, they can also be determined by other existing methods that associate categories with attributes. Taking CPV as an example, for instance, if the new product is a red round-neck pure cotton T-shirt, the category of the new product can be T-shirt, and the CPV can be <T-shirt, material, pure cotton>, <T-shirt, style, round neck>, or <T-shirt, color, red>, etc. That is, CPV can define the attributes of the new product from different perspectives.

[0107] Step 204: Obtain a prediction feature vector based on the time series of the first traffic feature and the time series of the second traffic feature.

[0108] The time series of the first traffic feature and the time series of the second traffic feature are as described above. The prediction feature vector is a feature vector that combines the first traffic feature and the second traffic feature for predicting the sales volume value.

[0109] Specifically, as Figure 4 shown, step 204 can be implemented through steps 2042 to 2046.

[0110] Step 2042: Respectively perform feature extraction on the time series of the first traffic feature and the time series of the second traffic feature to obtain the first time series vector of the first traffic feature and the second time series vector of the second traffic feature.

[0111] Feature extraction refers to extracting the main features from the multi-dimensional time series matrix formed by the time series of traffic features and processing them into a one-dimensional time series vector.

[0112] In practical applications, step 2042 can be implemented by a Convolutional Neural Network (CNN) model, a Long Short-Term Memory (LSTM), or a Gated Recurrent Unit (GRU) model. Below, Figure 5 taking the CNN model as an example for illustration.

[0113] First, the training process of the CNN model is explained.

[0114] Receive the time series of the number of exposures, number of people exposed, number of clicks, number of clicks, number of add-on purchases, number of people added to purchases, number of orders, and number of people in each of the multiple sample products input within n days, including the product itself, the category to which the product belongs, one or more of the stores to which the product belongs, and CPV, where the number is calculated by the page views PV and the number of people is calculated by the number of independent visitors UV. Since the time series of each first traffic feature and the time series of each second traffic feature are processed in the same way, and the time series vectors are obtained by using the CNN model trained for the corresponding ones, the following takes the traffic features of the sample products in the first traffic feature within n days as an example for explanation.

[0115] The traffic characteristics of the sample products within n days can be characterized as the time series of the number of exposures, number of exposure users, number of clicks, number of clickers, number of add-to-carts, number of add-to-carts, number of orders, and number of orders of the sample products within n days. That is, the traffic characteristics of the products themselves are characterized as the following k*n matrix, where n is the number of preset unit durations. The unit duration is a time measurement unit, which can be hours, days, days, months, or years. It can be determined according to actual needs. In this example, days are used for explanation. k represents k time series representing products, and both k and n are natural numbers. to is the number of exposures from 1 to n days, to is the number of exposures from 1 to n days, to is the number of orders from 1 to n days.

[0116]

[0117] The traffic characteristics of the category to which the product belongs, the traffic characteristics of the store to which the product belongs, and the traffic characteristics of each CPV are similar to the traffic characteristics of the product itself.

[0118] Figure 6The CNN model framework used in the feature extraction step in the sales forecasting method according to the embodiment of the present application is shown. As shown in the figure, the CNN model can have 2 convolutional layers and 2 pooling layers (only for example). The convolutional layer is used for feature extraction. There are generally two operations of the pooling layer, one is average pooling and the other is maximum pooling. The pooling layer compresses the input matrix, which makes the matrix smaller and simplifies the network calculation complexity on the one hand; on the other hand, it compresses the features and extracts the main features; the pooling layer is connected to the fully connected layer, and each neuron in the fully connected layer is fully connected with all neurons in the previous layer. The full connection can integrate the local information with category distinction in the pooling layer, and the output value of the fully connected layer is passed to the subsequent output layer to output the final calculation result. In this embodiment, the CNN model is used for feature extraction.

[0119] The above k*n time series matrix is ​​used as a data sample and the first time series vector with a dimension of [1, n-w+1] is used as a label to input the CNN model to be trained. The CNN model extracts the time series features of the product's own traffic characteristics and learns the CNN time series modeling t(x).

[0120] The CNN model uses a convolution kernel of [k, w] dimension. That is, the width of the convolution kernel is the number k of time series that represent each product, its category, its store, and CPV. The length of the convolution kernel is a hyperparameter w. If it is 3 / 7 / 15, it means that this layer of convolution can extract the magnitude and trend information of the recent 3-day / 7-day / 15-day sequences.

[0121] The trained CNN model generates a first time series matrix with a dimension of [k, n] based on the k time series of each first traffic feature of the new product within n days, and generates a second time series matrix with a dimension of [k, n] based on the k time series of each second traffic feature of the new product within n days.

[0122] Let’s take the traffic characteristics of a new product as an example to illustrate.

[0123]

[0124] The above k*n matrix is ​​input into the pre-trained CNN model to obtain the first time series vector with a dimension of [1, n-w+1].

[0125] The input of the CNN model is

[0126]

[0127] The output of the CNN model is

[0128] s 1 =[a 1 ,a 2 ,a3 ,a 4 ,…,a n-w ,a n-w+1 ],

[0129] Among them, s 1 is the first time series vector of the commodity’s own traffic characteristics, a 1 to a n-w+1 is the time series vector s extracted by the CNN model 1 Vector elements.

[0130] For each sample product's category traffic features, product store traffic features, and each CPV traffic features, the CNN model is trained separately. The CNN model can process the product category traffic features and product store traffic features of dimension k*n into the first time series vector s of dimension [1,n-w+1] 2 、s 3 , each CPV traffic feature with dimension k*n is processed into a second time series vector s with dimension [1,n-w+1] CPVi .

[0131] The time series matrix is ​​processed into a time series vector through CNN, which facilitates feature fusion with the aggregated time series vector of the second traffic feature.

[0132] Step 2044: Perform feature aggregation on the second time series vector of the second traffic feature to obtain a third time series vector.

[0133] Feature aggregation refers to reducing the dimension of the multi-dimensional time series matrix composed of the time series vectors of the second traffic feature into a third time series vector with a dimension of 1.

[0134] Specifically, step 2044 can be implemented by a graph convolutional neural network (GCN) model or an aggregation model such as LSTM. Figure 6 Take the GCN model as an example to illustrate. Figure 6 The GCN model framework used in the aggregation step in the sales forecasting method according to an embodiment of the present application is shown. The GCN model is a generalization of the CNN model in the graph domain. Similar to the CNN model, there are multiple hidden layers between the input layer and the output layer. The model can simultaneously perform end-to-end learning of node feature information and structural information.

[0135] First, train the GCN model.

[0136] For each sample commodity among the multiple sample commodities, a third time series matrix is ​​generated based on the second time series vector of each CPV flow feature of the sample commodity.

[0137] The generated third time series matrix is ​​used as a data sample and the third time series vector with a dimension of [1, n-w+1] (described in detail below) is input as a label into the GCN model to be trained for training to obtain a GCN model. The GCN model associates the third time series matrix of the sample product with the third time series vector of the sample product, wherein the nodes of the GCN model are the second time series vector and the third time series vector of each second flow feature of the sample product, and the edges are the relationship between the second time series vector and the third time series vector of each second flow feature of the sample product.

[0138] Specifically, each product has an indefinite number of CPV traffic features, set to m, for example, CPV 1 The second time series vector of traffic characteristics is s CPV1 =[b 1 ,b 2 ,b 3 ,b 4 ,…,b n-w ,b n-w+1 ], where b 1 To b n-w+1 is the vector s extracted by the CNN model CPV1 Vector element, CPV m The second time series vector of traffic characteristics is s CPVm =[f 1 ,f 2 ,f 3 ,f 4 ,…,f n-w ,f n-w+1 ], where f 1 to f n-w+1 is the vector s extracted by the CNN model CPVm Vector elements.

[0139] The input of the GCN module is a matrix composed of vectors of the m CPV traffic features of the product after feature extraction by CNN, where each element of the matrix is CPV1 is the feature vector of the above [1,n-w+1] dimension extracted by CNN.

[0140] The GCN model input is:

[0141]

[0142] That is, the matrix composed of the CPV feature vectors of each product [s CPV1 ,s CPV2 ,s CPV3 …s CPVmInput the GCN model. The aggregation based on the GCN aggregation strategy, such as average pooling, means using the GCN model function g(x) = avg(s CPV1 , s CPV2 , s CPV3 … s CPVm ). The output is the unified representation s CPV = [h 1 , h 2 , h 3 … h n-w+1 of the second time series vectors of all CPVs of the commodity, which is also a feature vector of dimension [1, n - w + 1]. Among them, h 1 to h n-w+1 are the aggregated elements of each corresponding vector element in s CPV1 to s CPVm .

[0143] In the embodiments of the present application, using the global average pooling (GAP) layer, the time series matrix of dimension "m, n - w + 1" is reduced to a third time series vector of dimension [1, n - w + 1] by taking the average value of each column element of the time series matrix. Or it can be trained using an existing neural network toolbox, and call the aggregation using DNN trainable parameters, that is, g(x) = dnn(s CPV1 , s CPV2 , s CPV3 … s CPVm ).

[0144] Through the GCN, the vectors obtained by feature extraction of the m CPV traffic features through the CNN are aggregated to realize the aggregation of discrete node information, which is convenient for modeling and prediction by taking into account the attribute features of the commodity.

[0145] In the embodiments of the present application, taking CPV1 as <T-shirt, material, cotton>, CPV2 as <T-shirt, style, round neck>, and CPV3 as <T-shirt, color, red> as an example for illustration.

[0146] As Figure 7 shown, using the global average pooling layer, the time series matrix of dimension "3, n - w + 1" is reduced to a third time series vector s CPV , s CPV of dimension [1, n - w + 1] by taking the average value. s

[0147] Step 2046: Perform feature fusion on the first time series vector of the first traffic feature and the third time series vector to obtain the prediction feature vector.

[0148] Feature fusion refers to combining the first time series vector with the third time series vector to achieve the combination of the first flow feature and the second flow feature.

[0149] Specifically, the first time series vector of the first flow feature is fused with the third time series vector through a function or model capable of fusing features of multiple time series vectors, thereby combining the first flow feature with the second flow feature.

[0150] Step 2046 can be implemented through a concatenation (concat) function, a Stack model, a CNN model, etc., and is explained below using a concatenation (concat) function.

[0151] The time series features of the product itself, the category to which the product belongs, and the store to which the product belongs are fused with the CPV time series features of the product. After the original time series of the product itself, the category to which the product belongs, and the store to which the product belongs pass through the CNN module, the [1, n-w+1]-dimensional time series vector representations of the same dimension are output respectively, denoted as s 1 、s 2 、s 3 After the CPV time series features of the product pass through the CNN and GCN modules, they also output a time series vector of [1, n-w+1] dimensions, denoted as s CPV Here, the feature fusion method is to concatenate the time series vectors of the product itself, the category to which the product belongs, the store to which the product belongs, and the CPV of the product. Then the prediction feature vector input = concat(s 1 ,s 2 ,s 3 ,s CPV ).

[0152] In summary, step 204 can overcome the defect that the traditional method cannot consider the discrete CPV traffic feature node information, realize the aggregation of discrete CPV traffic feature node information, and fuse the traffic characteristics of the object to be predicted itself, the traffic characteristics of the category to which it belongs, the traffic characteristics of the store to which it belongs, and the traffic characteristics of the object belonging to the same category as the object to be predicted and the same as one of the attributes of the object to be predicted, so as to take the attribute characteristics of the object to be predicted into account, and realize the aggregation of traffic characteristics of all commodities with the same attribute as one of the new commodity for prediction.

[0153] Step 206: Obtain the sales volume forecast value of the object to be forecasted according to the forecast feature vector.

[0154] Step 206 can be implemented by: inputting the prediction feature vector into a pre-trained sales prediction model to obtain a sales prediction value of the object to be predicted, wherein the sales prediction model associates the prediction feature vector with the sales value of the object to be predicted.

[0155] Sales forecasting models include deep neural network (DNN) models, Wide & Deep models, DeepFM models or linear regression models. Figure 8 Take the DNN model as an example to illustrate. Figure 8 The DNN model framework used in the sales forecasting method according to an embodiment of the present application is shown. In this embodiment, there are three hidden layers between the input layer and the output layer, and the layers are fully connected, that is, any neuron in the i-th layer must be connected to any neuron in the i+1-th layer.

[0156] As shown in Figure 3(a), the model training process includes the following steps:

[0157] Obtain the time series of exposure / clicks / add-to-cart / orders of multiple sample products within n days for one or more of the products, product categories, and product stores, as well as products that belong to the same category and have the same attributes as the product;

[0158] For each of the multiple sample products, obtain a first time series vector of one or more of the product, the category to which the product belongs, and the store to which the product belongs, for which features are extracted by the CNN model, and a second time series vector of a product that belongs to the same category as the product and has the same attributes as the product;

[0159] For each of the multiple sample products, obtain a third time series vector s after aggregation by the GCN model CPV ;

[0160] For each of the plurality of sample commodities, the first time series vector and the third time series vector s CPV Perform feature fusion to obtain the predicted feature vector;

[0161] The predicted feature vector of each sample commodity among the multiple sample commodities is used as a data sample and the sales value of the sample commodity is used as a label to be input into the sales prediction model to be trained for training, so as to obtain a sales prediction model. The sales prediction model associates the predicted feature vector of each sample commodity among the multiple sample commodities with the sales value of the sample commodity.

[0162] Specifically, the time series obtained by fusing the traffic features of multiple commodities and the actual sales volume y are input into the DNN model as samples to predict the sales volume forecast value of each commodity. Calculate the loss function with the actual sales volume y of the product The model parameters are supervised and learned based on the loss function. The DNN model is iteratively calculated based on the degree of difference between them and the parameters in the DNN model are updated. The converged DNN model is obtained and output as a sales prediction model. Specifically, the objective function of the training is defined, and the parameters of the deep neural network are optimized according to the stochastic gradient descent method. In each round of iteration, the data of the sample forward propagated to the output layer nodes is calculated, and the gradient is calculated and then back-propagated. In the process of back-propagation, the network model parameters are updated. After multiple rounds of iterations, stable model parameters are finally obtained, indicating that the objective function converges and the training process of the DNN model has converged. The final output DNN model is obtained as the sales prediction model.

[0163] In practical applications, for the regression problem of sales forecasting, the loss function Loss can be the mean square error (MSE), mean absolute error (MAE), or Huber Loss, etc. MSE is used as an example for explanation.

[0164]

[0165] Where M is the number of samples, m is between 1 and M, y m It’s real sales. is the predicted sales volume, and MSE represents the mean of the sum of squares of the errors between the actual sales volume and the predicted sales volume. The closer the MSE is to 0, the better the model selection and fit are, and the more successful the data prediction is.

[0166] After obtaining the trained DNN model, input the prediction sample, extract it through the CNN model, fuse it with the GCN model, and concatenate it with the concat function, and then input it into the trained DNN model to predict the sales forecast value of the product. This is the sales forecast result of the model.

[0167] This sales forecasting method can predict the sales value of the object to be predicted by integrating one or more of the traffic characteristics of the object to be predicted, the traffic characteristics of the category to which it belongs, the traffic characteristics of the store to which it belongs, and the traffic characteristics of objects that belong to the same category as the object to be predicted and have the same attributes as the object. The forecasting information is more comprehensive and the forecast is more accurate.

[0168] Corresponding to the above method embodiment, this specification also provides a sales forecasting device embodiment. Fig. 9 FIG. 1 is a schematic diagram showing the structure of a sales forecasting device provided by an embodiment of the present application. Fig. 9 As shown, the device comprises:

[0169] The first acquisition module 902 is configured to acquire a time series of a first traffic feature and a time series of a second traffic feature of the object to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the object to be predicted, a traffic feature of a category to which the object to be predicted belongs, and a traffic feature of a store to which the object to be predicted belongs, and the second traffic feature includes a traffic feature of an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted;

[0170] A first feature vector obtaining module 904 is configured to obtain a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0171] The first obtaining module 906 is configured to obtain the sales volume prediction value of the object to be predicted according to the prediction feature vector.

[0172] The sales forecasting device can realize the prediction of the sales value of the object to be predicted by integrating the traffic characteristics of the object to be predicted, the traffic characteristics of the category to which it belongs, the traffic characteristics of the store to which it belongs, and the traffic characteristics of objects belonging to the same category as the object to be predicted and having the same attributes as the object to be predicted, so that the prediction information is more comprehensive and the prediction is more accurate.

[0173] Optionally, the first feature vector obtaining module includes:

[0174] an extraction unit, configured to perform feature extraction on the time series of the first flow feature and the time series of the second flow feature respectively, to obtain a first time series vector of the first flow feature and a second time series vector of the second flow feature;

[0175] an aggregation unit, configured to perform feature aggregation on the second time series vector to obtain a third time series vector;

[0176] The fusion unit is configured to perform feature fusion on the first time series vector and the third time series vector to obtain the prediction feature vector.

[0177] Optionally, the extraction unit is further configured to:

[0178] Based on the k time series of each first traffic feature of the object to be predicted within the preset time interval, a first time series matrix with a dimension of [k, n] is generated; based on the k time series of each second traffic feature of the object to be predicted within the preset time interval, a second time series matrix with a dimension of [k, n] is generated, where n is the number of preset unit durations, and both k and n are natural numbers;

[0179] Obtaining the first time series vector according to the first time series matrix; and

[0180] The second time series vector is obtained according to the second time series matrix.

[0181] Optionally, the extraction unit is further configured to:

[0182] The generated first time series matrix is ​​input into a pre-trained convolutional neural network model to obtain first time series vectors with dimensions [1, n-w+1], respectively, wherein the pre-trained convolutional neural network model uses a convolution kernel with dimensions [k, w] to process the first time series matrix into the first time series vector, where w is a natural number.

[0183] Optionally, the extraction unit is further configured to:

[0184] The generated second time series matrix is ​​input into a pre-trained convolutional neural network model to obtain a second time series vector with a dimension of [1, n-w+1], wherein the pre-trained convolutional neural network model uses a convolution kernel with a dimension of [k, w] to process the second time series matrix into a second time series vector, where w is a natural number.

[0185] Optionally, the sales volume prediction device further includes a first training module configured to:

[0186] Receive a time series of a first flow feature of each sample object in a plurality of input sample objects within the preset time interval;

[0187] Generate a first time series matrix based on the time series of each first flow feature of each sample object in the plurality of sample objects within the preset time interval;

[0188] The convolutional neural network model to be trained is trained according to the generated first time series matrix and the first time series vector to obtain a convolutional neural network model, wherein the convolutional neural network model associates the first time series matrix of each sample object in the plurality of sample objects with the first time series vector of the object. Optionally, the number of the second traffic features is m, where m is a natural number, and the aggregation unit is further configured as follows:

[0189] Obtain m second time series vectors of the second traffic characteristics;

[0190] Based on the m second time series vectors of the second traffic features, generate a third time series matrix with a dimension of "m, n-w+1";

[0191] The generated third time series matrix is ​​input into a pre-trained graph convolutional neural network model to obtain a third time series vector with a dimension of [1, n-w+1], wherein the pre-trained graph convolutional neural network model associates the generated third time series matrix with the third time series vector.

[0192] Optionally, the sales volume prediction device further includes a second training module configured to:

[0193] For each of the plurality of received sample objects, generating a third time series matrix based on a second time series vector of a second flow feature of the sample object;

[0194] The generated third time series matrix is ​​used as a data sample and the third time series vector with a dimension of [1, n-w+1] is used as a label to be input into the graph convolutional neural network model to be trained, so as to obtain a graph convolutional neural network model, wherein the graph convolutional neural network model associates the third time series matrix of the sample object with the third time series vector of the sample object.

[0195] Among them, the nodes of the graph convolutional neural network model are the second time series vector and the third time series vector of each second flow feature of the sample object, and the edges are the relationship between the second time series vector and the third time series vector of each second flow feature of the sample object.

[0196] Optionally, the fusion unit is further configured to:

[0197] The first time series vector of the first traffic feature is concatenated with the third time series vector.

[0198] Optionally, the first obtaining module 906 is further configured to:

[0199] The prediction feature vector is input into a pre-trained sales prediction model to obtain a sales prediction value of the object to be predicted, wherein the sales prediction model associates the prediction feature vector with the sales value of the object to be predicted.

[0200] Optionally, the sales volume prediction device further includes a third training module configured to:

[0201] Acquire a time series of a first flow characteristic and a time series of a second flow characteristic of a plurality of sample objects within the preset time interval;

[0202] For each of the multiple sample objects, feature extraction is performed on the time series of the first flow feature and the time series of the second flow feature to obtain a first time series vector of the first flow feature and a second time series vector of the second flow feature;

[0203] For each of the plurality of sample objects, perform feature aggregation on the second time series vector to obtain a third time series vector;

[0204] For each of the multiple sample objects, perform feature fusion on the first time series vector and the third time series vector to obtain the prediction feature vector;

[0205] The predicted feature vector of each sample object among the multiple sample objects is used as a data sample and the sales value of the sample object is used as a label to input the sales prediction model to be trained to obtain a sales prediction model. The sales prediction model associates the predicted feature vector of each sample object among the multiple sample objects with the sales value of the sample object.

[0206] Optionally, the time series of the first traffic feature includes a time series of the number of times an action to generate traffic is executed and / or the number of people who execute it within the preset time interval for one or more of the object to be predicted, the category to which the object to be predicted belongs, and the store to which the object to be predicted belongs, and the time series of the second traffic feature includes a time series of the number of times an action to generate traffic is executed and / or the number of people who execute it within the preset time interval for an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted.

[0207] The above is a schematic solution of a sales forecasting device of this embodiment. It should be noted that the technical solution of the sales forecasting device and the technical solution of the sales forecasting method described above are of the same concept, and the details not described in detail in the technical solution of the sales forecasting device can be found in the description of the technical solution of the sales forecasting method described above.

[0208] This specification also provides a schematic scheme of a commodity processing method, including:

[0209] Obtaining a time series of a first traffic feature and a time series of a second traffic feature of the commodity to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the commodity to be predicted, a traffic feature of a category to which the commodity to be predicted belongs, and a traffic feature of a store to which the commodity to be predicted belongs, and the second traffic feature includes a traffic feature of a commodity that belongs to the same category as the commodity to be predicted and has the same attributes as the commodity to be predicted;

[0210] Obtaining a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0211] Obtaining a sales forecast value of the commodity to be forecasted according to the forecast feature vector;

[0212] Obtaining the inventory quantity of the commodity to be predicted and the replenishment cycle of the commodity to be predicted;

[0213] The replenishment quantity of the commodity to be predicted is determined according to the sales forecast value, the inventory quantity and the replenishment cycle.

[0214] Optionally, obtaining the inventory quantity of the commodity to be predicted and the replenishment cycle of the commodity to be predicted includes:

[0215] Sending an information collection request for the inventory quantity of the commodity to be predicted and the replenishment cycle of the commodity to be predicted;

[0216] receiving information inputted in response to the information collection request;

[0217] The inventory quantity of the commodity to be predicted and the replenishment cycle of the commodity to be predicted are obtained according to the information.

[0218] The commodity to be predicted is a commodity whose sales volume is to be predicted. The information collection request can be sent to the merchant via SMS, interactive interface, etc. The merchant enters information by replying to the request via SMS or entering information in the interactive interface. The methods listed here are only examples and are not limited to the methods mentioned here.

[0219] After obtaining the merchant's inventory and replenishment cycle for the product, reasonably determine the replenishment quantity based on the sales forecast, inventory and replenishment cycle.

[0220] This specification also provides a commodity processing device, including:

[0221] A second acquisition module is configured to acquire a time series of a first traffic feature and a time series of a second traffic feature of the commodity to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the commodity to be predicted, a traffic feature of a category to which the commodity to be predicted belongs, and a traffic feature of a store to which the commodity to be predicted belongs, and the second traffic feature includes a traffic feature of a commodity that belongs to the same category as the commodity to be predicted and has the same attributes as the commodity to be predicted;

[0222] A second feature vector obtaining module is configured to obtain a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0223] A second obtaining module is configured to obtain a sales forecast value of the commodity to be predicted according to the prediction feature vector;

[0224] A third acquisition module is configured to acquire the inventory of the commodity to be predicted and the replenishment cycle of the commodity to be predicted;

[0225] The second determination module is configured to determine the replenishment quantity of the to-be-predicted commodity according to the sales forecast value, the inventory quantity and the replenishment cycle.

[0226] Optionally, the third acquisition module is configured to:

[0227] Sending an information collection request for the inventory quantity of the commodity to be predicted and the replenishment cycle of the commodity to be predicted;

[0228] receiving information inputted in response to the information collection request;

[0229] The inventory quantity of the commodity to be predicted and the replenishment cycle of the commodity to be predicted are obtained according to the information.

[0230] This specification also provides a supply chain demand planning configuration method, including:

[0231] Obtaining a time series of a first traffic feature and a time series of a second traffic feature of the commodity to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the commodity to be predicted, a traffic feature of a category to which the commodity to be predicted belongs, and a traffic feature of a store to which the commodity to be predicted belongs, and the second traffic feature includes a traffic feature of a commodity that belongs to the same category as the commodity to be predicted and has the same attributes as the commodity to be predicted;

[0232] Obtaining a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0233] Obtaining a sales forecast value of the commodity to be forecasted according to the forecast feature vector;

[0234] According to the sales forecast value, based on a plurality of replenishment modes pre-stored in a replenishment mode library, a replenishment mode that matches the change rule of the sales forecast value is configured for the commodity to be predicted.

[0235] In one embodiment of the present application, configuring a replenishment mode that matches the change pattern of the sales forecast value for the object to be predicted includes: selecting a matching replenishment mode for the object to be predicted from a plurality of replenishment modes pre-stored in a replenishment mode library according to the sales forecast value of the object to be predicted in each prediction time period in a plurality of prediction time periods.

[0236] Replenishment modes can include cycle rolling mode, replenishment mode, centralized replenishment mode and order-driven mode. In addition to replenishment mode information, the demand plan can also include replenishment time. The specific replenishment time can be determined based on the storage method of the object to be processed and the length of preparation time to ensure timely replenishment. Storage methods include direct storage and transfer from other warehouses. If the method of transfer from other warehouses is adopted, the transportation distance between other warehouses and the warehouse to be replenished needs to be considered when determining the replenishment time. The length of preparation time refers to the length of time required to prepare a certain replenishment quantity. The preparation time is related to the shipment volume, logistics and transportation conditions, etc. For example, if it takes three days to prepare a week's worth of goods, replenishment needs to be started three days before the arrival of this week.

[0237] The embodiment of the present application can determine sales forecast values ​​within multiple consecutive forecast time periods, and then match a more accurate replenishment model according to the change pattern of the sales forecast values ​​corresponding to these multiple consecutive forecast time periods, thereby improving replenishment efficiency.

[0238] This specification also provides a supply chain demand planning configuration device, including:

[0239] a fourth acquisition module, configured to acquire a time series of a first traffic feature and a time series of a second traffic feature of the commodity to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the commodity to be predicted, a traffic feature of a category to which the commodity to be predicted belongs, and a traffic feature of a store to which the commodity to be predicted belongs, and the second traffic feature includes a traffic feature of a commodity that belongs to the same category as the commodity to be predicted and has the same attributes as the commodity to be predicted;

[0240] a third feature vector obtaining module, configured to obtain a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0241] A third obtaining module is configured to obtain a sales forecast value of the commodity to be predicted according to the prediction feature vector;

[0242] The matching module is configured to configure a replenishment mode that matches the change rule of the sales forecast value for the commodity to be predicted based on the sales forecast value and multiple replenishment modes pre-stored in a replenishment mode library.

[0243] Fig.10 The structure block diagram of a computing device 1000 provided according to an embodiment of the present application is shown. The components of the computing device 1000 include but are not limited to a memory 1010 and a processor 1020. The processor 1020 is connected to the memory 1010 via a bus 1030, and the database 1050 is used to store data.

[0244] The computing device 1000 also includes an access device 1040 that enables the computing device 1000 to communicate via one or more networks 1060. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 1040 may include one or more of any type of network interface (e.g., a network interface card (NIC)) that is wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a World Wide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, a near field communication (NFC) interface, and the like.

[0245] In one embodiment of the present specification, the above components of the computing device 1000 and Fig.10 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Fig.10 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0246] The computing device 1000 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or PC. The computing device 1000 may also be a mobile or stationary server.

[0247] The processor 1020 is used to execute the following computer executable instructions:

[0248] Obtaining a time series of a first traffic feature and a time series of a second traffic feature of the object to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the object to be predicted, a traffic feature of a category to which the object to be predicted belongs, and a traffic feature of a store to which the object to be predicted belongs, and the second traffic feature includes a traffic feature of an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted;

[0249] Obtaining a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0250] The sales volume forecast value of the object to be forecasted is obtained according to the forecast feature vector.

[0251] The above is a schematic scheme of a computing device of this embodiment. It should be noted that the technical scheme of the computing device and the technical scheme of the sales forecasting method described above are of the same concept, and the details not described in detail in the technical scheme of the computing device can be found in the description of the technical scheme of the sales forecasting method described above.

[0252] An embodiment of the present specification further provides a computer-readable storage medium storing computer instructions, which are used when executed by a processor to:

[0253] Obtaining a time series of a first traffic feature and a time series of a second traffic feature of the object to be predicted within a preset time interval, wherein the first traffic feature includes one or more of a traffic feature of the object to be predicted, a traffic feature of a category to which the object to be predicted belongs, and a traffic feature of a store to which the object to be predicted belongs, and the second traffic feature includes a traffic feature of an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted;

[0254] Obtaining a prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature;

[0255] The sales volume forecast value of the object to be forecasted is obtained according to the forecast feature vector.

[0256] The above is a schematic scheme of a computer-readable storage medium of this embodiment. It should be noted that the technical scheme of the storage medium and the technical scheme of the sales forecasting method described above are of the same concept, and the details not described in detail in the technical scheme of the storage medium can be found in the description of the technical scheme of the sales forecasting method described above.

[0257] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0258] The computer instructions include computer program codes, which may be in source code form, object code form, executable files or some intermediate forms, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0259] It should be noted that, for the convenience of description, the aforementioned method embodiments are all described as a series of action combinations, but those skilled in the art should be aware that this specification is not limited by the order of the actions described, because according to this specification, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this specification.

[0260] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0261] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of this specification, so that technicians in the relevant technical field can well understand and use this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A sales forecasting method, comprising: Obtain a time series of a first traffic feature and a time series of a second traffic feature of an object to be predicted within a preset time interval, wherein the first traffic feature includes one or more of the traffic feature of the object to be predicted, the traffic feature of the category to which the object to be predicted belongs, and the traffic feature of the store to which the object to be predicted belongs, and the second traffic feature includes the traffic feature of an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted; the time series of the first traffic feature includes a time series of the number of times and / or the number of people who have executed an action to generate traffic within the preset time interval for one or more of the object to be predicted, the category to which the object to be predicted belongs, and the store to which the object to be predicted belongs, and the time series of the second traffic feature includes a time series of the number of times and / or the number of people who have executed an action to generate traffic within the preset time interval for an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted, and the actions that generate traffic include exposure, click, add to cart, and purchase; A prediction feature vector is obtained according to the time series of the first flow feature and the time series of the second flow feature, wherein the method of obtaining the prediction feature vector according to the time series of the first flow feature and the time series of the second flow feature comprises: extracting features from the multidimensional time series matrix formed by the time series of the first flow feature and the multidimensional time series matrix formed by the time series of the second flow feature, respectively, and processing them into one-dimensional time series vectors to obtain a first time series vector of the first flow feature and a second time series vector of the second flow feature; reducing the dimension of the multidimensional time series matrix formed by the second time series vector into a third time series vector with a dimension of 1; and performing feature fusion on the first time series vector and the third time series vector to obtain the prediction feature vector; The sales volume forecast value of the object to be forecasted is obtained according to the forecast feature vector.

2. The sales forecasting method according to claim 1, wherein: Feature extraction is performed on the multidimensional time series matrix formed by the time series of the first flow feature and the multidimensional time series matrix formed by the time series of the second flow feature respectively, and they are processed into one-dimensional time series vectors to obtain a first time series vector of the first flow feature and a second time series vector of the second flow feature, including: Based on the k time series of each first traffic feature of the object to be predicted within the preset time interval, a first time series matrix with a dimension of [k, n] is generated; based on the k time series of each second traffic feature of the object to be predicted within the preset time interval, a second time series matrix with a dimension of [k, n] is generated, where n is the number of preset unit durations, and both k and n are natural numbers; Obtaining the first time series vector according to the first time series matrix; and The second time series vector is obtained according to the second time series matrix.

3. The sales forecasting method according to claim 2, wherein: Obtaining the first time series vector according to the first time series matrix includes: The generated first time series matrix is ​​input into a pre-trained convolutional neural network model to obtain a first time series vector with a dimension of [1, n-w+1], wherein the pre-trained convolutional neural network model uses a convolution kernel with a dimension of [k, w] to process the first time series matrix into the first time series vector, where w is a natural number.

4. The sales forecasting method according to claim 2, wherein: Obtaining the second time series vector according to the second time series matrix includes: The generated second time series matrix is ​​input into a pre-trained convolutional neural network model to obtain a second time series vector with a dimension of [1, n-w+1], wherein the pre-trained convolutional neural network model uses a convolution kernel with a dimension of [k, w] to process the second time series matrix into a second time series vector, where w is a natural number.

5. The sales volume forecasting method according to claim 3, wherein: The convolutional neural network model is trained through the following steps: Receive a time series of a first flow feature of each sample object in a plurality of input sample objects within the preset time interval; Generate a first time series matrix based on the time series of each first flow feature of each sample object in the plurality of sample objects within the preset time interval; The convolutional neural network model to be trained is trained according to the generated first time series matrix and the first time series vector to obtain a convolutional neural network model, wherein the convolutional neural network model associates the first time series matrix of each sample object among the multiple sample objects with the first time series vector of the object.

6. The sales forecasting method according to claim 1, wherein: The number of the second traffic features is m, where m is a natural number, and reducing the dimension of the multidimensional time series matrix formed by the second time series vectors into a third time series vector with a dimension of 1 includes: Obtain m second time series vectors of the second traffic characteristics; Based on the m second time series vectors of the second traffic features, generate a third time series matrix with a dimension of "m, n-w+1"; The generated third time series matrix is ​​input into a pre-trained graph convolutional neural network model to obtain a third time series vector with a dimension of [1, n-w+1], wherein the pre-trained graph convolutional neural network model associates the generated third time series matrix with the third time series vector.

7. The sales volume forecasting method according to claim 6, wherein: The graph convolutional neural network model is trained through the following steps: For each of the plurality of sample objects, generating a third time series matrix based on a second time series vector of a second flow feature of the sample object; The generated third time series matrix is ​​used as a data sample and the third time series vector with a dimension of [1, n-w+1] is used as a label to be input into the graph convolutional neural network model to be trained, so as to obtain a graph convolutional neural network model, which associates the third time series matrix of the sample object with the third time series vector of the sample object, wherein the nodes of the graph convolutional neural network model are the second time series vector and the third time series vector of each second flow feature of the sample object, and the edges are the relationship between the second time series vector and the third time series vector of each second flow feature of the sample object.

8. The sales forecasting method according to claim 1, wherein: Performing feature fusion on the first time series vector and the third time series vector to obtain the prediction feature vector includes: The first time series vector of the first traffic feature is concatenated with the third time series vector.

9. The sales forecasting method according to claim 1, wherein: Obtaining the sales forecast value of the object to be predicted based on the prediction feature vector includes: inputting the prediction feature vector into a pre-trained sales forecast model to obtain the sales forecast value of the object to be predicted, wherein the sales forecast model associates the prediction feature vector with the sales value of the object to be predicted.

10. The sales volume forecasting method according to claim 9, wherein: The sales forecasting model is trained through the following steps: Acquire a time series of a first flow characteristic and a time series of a second flow characteristic of a plurality of sample objects within the preset time interval; For each of the multiple sample objects, feature extraction is performed on the time series of the first flow feature and the time series of the second flow feature to obtain a first time series vector of the first flow feature and a second time series vector of the second flow feature; For each of the plurality of sample objects, perform feature aggregation on the second time series vector to obtain a third time series vector; For each of the multiple sample objects, perform feature fusion on the first time series vector and the third time series vector to obtain the prediction feature vector; The predicted feature vector of each sample object among the multiple sample objects is used as a data sample and the sales value of the sample object is used as a label to input the sales prediction model to be trained to obtain a sales prediction model. The sales prediction model associates the predicted feature vector of each sample object among the multiple sample objects with the sales value of the sample object.

11. A sales forecasting device, comprising: A first acquisition module is configured to acquire a time series of a first traffic feature and a time series of a second traffic feature of an object to be predicted within a preset time interval, wherein the first traffic feature includes one or more of the traffic feature of the object to be predicted, the traffic feature of the category to which the object to be predicted belongs, and the traffic feature of the store to which the object to be predicted belongs, and the second traffic feature includes the traffic feature of an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted; the time series of the first traffic feature includes a time series of the number of executions of traffic-generating actions and / or the number of execution persons of one or more of the object to be predicted, the category to which the object to be predicted belongs, and the store to which the object to be predicted belongs within the preset time interval, and the time series of the second traffic feature includes a time series of the number of executions of traffic-generating actions and / or the number of execution persons of an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted within the preset time interval, and the traffic-generating actions include exposure, click, add to cart, and purchase; The first feature vector obtaining module is configured to obtain a prediction feature vector according to the time series of the first traffic feature and the time series of the second traffic feature, wherein the first feature vector obtaining module is specifically configured to perform feature extraction on the multidimensional time series matrix composed of the time series of the first traffic feature and the multidimensional time series matrix composed of the time series of the second traffic feature, respectively, and process them into one-dimensional time series vectors to obtain a first time series vector of the first traffic feature and a second time series vector of the second traffic feature; reduce the dimension of the multidimensional time series matrix composed of the second time series vector into a third time series vector with a dimension of 1; perform feature fusion on the first time series vector and the third time series vector to obtain a prediction feature vector; The first obtaining module is configured to obtain the sales volume prediction value of the object to be predicted according to the prediction feature vector.

12. A commodity processing method, comprising: Obtain a time series of a first traffic feature and a time series of a second traffic feature of the commodity to be predicted within a preset time interval, wherein the first traffic feature includes one or more of the traffic feature of the commodity to be predicted, the traffic feature of the category to which the commodity to be predicted belongs, and the traffic feature of the store to which the commodity to be predicted belongs, and the second traffic feature includes the traffic feature of commodities that belong to the same category as the commodity to be predicted and have the same attributes as the commodity to be predicted; the time series of the first traffic feature includes a time series of the number of executions of traffic-generating actions and / or the number of execution persons of one or more of the commodity to be predicted, the category to which the commodity to be predicted belongs, and the store to which the commodity to be predicted belongs within the preset time interval, and the time series of the second traffic feature includes a time series of the number of executions of traffic-generating actions and / or the number of execution persons of commodities that belong to the same category as the commodity to be predicted and have the same attributes as the commodity to be predicted within the preset time interval, and the traffic-generating actions include exposure, click, add to cart, and purchase; A prediction feature vector is obtained according to the time series of the first flow feature and the time series of the second flow feature, wherein the prediction feature vector is obtained according to the time series of the first flow feature and the time series of the second flow feature, including: extracting features from the multidimensional time series matrix formed by the time series of the first flow feature and the multidimensional time series matrix formed by the time series of the second flow feature, respectively, processing them into one-dimensional time series vectors, and obtaining a first time series vector of the first flow feature and a second time series vector of the second flow feature; reducing the dimension of the multidimensional time series matrix formed by the second time series vector into a third time series vector with a dimension of 1; and performing feature fusion on the first time series vector and the third time series vector to obtain a prediction feature vector; Obtaining a sales forecast value of the commodity to be forecasted according to the forecast feature vector; Obtaining the inventory quantity of the commodity to be predicted and the replenishment cycle of the commodity to be predicted; The replenishment quantity of the commodity to be predicted is determined according to the sales forecast value, the inventory quantity and the replenishment cycle.

13. The commodity processing method according to claim 12, wherein: Acquiring the inventory of the commodity to be predicted and the replenishment cycle of the commodity to be predicted includes: Sending an information collection request for the inventory quantity of the commodity to be predicted and the replenishment cycle of the commodity to be predicted; receiving information inputted in response to the information collection request; The inventory quantity of the commodity to be predicted and the replenishment cycle of the commodity to be predicted are obtained according to the information.

14. A commodity processing device, comprising: The second acquisition module is configured to acquire a time series of a first traffic feature and a time series of a second traffic feature of the commodity to be predicted within a preset time interval, wherein the first traffic feature includes one or more of the traffic feature of the commodity to be predicted, the traffic feature of the category to which the commodity to be predicted belongs, and the traffic feature of the store to which the commodity to be predicted belongs, and the second traffic feature includes the traffic feature of commodities belonging to the same category as the commodity to be predicted and having the same attributes as the commodity to be predicted; the time series of the first traffic feature includes a time series of the number of executions of traffic-generating actions and / or the number of execution persons of one or more of the commodity to be predicted, the category to which the commodity to be predicted belongs, and the store to which the commodity to be predicted belongs within the preset time interval, and the time series of the second traffic feature includes a time series of the number of executions of traffic-generating actions and / or the number of execution persons of commodities belonging to the same category as the commodity to be predicted and having the same attributes as the commodity to be predicted within the preset time interval, and the traffic-generating actions include exposure, click, add to cart, and purchase; A second feature vector obtaining module is configured to obtain a prediction feature vector according to the time series of the first traffic feature and the time series of the second traffic feature, wherein the second feature vector obtaining module is specifically configured to perform feature extraction on the multidimensional time series matrix composed of the time series of the first traffic feature and the multidimensional time series matrix composed of the time series of the second traffic feature, respectively, and process them into one-dimensional time series vectors to obtain a first time series vector of the first traffic feature and a second time series vector of the second traffic feature; reduce the dimension of the multidimensional time series matrix composed of the second time series vector into a third time series vector with a dimension of 1; perform feature fusion on the first time series vector and the third time series vector to obtain a prediction feature vector; A second obtaining module is configured to obtain a sales forecast value of the commodity to be predicted according to the prediction feature vector; A third acquisition module is configured to acquire the inventory of the commodity to be predicted and the replenishment cycle of the commodity to be predicted; The second determination module is configured to determine the replenishment quantity of the to-be-predicted commodity according to the sales forecast value, the inventory quantity and the replenishment cycle.

15. A supply chain demand planning configuration method, comprising: Obtain a time series of a first traffic feature and a time series of a second traffic feature of the commodity to be predicted within a preset time interval, wherein the first traffic feature includes one or more of the traffic feature of the commodity to be predicted, the traffic feature of the category to which the commodity to be predicted belongs, and the traffic feature of the store to which the commodity to be predicted belongs, and the second traffic feature includes the traffic feature of commodities that belong to the same category as the commodity to be predicted and have the same attributes as the commodity to be predicted; the time series of the first traffic feature includes a time series of the number of executions of traffic-generating actions and / or the number of execution persons of one or more of the commodity to be predicted, the category to which the commodity to be predicted belongs, and the store to which the commodity to be predicted belongs within the preset time interval, and the time series of the second traffic feature includes a time series of the number of executions of traffic-generating actions and / or the number of execution persons of commodities that belong to the same category as the commodity to be predicted and have the same attributes as the commodity to be predicted within the preset time interval, and the traffic-generating actions include exposure, click, add to cart, and purchase; A prediction feature vector is obtained according to the time series of the first flow feature and the time series of the second flow feature, wherein the prediction feature vector is obtained according to the time series of the first flow feature and the time series of the second flow feature, including: extracting features from the multidimensional time series matrix formed by the time series of the first flow feature and the multidimensional time series matrix formed by the time series of the second flow feature, respectively, processing them into one-dimensional time series vectors, and obtaining a first time series vector of the first flow feature and a second time series vector of the second flow feature; reducing the dimension of the multidimensional time series matrix formed by the second time series vector into a third time series vector with a dimension of 1; and performing feature fusion on the first time series vector and the third time series vector to obtain a prediction feature vector; Obtaining a sales forecast value of the commodity to be forecasted according to the forecast feature vector; According to the sales forecast value, based on a plurality of replenishment modes pre-stored in a replenishment mode library, a replenishment mode that matches the change pattern of the sales forecast value is configured for the commodity to be predicted.

16. A supply chain demand planning configuration device, comprising: A fourth acquisition module is configured to acquire a time series of a first traffic feature and a time series of a second traffic feature of the commodity to be predicted within a preset time interval, wherein the first traffic feature includes one or more of the traffic feature of the commodity to be predicted, the traffic feature of the category to which the commodity to be predicted belongs, and the traffic feature of the store to which the commodity to be predicted belongs; the second traffic feature includes the traffic feature of commodities belonging to the same category as the commodity to be predicted and having the same attributes as the commodity to be predicted; the time series of the first traffic feature includes a time series of the number of executions of traffic-generating actions and / or the number of execution persons of one or more of the commodity to be predicted, the category to which the commodity to be predicted belongs, and the store to which the commodity to be predicted belongs within the preset time interval; the time series of the second traffic feature includes a time series of the number of executions of traffic-generating actions and / or the number of execution persons of commodities belonging to the same category as the commodity to be predicted and having the same attributes as the commodity to be predicted within the preset time interval, and the traffic-generating actions include exposure, click, add to cart, and purchase; The third feature vector obtaining module is configured to obtain a prediction feature vector according to the time series of the first traffic feature and the time series of the second traffic feature, wherein the third feature vector obtaining module is specifically configured to perform feature extraction on the multidimensional time series matrix composed of the time series of the first traffic feature and the multidimensional time series matrix composed of the time series of the second traffic feature, respectively, and process them into one-dimensional time series vectors to obtain the first time series vector of the first traffic feature and the second time series vector of the second traffic feature; reduce the dimension of the multidimensional time series matrix composed of the second time series vector into a third time series vector with a dimension of 1; perform feature fusion on the first time series vector and the third time series vector to obtain a prediction feature vector; A third obtaining module is configured to obtain a sales forecast value of the commodity to be predicted according to the prediction feature vector; The matching module is configured to configure a replenishment mode that matches the change rule of the sales forecast value for the commodity to be predicted based on the sales forecast value and multiple replenishment modes pre-stored in a replenishment mode library.

17. A computing device comprising: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions: Obtain a time series of a first traffic feature and a time series of a second traffic feature of an object to be predicted within a preset time interval, wherein the first traffic feature includes one or more of the traffic feature of the object to be predicted, the traffic feature of the category to which the object to be predicted belongs, and the traffic feature of the store to which the object to be predicted belongs, and the second traffic feature includes the traffic feature of an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted; the time series of the first traffic feature includes a time series of the number of times and / or the number of people who have executed an action to generate traffic within the preset time interval for one or more of the object to be predicted, the category to which the object to be predicted belongs, and the store to which the object to be predicted belongs, and the time series of the second traffic feature includes a time series of the number of times and / or the number of people who have executed an action to generate traffic within the preset time interval for an object that belongs to the same category as the object to be predicted and has the same attributes as the object to be predicted, and the actions that generate traffic include exposure, click, add to cart, and purchase; A prediction feature vector is obtained according to the time series of the first flow feature and the time series of the second flow feature, wherein the prediction feature vector is obtained according to the time series of the first flow feature and the time series of the second flow feature, including: extracting features from the multidimensional time series matrix formed by the time series of the first flow feature and the multidimensional time series matrix formed by the time series of the second flow feature, respectively, processing them into one-dimensional time series vectors, and obtaining a first time series vector of the first flow feature and a second time series vector of the second flow feature; reducing the dimension of the multidimensional time series matrix formed by the second time series vector into a third time series vector with a dimension of 1; and performing feature fusion on the first time series vector and the third time series vector to obtain a prediction feature vector; The sales volume forecast value of the object to be forecasted is obtained according to the forecast feature vector.

18. A computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the steps of the sales forecasting method according to any one of claims 1 to 10.

Images