Sales volume correction method, sales volume correction device and computer readable storage medium
By using sales correction methods on e-commerce platforms, combined with interval classification and residual prediction models, the problem of insufficient sales prediction capabilities of e-commerce platforms in medium and long-tail sales scenarios is solved, and the completion and accuracy of sales data is improved.
Patent Information
- Application Number
- CN202411547541.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-01
- Publication Date
- 2024-11-29
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing technology has limited predictive capabilities for sales forecasts for long-tail sales scenarios in e-commerce platforms, especially the high proportion of sales of 0-200 and low proportion of sales of more than 10,000, and due to the cost and difficulty of collecting samples. , the public sales data obtained are incomplete, resulting in the prediction accuracy needs to be improved.
By using the sales volume correction method, the target product category is obtained based on the initial results and sales feature data of the initial sales volume completion, and the trained interval classification model and residual prediction model are used to perform sales volume interval classification and residual prediction, and the initial sales volume compensation is corrected. All results will improve the accuracy of final sales completion results.
The sales data completion of long-tail sales scenarios has been achieved, the prediction accuracy has been improved, and the integrity and accuracy of the sales data of the full product are ensured.
Smart Images

Figure CN119048158A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to a sales correction method, a sales correction device and a computer-readable storage medium. Background Art
[0002] Evaluating and predicting product sales plays an important role in formulating product launch plans for merchants. At present, sales forecasts for e-commerce platforms are often based on traditional forecasting methods such as linear regression, support vector machines, and neural networks. For scenarios such as long-tail sales scenarios, sales of 0-200 account for more than 80%, and sales of more than 10,000 account for less than 3%. In this regard, the current sales forecasting methods have limited forecasting capabilities and the accuracy of forecasts needs to be improved. Recently, the Amazon platform has gradually opened up the sales of product variants corresponding to each category. Due to the cost and difficulty of collecting samples, only part of the public sales can be obtained.
[0003] Therefore, there is still a problem of incomplete sales data acquisition in the related technology. Summary of the invention
[0004] In this embodiment, a sales correction method, a sales correction device and a computer-readable storage medium are provided to solve the problem of incomplete sales data acquisition in the related art.
[0005] In a first aspect, a sales volume correction method is provided in this embodiment, including: Obtaining an initial sales volume completion result of the target product category based on the initial sales volume completion, and obtaining sales volume characteristic data of each product in the target product category; According to the trained interval classification model, the sales feature data is classified into sales intervals to obtain the target sales interval corresponding to each product in the target product category; The residual prediction of the sales feature data is performed according to the trained residual prediction model to obtain a residual prediction result; wherein the trained residual prediction model is obtained based on a preset training sample set and a target residual training; the target residual is the residual between the sales complement value and the real sales value corresponding to the training sample set; According to the target sales volume range, the residual prediction result and the initial sales volume completion result, the final sales volume completion result corresponding to each commodity under the target commodity category is determined.
[0006] In some of the embodiments, obtaining the initial sales volume completion result of the target product category based on the initial sales volume completion includes: Obtaining sales ranking information and public sales data corresponding to each standardized product unit under the target product category; Establishing a relationship pair between the sales ranking information and the public sales data, and completing the sales data of the standardized product units with missing sales under the target product category based on the relationship pair based on interpolation processing to obtain the completed sales data of the standardized product units with missing sales; The sales data of each inventory unit under each standardized product unit is divided into sub-bodies according to the completed sales data to obtain the initial sales completion result of all commodities under the target commodity category.
[0007] In some of the embodiments, obtaining sales characteristic data of each commodity in the target commodity category includes: Extracting multi-dimensional initial feature data of each product in the target product category; According to different feature screening models, the multi-dimensional initial feature data are sorted by feature importance to obtain feature importances calculated by the different feature screening models; According to a preset weight allocation rule, weighted summation is performed on the feature importances respectively calculated by the different feature screening models to obtain a feature weight score; The sales feature data is obtained by screening the initial feature data according to the feature weight score.
[0008] In some of the embodiments, the multi-dimensional initial feature data includes at least: sales ranking time-varying features, category features, price features, collection frequency features, keyword features, price rating features, and comment features.
[0009] In some embodiments, the training process of the trained residual prediction model includes: Taking the feature data in the preset training sample set as input and the target residual corresponding to the training sample set as prediction output, the initial gradient boosting model is trained until the preset training termination condition is reached, thereby obtaining the trained residual prediction model.
[0010] In some of the embodiments, when training the initial gradient boosting model, the method further includes: Performing a hyperparameter search on the tree depth of the decision tree in the gradient boosting model based on a grid search to determine the optimal tree depth of the gradient boosting model; When iterating on the verification set divided from the training sample set, the iteration round in which the training loss function value decreases during the iteration and the accuracy rate corresponding to the verification set is the highest is taken as the final training round.
[0011] In some of the embodiments, determining the final sales forecast result corresponding to each product under the target product category according to the target sales range, the residual forecast result and the initial sales completion result includes: When the initial sales completion result is consistent with the target sales interval, the final sales completion result is determined to be the sum of the exponential operation result of the residual prediction result and the initial sales completion result.
[0012] In some of the embodiments, determining the final sales forecast result corresponding to each product under the target product category according to the target sales range, the residual forecast result and the initial sales completion result includes: When the initial sales completion result is inconsistent with the target sales interval, the final sales forecast result is determined to be the sum of the exponential operation result of the residual forecast result and the lower limit value of the target sales interval.
[0013] In a second aspect, a sales volume correction device is provided in this embodiment, including: an acquisition module, a classification module, a residual prediction module and a correction module; wherein: The acquisition module is used to acquire the initial sales volume completion result of the target commodity category based on the initial sales volume completion, and acquire the sales volume characteristic data of each commodity in the target commodity category; The classification module is used to classify the sales feature data into sales intervals according to the trained interval classification model to obtain the target sales interval corresponding to each commodity in the target commodity category; The residual prediction module is used to perform residual prediction on the sales feature data according to the trained residual prediction model to obtain a residual prediction result; wherein the trained residual prediction model is obtained based on a preset training sample set and a target residual training; the target residual is the residual between the sales complement value and the real sales value respectively corresponding to the training sample set; The correction module is used to determine the final sales completion result corresponding to each product under the target product category according to the target sales range, the residual prediction result and the initial sales completion result.
[0014] According to a third aspect, a storage medium is provided in this embodiment, on which a computer program is stored, and when the program is executed by a processor, the sales volume correction method described in the first aspect is implemented.
[0015] Compared with the related art, the present embodiment provides a sales correction method, a sales correction device and a computer-readable storage medium. The sales correction method obtains the initial sales completion result of the target commodity category based on the initial sales completion, and obtains the sales feature data of each commodity in the target commodity category; classifies the sales feature data into sales intervals according to the trained interval classification model to obtain the target sales interval corresponding to each commodity in the target commodity category; predicts the sales feature data according to the trained residual prediction model to obtain the residual prediction result; wherein the trained residual prediction model is trained based on the preset training sample set and the target residual; the target residual is the residual between the sales prediction value and the actual sales value corresponding to the training sample set; according to the target sales interval, the residual prediction result and the initial sales completion result, the final sales completion result corresponding to each commodity under the target commodity category is determined. It realizes the classification of the overall sales trend through the classification model, determines the residual between the initial sales completion result and the actual sales through the residual prediction, and then realizes the correction of the initial sales completion result, improves the accuracy of the final sales completion result, and realizes the completion of the sales data of all commodities.
[0016] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 is a hardware structure block diagram of a terminal of the sales volume correction method of this embodiment; Figure 2 is a flow chart of the sales volume correction method of this embodiment; Figure 3 This is a schematic diagram of the corresponding relationship between sales volume and BSR data in this embodiment; Figure 4 is a flow chart of a sales volume correction method of some of the embodiments; Figure 5 It is a structural block diagram of the sales volume correction device of this embodiment. DETAILED DESCRIPTION
[0018] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments.
[0019] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meaning understood by people with general skills in the technical field to which this application belongs. The words "one", "a", "the", "these" and the like in this application do not indicate a quantitative limitation, and they may be singular or plural. The terms "include", "comprise", "have" and any variants thereof involved in this application are intended to cover non-exclusive inclusions; for example, a process, method and system, product or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" may mean: A exists alone, A and B exist at the same time, and B exists alone. Generally, the character " / " indicates that the objects associated with each other are in an "or" relationship. The terms "first", "second", "third", etc. in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.
[0020] The method embodiment provided in this embodiment can be executed in a terminal, a computer or a similar computing device. For example, running on a terminal, Figure 1 FIG. 1 is a hardware structure diagram of a terminal of the sales volume correction method of this embodiment. Figure 1 As shown, the terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 and a memory 104 for storing data, wherein the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA. The above terminal may also include a transmission device 106 and an input and output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is for illustration only and does not limit the structure of the above terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.
[0021] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the sales volume correction method in this embodiment. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the terminal via a network. Examples of the above network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0022] The transmission device 106 is used to receive or send data via a network. The above network includes a wireless network provided by the communication provider of the terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (Radio Frequency, referred to as RF) module, which is used to communicate with the Internet wirelessly.
[0023] In this embodiment, a sales volume correction method is provided. Figure 2 is a flow chart of the sales volume correction method of this embodiment. Figure 2 As shown, the process includes the following steps: Step S210, obtaining the initial sales volume completion result of the target product category based on the initial sales volume completion, and obtaining the sales volume characteristic data of each product in the target product category.
[0024] Specifically, the target product category can first be preliminarily supplemented with sales based on a preset sales completion algorithm to obtain the corresponding initial sales completion result. The preset sales completion algorithm can be to establish a mapping relationship between BSR and public sales, and to complete the missing sales based on interpolation processing to obtain the initial sales completion result.
[0025] The commodities in the target commodity category may specifically be standardized product units of the target commodity category and corresponding inventory units under the standardized product units. Among them, the Standard Product Unit (SPU) is a product or commodity with a unique identifier and a set of standardized attributes under the target commodity category of the e-commerce platform. The Stock Keeping Unit (SKU) is a code or identifier used to identify and manage each unique commodity or product variant on the e-commerce platform. It is a key tool for merchants to track, manage and control inventory when selling commodities on e-commerce platforms (such as Amazon platforms). Multiple inventory units may correspond to one standardized product unit. Therefore, a standardized product unit may be used as a parent, and the corresponding inventory unit may be used as a child of the parent. The correspondence between the parent and the child may be one-to-one or one-to-many. The sales feature data of each parent and child under the target commodity category may be extracted respectively. The above-mentioned sales feature data may be one or more features of statistical information representing the sales changes of commodities.
[0026] Step S220, classifying the sales feature data into sales intervals according to the trained interval classification model, and obtaining the target sales interval corresponding to each product in the target product category.
[0027] In this step, first, sales completion is considered as a classification problem, which is used to determine the actual sales interval of the product. The interval can be expressed as [interval_low, interval_high]. Interval_low represents the lower limit, and interval_high represents the upper limit. For example, the sales interval can be divided into 0 to 50, 50 to 100, 100 to 900, 1000 to 10000, etc., and 20 categories can be divided according to the sales interval, that is, one interval corresponds to one category. Therefore, the target sales interval corresponding to each product in the target product category can be classified by classification.
[0028] Therefore, based on the trained interval classification model, the sales feature data can be used as input data and input into the trained interval classification model to classify the sales interval, thereby obtaining the target sales interval corresponding to each commodity in the target commodity category. In particular, the interval classification model can be implemented based on the Extreme Gradient Boosting (xgboost) model. In addition, those skilled in the art can also select other models suitable for classification according to the needs of actual application scenarios. Based on this, this step can realize the classification of the sales interval of each commodity in the target commodity category. The specific training process of training the above-mentioned gradient boosting model to obtain the trained interval classification model can select an existing model training method for training according to actual conditions, and this embodiment does not make specific limitations on this.
[0029] Step S230, performing residual prediction on the sales feature data according to the trained residual prediction model to obtain a residual prediction result; wherein the trained residual prediction model is obtained based on a preset training sample set and a target residual training; the target residual is the residual between the sales completion value and the true sales value corresponding to the training sample set.
[0030] Figure 3 FIG. 1 is a schematic diagram of a data correspondence between sales volume and BSR in this embodiment. Figure 3 As shown, Figure 3 In the distribution diagrams on the left and right sides, the horizontal axis is BSR and the vertical axis is actual sales data. The points in the figure are the corresponding data distribution, and the curve is the fitting curve corresponding to the functional relationship between BSR and sales. For example, the expression of the fitting curve can be: f(x)=8728×e^(-8×x)+46. Where “^” represents the power operation.
[0031] This embodiment takes into account that the public sales value is a value within a certain range, such as above 50, above 100, above 200, etc. Therefore, it is difficult to directly complete the public sales using a machine learning regression model. This embodiment regards sales completion as a classification problem. In this regard, in the training process of the residual prediction model, based on the data in the training sample set where the error between the sales completion value and the true sales value does not exceed 20%, the corresponding residual value is used as the Y value for model training.
[0032] Specifically, the trained residual prediction model is used to process the input sales feature data and output the residual prediction result. The residual prediction result represents the residual between the completed sales result and the actual sales. In the model training stage, the initial xgboost model can be trained by using the residual between the sales prediction value obtained by the initial completion of the training sample set and the actual sales value of the training sample set itself to obtain the above-mentioned fully trained residual prediction model.
[0033] Specifically, during the training phase, the sales completion based on the sales interpolation processing can be used to complete the sales of the training sample set and obtain the sales completion value. The residual between the sales completion value and the true sales value is calculated, and the initial xgboost model is trained based on the residual and the training sample set to obtain a trained residual prediction model.
[0034] Step S240, determining the final sales completion result corresponding to each product under the target product category according to the target sales range, the residual prediction result and the initial sales completion result.
[0035] In step S220 and step S230, the classification of sales intervals and the prediction of sales residuals are respectively realized. Then, according to the results obtained in step S220 and step S230, combined with the initial sales completion results obtained in step S210, the final sales completion results of each product under the target product category are determined. It can be understood that the final sales completion results can include the sales completion results corresponding to each child body under the parent body.
[0036] Among them, the above steps S210 to S240 can be applied to both offline sales completion scenarios and real-time sales completion scenarios. For example, after completing the initial sales completion of the target product category every day, the sales completion results can be corrected based on the above steps S210 to S240. Alternatively, the real-time data collection system kappa can be combined to obtain the corresponding sales feature data in real time, and the real-time sales completion interface can be called to obtain the real-time initial sales completion results, and the real-time sales correction can be achieved based on the above steps S210 to S240.
[0037] In the related art, the sales volume of goods is often predicted based on inventory or sales ranking. However, for specific scenarios, such as long-tail sales scenarios, sales of 0-200 account for more than 80%, and sales of more than 10,000 account for less than 3%. In this regard, the current sales forecasting method has limited forecasting capabilities and the problem of prediction accuracy needs to be improved. In addition, due to the cost and difficulty of collecting samples of public sales data open to e-commerce platforms, only part of the public sales can often be obtained. To this end, this embodiment first performs initial data completion on the missing sales data based on the public sales data through the above steps S210 to S240, and then, based on the interval classification model, the sales interval classification corresponding to the missing sales data is realized, and based on the residual prediction model, the prediction of further sales residuals is realized, and the residual between the initial sales completion result and the actual sales is predicted, so that the initial sales completion result can be corrected, the accuracy of the final sales completion result is improved, and the completion of the full amount of commodity sales data is realized.
[0038] In the above steps S210 to S240, the initial sales volume completion result obtained based on the initial sales volume completion of the target commodity category is obtained, and the sales volume characteristic data of each commodity in the target commodity category is obtained; the sales volume characteristic data is classified into sales volume intervals according to the trained interval classification model to obtain the target sales volume intervals corresponding to each commodity in the target commodity category; the sales volume characteristic data is residually predicted according to the trained residual prediction model to obtain the residual prediction result; wherein, the trained residual prediction model is obtained based on the preset training sample set and the target residual training; the target residual is the residual between the sales volume prediction value and the actual sales volume value respectively corresponding to the training sample set; according to the target sales volume interval, the residual prediction result and the initial sales volume completion result, the final sales volume completion result corresponding to each commodity under the target commodity category is determined. It realizes the classification of the overall sales volume trend through the classification model, determines the residual between the initial sales volume completion result and the actual sales volume through the residual prediction, and then realizes the correction of the initial sales volume completion result, improves the accuracy of the final sales volume completion result, and realizes the completion of the sales volume data of all commodities.
[0039] In one embodiment, based on the above step S210, obtaining the initial sales volume completion result of the target product category based on the initial sales volume completion may specifically include: Obtain the sales ranking information and public sales data corresponding to each standardized product unit under the target commodity category; establish a relationship pair between the sales ranking information and the public sales data, and based on the relationship pair, complete the sales data of the standardized product units with missing sales under the target commodity category based on interpolation processing to obtain the completed sales data of the standardized product units with missing sales; divide the sales data of each inventory unit under each standardized product unit into sub-bodies based on the completed sales data to obtain the initial sales completion result of all commodities under the target commodity category.
[0040] Among them, the public sales volume corresponding to each inventory unit (i.e., child body) sampled and extracted from the e-commerce platform. According to the corresponding relationship between the parent body and the child body, the public sales volume data of all child bodies corresponding to a parent body can be aggregated to obtain the public sales volume data corresponding to the parent body.
[0041] In this embodiment, considering that the sales ranking information can reflect the sales performance of the product in its category, there is a strong correlation between the sales ranking information and the sales volume. Therefore, after obtaining the public sales volume of the parent, the mapping relationship between the public sales volume of the parent and the sales ranking information of the parent can be matched to establish a relationship pair between the above sales ranking information and the public sales volume data of the parent. Afterwards, based on the arrangement relationship between different sales rankings in the relationship pair, for the parent that already has sales data but is missing sales data, interpolation processing can be used to complete the sales data corresponding to the parent with missing sales based on other parents with public sales, so as to complete the missing sales volume of the parent. For example, the missing sales volume can be completed based on the nearest neighbor interpolation algorithm. The specific completion process can be:
[0042] in is the complete sales volume when BSR is k, y k+i It is the public sales volume of the parent corresponding to the i-th BSR with sales volume that is closest to the k-th BSR. In this way, the corresponding relationship between BSR and public sales volume can be obtained under each category to complete the missing sales volume of the parent.
[0043] After completing the completion of the parent body sales, the child body sales data is divided according to the mapping relationship between the parent body and the child body. Among them, for the child body whose public sales have been obtained, the original corresponding public sales can be used directly; for the child body whose public sales are missing, the child body division of the sales data can be realized based on the corresponding parent body’s completed public sales, combined with other sales reference information, such as historical public sales, number of comments, etc. Based on this, for a parent body, the sales data of all child bodies under the parent body can be obtained. Based on this, this step can obtain the full sales data of all child bodies under the target product category.
[0044] For example, the obtained child sales are subtracted from the public sales corresponding to the parent, and the public sales of the child with missing sales are divided according to the number of comments, historical public sales data and other reference indicators. The more comments a child has, the more public sales it can get. For example, the sales data obtained by dividing the child with missing sales can be:
[0045] in, l k Indicates the sales data obtained by completing the sub-body; Indicates the completed sales data of the parent body; L known Indicates the sum of the public sales data of the sub-body that has been obtained. c i Indicates i The number of comments for the sub-bodies with missing sales volume is calculated; if the number of comments for all sub-bodies with missing sales volume is 0, the public sales volume data to be divided will be evenly divided according to the number of sub-bodies with missing sales volume.
[0046] In this way, this embodiment achieves the initial data completion of the sales data by interpolation completion, and obtains the above-mentioned initial sales completion result. It can be understood that in the process of training the residual prediction model of the above-mentioned step S230, the sales completion value of the corresponding training sample set can also be obtained based on the interpolation completion method of this embodiment, which will not be described in detail.
[0047] Additionally, in one embodiment, based on the above step S210, obtaining sales characteristic data of each commodity in the target commodity category may include: Extract multi-dimensional initial feature data of each product in the target product category; sort the multi-dimensional initial feature data by feature importance according to different feature screening models to obtain feature importances calculated by different feature screening models; perform weighted summation of feature importances calculated by different feature screening models according to preset weight allocation rules to obtain feature weight scores; and filter sales feature data from the initial feature data according to the feature weight scores.
[0048] The sales feature data of different dimensions can be extracted from the parent and child levels respectively. In this embodiment, in order to improve the data processing efficiency, it is necessary to screen out effective features from the initial feature data of many different dimensions as the sales feature data for the final sales completion, and eliminate redundant data. Therefore, in this embodiment, the three feature screening models of random forest, logistic regression, and gradient boosting decision tree (GBDT) are used to calculate the feature importance of each dimensional feature, and screen out the features with the top preset importance ranking (for example, the top 50). Among them, during the screening process, the above-mentioned feature screening model sorts the features of each dimension according to the feature importance, and assigns a serial number as the corresponding feature importance.
[0049] For the feature data of the preset number of digits screened out by the feature screening model, the feature importance is weighted to obtain the feature weight score, and the feature data with the top several digits (for example, 20 digits) is obtained as the sales feature data. The specific process of calculating the feature weight score can be: fea_imp=0.5×gbdt+0.3×rf+0.2×lr Among them, fea_imp is the feature weight score; gbdt represents the feature importance calculated by the GBDT model, and 0.5 is the weight corresponding to the GBDT model; rf is the feature importance calculated by the random forest model, and 0.3 is the weight corresponding to the random forest model; lr is the feature importance calculated by the logistic regression model, and 0.2 is the weight corresponding to the logistic regression model.
[0050] In one embodiment, the multi-dimensional initial feature data includes at least: sales ranking time-varying features, category features, price features, collection frequency features, keyword features, price rating features, and comment features.
[0051] The sales ranking time-varying feature may specifically be a BSR time-varying feature, which may specifically be the following statistics of the BSR of the child (asin) and the parent (pasin) in the last 7 days, 15 days, 30 days, 60 days and 90 days: average, maximum, minimum and quantile (for example, quantile [0.25, 0.5, 0.75]).
[0052] The above-mentioned category features may include the unique identifier (ID) of the category to which the child belongs, whether there has been a category change in the past 7 days, 15 days, 30 days, 60 days, or 90 days; and whether the category is shared. Among them, the parent and the corresponding child share the BSR ranking, that is, the BSR of the parent under the category is exactly the same as the BSR of the child, otherwise the category is not shared.
[0053] The price characteristics may specifically be the price average, maximum price, and minimum price of the past 7 days, 15 days, 30 days, 60 days, or 90 days.
[0054] The collection frequency feature may be the number of collections in the past 7 days, 15 days, 30 days, 60 days, or 90 days.
[0055] Keyword features are used to filter out searches and words with large exposure (such as words with a daily search volume greater than 10 million) based on word frequency. Specifically, keyword features can be the keywords with the highest exposure rate under the sub-body, and the corresponding exposure, click volume, and search volume can be extracted; as well as the number of keywords with a corresponding exposure volume greater than 10,000; whether the top-ranked keyword has changed in the past 30 days; and the number of changes to the keyword with the highest exposure rate in the past 90 days.
[0056] The price rating feature may specifically be the average number of new price ratings in the past 7 days, 15 days, 30 days, 60 days, or 90 days, and the percentage of evaluation levels from the lowest to the highest.
[0057] The comment feature may specifically be an average number of new price comments in the past 7 days, 15 days, 30 days, 60 days, or 90 days.
[0058] Thus, the multi-dimensional initial feature data may include 7 categories and nearly 200 dimensions of features. This embodiment determines effective sales feature data from the comprehensive functions of the initial feature data of different dimensions, which can improve the accuracy of sales volume completion.
[0059] Additionally, in one embodiment, the training process of the trained residual prediction model may include: Taking the feature data in the preset training sample set as input and the target residual corresponding to the training sample set as the prediction output, the initial gradient boosting model is trained until the preset training termination condition is reached to obtain the trained residual prediction model.
[0060] Specifically, the residual can be calculated as follows: err = log(y-pred1) Among them, err represents the residual of the actual prediction, y represents the true sales value of the training sample set, and pred1 represents the completed sales value of the training sample set.
[0061] When training the initial gradient boosting model, the training sample set can be divided into a training set, a validation set, and a test set, and the gradient boosting model is trained until a preset training termination condition is reached. The training termination condition may be reaching a preset number of training iterations, or the accuracy of the model output reaches a preset accuracy requirement.
[0062] In particular, in one embodiment, when training the initial gradient boosting model, a hyperparameter search is performed on the tree depth of the decision tree in the gradient boosting model based on a grid search to determine the optimal tree depth of the gradient boosting model; when iterating on the validation set divided from the training sample set, the iteration round in which the training loss function value decreases during the iteration process and the accuracy rate corresponding to the validation set is the highest is taken as the final training round.
[0063] The gradient boosting model xgboost includes a decision tree, and a hyperparameter search can be performed on the tree depth according to the grid search to determine the optimal tree depth. For example, a tree depth of 6 is determined as the optimal tree depth. Iterate on the validation set, and select the iteration round with the lowest training loss function value and the highest validation set accuracy as the final training round to terminate the iterative training. In this way, the xgboost model obtained by the final training can be used as the residual prediction model after the above training.
[0064] In addition, in one embodiment, based on the above step S240, according to the target sales range, the residual prediction result and the initial sales completion result, the final sales prediction result corresponding to each product under the target product category is determined, which may specifically include: When the initial sales completion result is consistent with the target sales range, the final sales completion result is determined to be the sum of the exponential operation result of the residual prediction result and the initial sales completion result.
[0065] When the initial sales completion result is consistent with the target sales range obtained by classification in step S220, the initial sales completion result is corrected according to the residual prediction result obtained in step S230, and the final sales completion result is determined as the sum of the exponential calculation result of the residual prediction result and the initial sales completion result, that is: pred_new=exp(err_pred)+pred2 Where pred_new represents the final sales completion result; exp is the exponential operation, exp(a)=e a , where e is a constant, approximately 2.71828. err_pred represents the residual prediction result; pred2 is the initial sales completion result.
[0066] In this embodiment, the initial sales volume completion result can be accurately corrected, thereby improving the accuracy of the final sales volume forecast result.
[0067] In addition, in one embodiment, based on the above step S240, the final sales completion result corresponding to each product under the target product category is determined according to the target sales range, the residual prediction result, and the initial sales completion result, which may specifically include: When the initial sales completion result is inconsistent with the target sales range, the final sales completion result is determined to be the sum of the exponential operation result of the residual prediction result and the lower limit value of the target sales range.
[0068] When the initial sales completion result is inconsistent with the target sales range, the final sales completion result pred_new can be: pred_new=exp(err_pred)+interval_low Among them, pred_new represents the final sales completion result; exp is the exponential operation, exp(a)=ea, where e is a constant, approximately 2.71828. err_pred represents the residual prediction result; interval_low is the lower limit of the target sales interval.
[0069] In this embodiment, when the target sales volume range of the classification is inconsistent with the initial sales volume completion result, it is possible to choose to accurately correct the initial sales volume completion result, thereby improving the accuracy of the final sales volume completion result.
[0070] Figure 4 is a flow chart of a sales volume correction method in some embodiments, such as Figure 4 As shown, the sales volume correction method includes the following steps: Step S401, obtaining sales ranking information and public sales data corresponding to each standardized product unit under the target commodity category.
[0071] Step S402, establish a relationship pair between the sales ranking information and the public sales data, and based on the relationship pair, supplement the sales data of the standardized product units with missing sales under the target product category based on interpolation processing to obtain the supplemented sales data of the standardized product units with missing sales.
[0072] Step S403, dividing the sales data of each inventory unit under each standardized product unit into sub-bodies according to the completed sales data, and obtaining the initial sales completion result of all commodities under the target commodity category.
[0073] Step S404: extracting multi-dimensional initial feature data of each product in the target product category.
[0074] Step S405 , according to different feature screening models, the feature importance of the multi-dimensional initial feature data is sorted respectively, and the feature importance calculated by different feature screening models is obtained.
[0075] Step S406, according to a preset weight allocation rule, weighted sum is performed on the feature importances respectively calculated by different feature screening models to obtain a feature weight score.
[0076] Step S407: Filter the initial feature data to obtain sales feature data according to the feature weight score.
[0077] Step S408, classifying the sales feature data into sales intervals according to the trained interval classification model, and obtaining the target sales interval corresponding to each product in the target product category.
[0078] Step S409, perform residual prediction on the sales feature data according to the trained residual prediction model to obtain a residual prediction result; wherein the trained residual prediction model is obtained based on a preset training sample set and a target residual training; the target residual is the residual between the sales complement value and the true sales value corresponding to the training sample set. Taking the feature data in the preset training sample set as input and the target residual corresponding to the training sample set as the prediction output, the initial gradient boosting model is trained until the preset training termination condition is reached to obtain a trained residual prediction model.
[0079] Step S410, when the initial sales completion result is consistent with the target sales range, the final sales completion result is determined to be the sum of the exponential operation result of the residual prediction result and the initial sales completion result.
[0080] Step S411, when the initial sales completion result is inconsistent with the target sales range, the final sales completion result is determined to be the sum of the exponential operation result of the residual prediction result and the lower limit value of the target sales range.
[0081] The above steps S401 to S411 realize the classification of the overall sales trend through the classification model, determine the residual between the initial sales completion result and the actual sales through residual prediction, and then realize the correction of the initial sales completion result, improve the accuracy of the final sales completion result, and realize the completion of the sales data of all commodities.
[0082] In this embodiment, a sales volume correction device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. The terms "module", "unit", "subunit", etc. used below may be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and conceivable.
[0083] Figure 5 is a structural block diagram of the sales volume correction device 50 of this embodiment. Figure 5 As shown, the sales volume correction device 50 includes: an acquisition module 52, a classification module 54, a residual prediction module 56 and a correction module 58; wherein: An acquisition module 52 is used to obtain the initial sales volume completion result of the target commodity category based on the initial sales volume completion, and obtain the sales volume characteristic data of each commodity in the target commodity category; a classification module 54 is used to perform sales volume interval classification on the sales volume characteristic data according to the trained interval classification model, and obtain the target sales volume interval corresponding to each commodity in the target commodity category; a residual prediction module 56 is used to perform residual prediction on the sales volume characteristic data according to the trained residual prediction model, and obtain the residual prediction result; wherein the trained residual prediction model is trained based on a preset training sample set and a target residual; the target residual is the residual between the sales volume completion value and the true sales volume value respectively corresponding to the training sample set; a correction module 58 is used to determine the final sales volume completion result corresponding to each commodity under the target commodity category according to the target sales volume interval, the residual prediction result and the initial sales volume completion result.
[0084] It should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.
[0085] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation modes, and will not be repeated in this embodiment.
[0086] In addition, in combination with the sales volume correction method provided in the above embodiments, a storage medium may be provided in this embodiment to implement the sales volume correction method. The storage medium stores a computer program; when the computer program is executed by a processor, any sales volume correction method in the above embodiments is implemented.
[0087] It should be understood that the specific embodiments described herein are only used to explain the application, rather than to limit it. Based on the embodiments provided in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of this application.
[0088] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.
[0089] Obviously, the drawings are only some examples or embodiments of the present application. For ordinary technicians in the field, the present application can also be applied to other similar situations based on these drawings without creative work. In addition, it is understandable that although the work done in this development process may be complicated and lengthy, for ordinary technicians in the field, certain changes in design, manufacturing or production based on the technical content disclosed in this application are only conventional technical means and should not be regarded as insufficient content disclosed in this application.
[0090] The term "embodiment" in this application refers to a specific feature, structure or characteristic described in conjunction with the embodiment that can be included in at least one embodiment of the present application. The appearance of this phrase in various locations in the specification does not necessarily mean the same embodiment, nor does it mean that it is mutually exclusive with other embodiments and is independent or optional. It is clearly or implicitly understood by those of ordinary skill in the art that the embodiments described in this application can be combined with other embodiments without conflict.
[0091] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of patent protection. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the scope of protection of the present application. Therefore, the scope of protection of the present application shall be subject to the attached claims.
Claims
1. A sales volume correction method, characterized in that: include: Obtaining an initial sales volume completion result of the target product category based on the initial sales volume completion, and obtaining sales volume characteristic data of each product in the target product category; According to the trained interval classification model, the sales feature data is classified into sales intervals to obtain the target sales interval corresponding to each product in the target product category; The residual prediction of the sales feature data is performed according to the trained residual prediction model to obtain a residual prediction result; wherein the trained residual prediction model is obtained based on a preset training sample set and a target residual training; the target residual is the residual between the sales complement value and the real sales value corresponding to the training sample set; According to the target sales volume range, the residual prediction result and the initial sales volume completion result, the final sales volume completion result corresponding to each commodity under the target commodity category is determined.
2. The sales volume correction method according to claim 1, characterized in that: Get the initial sales completion results of the target product category based on the initial sales completion, including: Obtaining sales ranking information and public sales data corresponding to each standardized product unit under the target product category; Establishing a relationship pair between the sales ranking information and the public sales data, and completing the sales data of the standardized product units with missing sales under the target product category based on the relationship pair based on interpolation processing to obtain the completed sales data of the standardized product units with missing sales; The sales data of each inventory unit under each standardized product unit is divided into sub-bodies according to the completed sales data to obtain the initial sales completion result of all commodities under the target commodity category.
3. The sales volume correction method according to claim 1, characterized in that: Obtain sales characteristic data of each product in the target product category, including: Extracting multi-dimensional initial feature data of each product in the target product category; According to different feature screening models, the multi-dimensional initial feature data are sorted by feature importance to obtain feature importances calculated by the different feature screening models; According to a preset weight allocation rule, weighted summation is performed on the feature importances respectively calculated by the different feature screening models to obtain a feature weight score; The sales feature data is obtained by screening the initial feature data according to the feature weight score.
4. The sales volume correction method according to claim 3, characterized in that: The multi-dimensional initial feature data includes at least: sales ranking time-varying features, category features, price features, collection frequency features, keyword features, price rating features, and comment features.
5. The sales volume correction method according to claim 1, characterized in that: The training process of the trained residual prediction model includes: Taking the feature data in the preset training sample set as input and the target residual corresponding to the training sample set as prediction output, the initial gradient boosting model is trained until the preset training termination condition is reached, thereby obtaining the trained residual prediction model.
6. The sales volume correction method according to claim 5, characterized in that: When training the initial gradient boosting model, the method further includes: Performing a hyperparameter search on the tree depth of the decision tree in the gradient boosting model based on a grid search to determine the optimal tree depth of the gradient boosting model; When iterating on the verification set divided from the training sample set, the iteration round in which the training loss function value decreases during the iteration and the accuracy rate corresponding to the verification set is the highest is taken as the final training round.
7. The sales volume correction method according to any one of claims 1 to 6, characterized in that: According to the target sales range, the residual prediction result and the initial sales completion result, a final sales completion result corresponding to each product under the target product category is determined, including: When the initial sales completion result is consistent with the target sales interval, the final sales completion result is determined to be the sum of the exponential operation result of the residual prediction result and the initial sales completion result.
8. The sales volume correction method according to any one of claims 1 to 6, characterized in that: According to the target sales range, the residual prediction result and the initial sales completion result, a final sales completion result corresponding to each product under the target product category is determined, including: When the initial sales completion result is inconsistent with the target sales interval, the final sales completion result is determined to be the sum of the exponential operation result of the residual prediction result and the lower limit value of the target sales interval.
9. A sales volume correction device, characterized in that: include: Acquisition module, classification module, residual prediction module and correction module; wherein: The acquisition module is used to acquire the initial sales volume completion result of the target commodity category based on the initial sales volume completion, and acquire the sales volume characteristic data of each commodity in the target commodity category; The classification module is used to classify the sales feature data into sales intervals according to the trained interval classification model to obtain the target sales interval corresponding to each commodity in the target commodity category; The residual prediction module is used to perform residual prediction on the sales feature data according to the trained residual prediction model to obtain a residual prediction result; wherein the trained residual prediction model is obtained based on a preset training sample set and a target residual training; the target residual is the residual between the sales complement value and the real sales value respectively corresponding to the training sample set; The correction module is used to determine the final sales completion result corresponding to each product under the target product category according to the target sales range, the residual prediction result and the initial sales completion result.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the sales volume correction method according to any one of claims 1 to 8 are implemented.
Citation Information
Patent Citations
Commodity sales prediction method and device, electronic equipment and readable storage medium
CN111652655A
Sales prediction method and device, computer equipment and storage medium
CN117314492A
Goods sorting mode determination method and device, equipment and storage medium
CN117726361A
