CN117726361A - Method, device, equipment and storage medium for determining goods distribution mode - Google Patents

Abstract

The disclosure relates to a method, a device, equipment and a storage medium for determining a goods distribution mode, and relates to the technical field of machine learning. The method comprises the following steps: acquiring sales process information of a shop selling target goods in a specified period; determining sales volume prediction labels of the target goods based on the sales process information and a pre-trained combined prediction model, wherein the combined prediction model comprises a plurality of classification models; and determining the goods distribution mode of the target goods in the store based on the sales volume prediction label. Therefore, the accuracy of predicting the sales volume of goods is improved, the problem that the model prediction deviation is increased due to data sparseness is solved through a combined prediction mode of a plurality of two-classification models, reasonable goods distribution of shops is facilitated, and out-of-sales and zero-sales are avoided.

Description

Method, device, equipment and storage medium for determining goods distribution mode

Technical Field

The disclosure relates to the technical field of machine learning, and in particular relates to a method, a device, equipment and a storage medium for determining a goods distribution mode.

Background

Store-sku (stock keeping unit, stock unit) dimension cell phone commodity sales prediction is of great importance in solving store commodity distribution problems and improving turnover. By accurately predicting sales of each commodity sku for a store, the store can be aided in accurately knowing the demand of each commodity sku, ensuring that each store can provide sufficient inventory to meet the purchasing needs of customers. In addition, inventory days of stores can be reasonably controlled according to different sales speeds and inventory levels through sales volume prediction, so that stock backlog or backorder can be avoided, and fund turnover efficiency is improved.

In the related art, regression prediction is usually adopted, but sales data are sparse in many times, prediction deviation is easily increased by the regression method, so that a goods distribution mode is unreasonable, and even a store is in a zero sale or out of sale in a serious case.

Disclosure of Invention

The present disclosure aims to solve, at least to some extent, one of the technical problems in the related art.

An embodiment of a first aspect of the present disclosure provides a method for determining a goods distribution mode, including:

acquiring sales process information of a shop selling target goods in a specified period;

determining sales volume prediction labels of the target goods based on the sales process information and a pre-trained combined prediction model, wherein the combined prediction model comprises a plurality of classification models;

and determining the goods distribution mode of the target goods in the store based on the sales volume prediction label.

An embodiment of a second aspect of the present disclosure provides a determining apparatus for a goods distribution mode, including:

the acquisition module is used for acquiring sales process information of a shop selling target goods in a specified period;

the first determining module is used for determining sales volume prediction labels of the target goods based on the sales process information and a pre-trained combined prediction model, wherein the combined prediction model comprises a plurality of classification models;

And the second determining module is used for determining the goods distribution mode of the target goods in the store based on the sales quantity prediction label.

An embodiment of a third aspect of the present disclosure provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the method for determining the goods distribution mode as set forth in the first aspect when the processor executes the program.

An embodiment of a fourth aspect of the present disclosure proposes a computer-readable storage medium, on which a computer program is stored, which program, when being executed by a processor, implements a method as described in the first aspect.

An embodiment of a fifth aspect of the present disclosure proposes a computer program product comprising a computer program which, when executed by a processor, implements the steps of the method according to the first aspect.

In the embodiment of the disclosure, sales process information of a target commodity sold by a shop in a designated period is firstly obtained, then a sales volume prediction label of the target commodity is determined based on the sales process information and a pre-trained combined prediction model, wherein the combined prediction model comprises a plurality of classification models, and finally a commodity distribution mode of the target commodity in the shop is determined based on the sales volume prediction label. Therefore, the accuracy of predicting the sales volume of goods is improved, the problem that the model prediction deviation is increased due to data sparseness is solved through a combined prediction mode of a plurality of two-classification models, reasonable goods distribution of shops is facilitated, and out-of-sales and zero-sales are avoided.

Additional aspects and advantages of the disclosure will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the disclosure.

Drawings

Fig. 1 is a flow chart of a method of determining a manner of dispensing items according to a first embodiment of the present disclosure;

FIG. 2 is a schematic diagram of a model training and prediction architecture;

fig. 3 is a flow chart of a method of determining a manner of dispensing items according to a second embodiment of the present disclosure;

FIG. 4 is a flow chart of super-parameter optimization in the model training and prediction process;

fig. 5 is a flow chart of a method of determining a manner of dispensing items according to a third embodiment of the present disclosure;

fig. 6 is a schematic structural view of a determining apparatus of an article dispensing mode according to an embodiment of the present disclosure;

fig. 7 illustrates a block diagram of an exemplary electronic device suitable for use in implementing embodiments of the present disclosure.

Detailed Description

Embodiments of the present disclosure are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and intended for the purpose of explaining the present disclosure and are not to be construed as limiting the present disclosure.

In the process of stock, replenishment and distribution of store commodity in new retail online, future sales conditions of each store-sku need to be accurately estimated, and store commodity is distributed according to the prediction result, so that store sales are met as much as possible, zero sales and sales rate are controlled, turnover and sales of stores are improved, and accurate sales prediction is important for the overall distribution condition.

In the scene of store-sku mobile phone sales prediction, as the data is very sparse (0 value is higher than 50%), label (tag) is concentrated, 98% label is concentrated in the range of 0-4, and the regression method predicts that the following defects exist: 1. the solution space is too large, and is usually valued in real space, which is far larger than the actual value (finite integer), so that the model learning burden is 2, the model learning is easily affected by abnormal values, and the prediction range of the model is usually affected although the ratio of the solution space is very low for a few large values label, so that the model prediction deviation is increased. The scheme can be used for solving the regression prediction problem in sparse and label concentration of the data, and provides a novel method for solving the ordinal classification problem based on integrated learning and global search optimal threshold, and the method shows the effect remarkably superior to other time sequence prediction schemes on the mobile phone sales prediction of store-sku granularity.

The following describes a method, apparatus, device and storage medium for determining a mode of distributing goods according to an embodiment of the present disclosure with reference to the accompanying drawings.

Fig. 1 is a flowchart of a method for determining a goods distribution mode according to an embodiment of the present disclosure. As shown in fig. 1, the method for determining the way of distributing goods may include the steps of:

it should be noted that the number of the substrates,

step 101, obtaining sales process information of a shop selling a target goods in a specified period.

The goods may be any goods sold by shops, such as mobile phones, headphones, tablets, home appliances, mobile phone cases, automobiles, charging lines, etc., and the goods sold by different shops are different and are not limited herein.

The target item may be any type of item to be predicted. For example, the target goods may be any specific type of mobile phone, which is not limited herein.

The store may be a store selling the target goods, for example, may be an online store, an electronic commerce platform, or may be a store, a retail store, a monopoly store, a store in store, etc., and the types of stores may be numerous, which are not limited herein.

The specified period may be a historical period of a certain period, such as one week, two weeks, one month, three months, half year, and one year, which is not limited herein.

The sales process information may be information related to the store when selling the target item, such as order information of the target item, information of the store, time information of sales, sales amount, and the like, which are not limited herein. Optionally, the sales process information includes at least one or more of: store information, order information, competitive goods information, time information, weather information, and promotional program information.

It will be appreciated that in the embodiments of the present disclosure, store information, order information, competitive goods information, time information, weather information, promotional activity information, or other sales process information that may affect sales of a target good may be used as information to be acquired, so that these sales process information may be used later to predict sales of the target good, providing powerful, efficient, and reliable data support.

The store information may include geographical location of the store, business hours, area of the store, decoration style, traffic of people in the region where the store is located, customer evaluation information of the store, number of store personnel of the store, and adjacent store information and bid store information of the location where the store is located, and the like, and is not limited herein.

The weather information may include weather conditions during sales, such as sunny days, rainy days, etc., for analyzing weather effects on sales and possibly adjusting promotional policies or inventory management, among others.

The competitive goods information may include, but is not limited to, a product type, price, sales time, sales strategy, sales promotion, sales volume, etc. of a competitor for analyzing market competition.

The order information may include sales of target goods, order quantity, payment method, distribution method, customer information of purchased goods, purchase behavior information of customers, order flow information, etc., which are not limited herein.

The time information may include, but is not limited to, sales time, transaction period, seasonal sales scenario, and the like.

Wherein, the promotion activity information: including sales promotion, discount information, gift information for the store, and sales promotion and discount information for the target item, etc., are not limited herein.

It should be noted that the sales process information may also include other types of information related to the sales process of the target goods, which is not described in detail herein.

Step 102, determining sales volume prediction labels of target goods based on sales process information and a pre-trained combined prediction model, wherein the combined prediction model comprises a plurality of classification models.

The sales predicting label can be a label which is obtained by evaluating sales of the target goods in a certain future period based on the collected sales process information of the historical period. The sales prediction tag can represent future sales characteristics of the store for the target item.

For example, sales prediction tags can be categorized into V0, V1, V2, V3. Where V0 represents the sales of store Y as 0 during the future time T period. V1 represents that sales K of store Y over a future time T period satisfy the relationship: 0 < K.ltoreq.K1, V2 represents that sales K of store Y over a future time T period satisfy the relationship: k1 < K.ltoreq.K2, V3 represents that sales K of store Y over a future time period T satisfy the relationship: k2 < K.ltoreq.K3, V4 represents that sales K of store Y over a future time period T satisfy the relationship: k3 is less than or equal to K4, and is not limited herein.

Wherein the classification model is a machine learning model for classifying the input samples into two different classes. The output of the two classification models may be a predictive probability. For example, it may be a probability of whether the sales K can be greater than K1 in the future time T period.

In the embodiment of the disclosure, the joint prediction model may be a multi-label classification model, and the joint prediction model may combine output results of a plurality of classification models to generate a comprehensive final output result, that is, a sales prediction label.

It should be noted that, the joint prediction model uses the mutual constraint relationship between the multiple two classifiers to jointly predict. Each bi-classifier only predicts whether the sales are greater than a certain value, e.g., whether the sales are greater than 0.

Fig. 2 is a schematic diagram of a model training and prediction architecture, as shown in fig. 2, where the target goods are mobile phones, and the collected sales process information (such as mobile phone data, store data, order data, bid data.) may be normalized and preprocessed, including missing value processing, outlier processing, data deduplication, data format conversion, etc., and useful feature vectors such as date features, commodity features, store promotion features, new product promotion features, external features, etc. are extracted through feature engineering as input for the model. And then, the characteristic information can be input into the joint prediction model, corresponding output results are respectively output through each of the two classification models in the joint prediction model, and then the output results of the two classification models can be fused to obtain a final sales prediction label. The joint prediction model comprises two classifiers 0, 1, 2, 3, 4 and 5. There are 5 sales tags, respectively, table 0, table 1, table 2, table 3, table 4, table 5, representing 6 cases where sales are 0, 1, 2, 3, 4, and sales are greater than 4. The two classifiers 0, 1, 2, 3, 4 and 5 are used for judging whether the sales are greater than 0, greater than 1, greater than 2, greater than 3, greater than 4 and which sales interval they belong to respectively.

Optionally, the prediction results of the multiple models can be fused in a weighted average mode, a voting mode or the like, so that the prediction accuracy is improved. In addition, the prediction result can be evaluated and optimized, and model parameters and feature selection can be adjusted according to indexes such as prediction errors, so that the accuracy and generalization capability of the model are improved.

And step 103, determining the distribution mode of the target goods in the store based on the sales volume prediction label.

The goods distribution mode can be a goods distribution mode of target goods in shops.

Alternatively, a first item quantity associated with a sales prediction tag may be first determined, and a second item quantity currently held by a store, and then a distribution quantity for the target item in the store may be determined based on the first item quantity and the second item quantity.

Wherein the first quantity of items may be a quantity of items corresponding to the sales prediction tag.

For example, the sales prediction tag may directly associate an item quantity, or may also associate an item quantity interval. For example, if the item number interval associated with the sales prediction tag is (a, b ], any integer in the value interval (a, b) may be used as the first item number, and the present invention is not limited thereto.

Wherein the second item quantity may be a store current inventory quantity.

Optionally, the first number of goods is denoted as X, the second number of goods is denoted as Y, if Y is greater than or equal to X, the stores are not required to be sorted, if Y is less than X, the stores are required to be sorted, and the number of sorted goods can be at least "X-Y".

In the embodiment of the disclosure, sales process information of a target commodity sold by a shop in a designated period is firstly obtained, then a sales volume prediction label of the target commodity is determined based on the sales process information and a pre-trained combined prediction model, wherein the combined prediction model comprises a plurality of classification models, and finally a commodity distribution mode of the target commodity in the shop is determined based on the sales volume prediction label. Therefore, the accuracy of predicting the sales volume of goods is improved, the problem that the model prediction deviation is increased due to data sparseness is solved through a combined prediction mode of a plurality of two-classification models, reasonable goods distribution of shops is facilitated, and out-of-sales and zero-sales are avoided.

Fig. 3 is a flowchart of a method for determining a goods distribution mode according to a second embodiment of the present disclosure. As shown in fig. 3, the method for determining the way of distributing goods may include the steps of:

Step 201, obtaining sales process information of a shop selling a target commodity in a specified period.

It should be noted that, the specific implementation manner of step 201 may refer to the above embodiment, and will not be described herein.

Step 202, obtaining a target value interval associated with each classification model.

The target value interval may be a parameter search space for determining a super parameter of the classification model. That is, the probability threshold of the prediction probability corresponding to each classification model may be determined in a target value interval, and the target value interval may be a value range of the super parameters of the classification models.

In the embodiment of the disclosure, a probability threshold value of the prediction probability of the classification model is used as a super parameter, and a target value interval corresponding to a value range of the probability threshold value is determined.

In the embodiment of the disclosure, the numerical space of the interval [0,1] may be used as the target value interval associated with each classification model, or the numerical space of the interval [0.25,0.75] may be used as the target value interval associated with each classification model.

Alternatively, a mapping relationship between each of the two classification models and the corresponding target value interval may be established in advance, for example, the target value space of the two classification model a is [0.25,0.5], and the target value space of the two classification model B is [0.3,0.8], which is not limited herein.

It should be noted that, in prediction, the prediction probability generally fluctuates within a range of values, and the range of values of the fluctuation corresponding to the prediction probability output by different classification models may be different. Therefore, the mapping relation between each classification model and the corresponding target value interval can be established in advance, so that the numerical value in the target value interval is traversed later, the calculated amount can be reduced, the calculation force is saved, the time is saved, the calculation speed is improved, and the calculation complexity is reduced.

And 203, dividing the target value interval to obtain a plurality of subintervals corresponding to the classification model.

Specifically, the target value space may be divided into a plurality of subintervals. For example, the target value interval [0,1] may be divided into 456 sub-intervals, or 100 sub-intervals. In the case of dividing the target value interval, the target value interval may be uniformly divided or unevenly divided, and the present invention is not limited thereto. Meanwhile, in the embodiment of the disclosure, the number of subintervals obtained by dividing the target value interval is not limited, and can be adjusted and set according to actual experience.

And 204, determining a plurality of super-parameter combinations according to the values of the two classification models in the corresponding subintervals.

Wherein, the hyper-parameter combination contains the value of each classification model in any subinterval.

For example, if there are 3 classification models, model1, model2, and model3, respectively, and the target value intervals corresponding to the three classification models are all [0,1]. The target value interval [0,1] of the model1 is divided into 4 sub-intervals of [0,0.25 ], [0.25,0.5 ], [0.5,0.75 ] and [ 0.75,1]. The target values of model2 and model3 were also divided into [0,0.25 ], [0.25,0.5 ], [0.5,0.75 ] and [ 0.75,1].

Then, a plurality of combinations of super parameters can be determined based on the subintervals [0,0.25 ], [0.25,0.5 ], [0.5, 0.75), (0.75,1 ] corresponding to model1, model2, and model3, respectively.

Each of the classification models may take any value in any subinterval, for example, model1 may take 0, 0.11, 0.2, 0.23 in subinterval [0,0.25 ]).

For example, model1 takes 4 arbitrary values D1, D2, D3, D4 in the 4 sub-intervals [0,0.25 ], [0.25,0.5 ], [0.5,0.75 ], and [ 0.75,1], where D1E [0,0.25 ], D2E [0.25,0.5 ], D3E [0.5,0.75 ], and D4E (0.75,1 ].

Similarly, the model2 takes 4 arbitrary values of R1, R2, R3, and R4 in the 4 sub-sections, and the model3 takes 4 arbitrary values of Y1, Y2, Y3, and Y4 in the 4 sub-sections. Then D1, D2, D3, D4 and R1, R2, R3, R4, and Y1, Y2, Y3, Y4 together form 96 sets of hyper-parametric combinations, i.e., 4x4=96, such as { D1, R1, Y1}, { D1, R1, Y2}, { D1, R1, Y3}, { D1, R1, Y4}, { D2, R1, Y1}, { D3, R1, Y1}, { D4, R1, Y1}, { D1, R2, Y1}, not to be exhaustive herein.

Step 205, obtaining model evaluation scores of the initial joint prediction model when each super-parameter combination is configured.

The initial joint prediction model may be a machine learning model that is trained to predict sales of a store for selling a target good over a period of time in the future (e.g., half a month). The initial combined prediction model is a model which does not reach the available degree, and the combined prediction model is a model which reaches the available degree.

The model evaluation score may be a measurement value obtained based on a measurement index measuring the performance of the model or the completion degree of the evaluation task. In machine learning and statistics, various metrics may be used to evaluate the accuracy, effectiveness, or other specific performance metrics of a model.

Alternatively, a verification data set may be first obtained, and an evaluation index for verifying the initial joint prediction model may be obtained, then the initial joint prediction model may be configured based on a plurality of super parameter combinations to obtain an initial joint prediction model of each configured super parameter combination, then the initial joint prediction model of each configured super parameter combination may be verified based on the verification data set to obtain a model prediction result, and finally the model prediction result may be evaluated based on the evaluation index to obtain a model evaluation score.

The verification data set may include sales process characteristics corresponding to the sales process information, and real sales volume tags corresponding to the sales process characteristics. And repeatedly and iteratively correcting the initial joint prediction model through the difference between the prediction sales label of the initial joint prediction model and the real sales label in the verification data set, so that the optimal super-parameter set can be obtained by optimizing, and the initial joint prediction model achieves the optimal prediction effect.

The evaluation index may be WAPE (Weighted Absolute Percentage Error ), recall (Recall), F1 Score (F1 Score), mean Squared Error (MSE, mean square error), RMSE (Root Mean Squared Error, root mean square error), accuracy (Accuracy), which are not limited herein.

FIG. 4 is a flow chart of finding the optimal hyper-parameters through Monte Carlo simulation. As shown in FIG. 4, a training set, a verification set and a test set are first obtained, and the training set is trained to obtain a Model, namely an initial joint prediction Model, which can perform multi-Model joint prediction. Optionally, the prediction deviation can be gradually reduced based on the model series rule, and the confidence coefficient of the model is improved. Wherein, the super parameter v ₀ 、v ₁ 、v ₂ 、v ₃ 、v ₄ A monte carlo simulation needs to be performed on a given validation set to find the optimal hyper-parameters. For each classifier, each parameter is a target value space [0,1 ]]And (3) carrying out finite equal proportion segmentation on the target value space by real numbers, dividing the interval into 5 mutually orthogonal numerical value spaces, carrying out numerical value traversal, simulating to generate prediction results of different thresholds, and calculating corresponding Metric scores (model evaluation scores), wherein the optimal parameters are super parameters corresponding to the maximum Metric. Wherein cls_0, cls_1, cls_2, cls_3, cls_4 in fig. 4 are classifiers, and multi-cls are multi-classifiers.

And 206, determining a target hyper-parameter combination and a joint prediction model according to the evaluation scores of the models.

The value of each classification model in the target hyper-parameter combination in any subinterval is a probability threshold corresponding to the classification model.

Wherein the target hyper-parameter combination may be a hyper-parameter combination of model parameters that eventually are the initial joint prediction model.

For example, in connection with step 204, if { D4, R4, Y4} is the target hyper-parameter combination, and d4=0.9, r4=0.92, y1=0.88, then 0.9 can be used as the probability threshold for the prediction probability of model1, 0.92 can be used as the probability threshold for the prediction probability of model2, and 0.88 can be used as the probability threshold for the prediction probability of model 3.

Wherein 0.9 is the value of model1 in subinterval (0.75,1), 0.92 is the value of model2 in subinterval (0.75,1), and 0.88 is the value of model3 in subinterval (0.75,1).

The above examples are only illustrative, and are not intended to limit the present disclosure.

Alternatively, the hyper-parameter combination corresponding to the highest model evaluation score in the model evaluation scores may be used as the target hyper-parameter combination, and the initial joint prediction model configured with the target hyper-parameter combination may be used as the usable joint prediction model.

For example, if the model evaluation scores corresponding to the super-parameter combinations β1, β2, and β3 are 50 points, 65 points, and 98 points, respectively, which means that the model evaluation score corresponding to β3 is the highest, β3 may be used as the target super-parameter combination.

Further, if the initial joint prediction model includes two classification models, model1, model2, and model3. If the superparameters included in the target superparameter combination β3 include F1, F2, and F3, the probability thresholds corresponding to the model1, the model2, and the model3 may be defined as F1, F2, and F3, respectively.

In the case where the model1, the model2, the model3 are respectively configured with the corresponding probability thresholds F1, F2, F3, the initial joint prediction model composed of the configured model1, model2, model3 can be used as a usable joint prediction model, and the joint prediction model can then be used for sales prediction of the target good.

In step 207, feature extraction is performed on the sales process information to obtain sales process features of the target article corresponding to the designated time period and the store.

In particular, collected sales process information may be normalized and preprocessed, including missing value processing, outlier processing, data deduplication, data format conversion, etc., and useful feature vectors, such as date features, merchandise features, store features, new product promotion features, and external features, etc., are extracted by feature engineering.

For example, store information may be converted into category type features, store information may be converted into feature vectors using one-time encoding or other encoding methods, and extracting features such as geographic location, size, and type of store may be considered. And extracting the attributes of the target goods, including the information of the type, brand, specification, cost, pricing and the like of the goods. The selling date/time can be converted into time-related characteristics such as year, month, day, season and the like, and information such as the specific time point of sales, influence of day of week on sales and the like can be extracted, and details are omitted here.

Step 208, inputting the sales process characteristics into the pre-trained joint prediction model to obtain the prediction probability output by each classification model.

Alternatively, the sales process features may first be combined with the corresponding tag data (i.e., the bi-classification results that need to be predicted) into a training set and a test set. The pre-trained joint prediction model may then be loaded, using corresponding libraries or tools to load model parameters according to the pre-trained model selected.

Further, a loaded joint prediction model may be used to input sales process features into the model and obtain the prediction probabilities output by each of the classification models. If the predictive model is a combination of multiple, bi-classified models, it is contemplated that model integration techniques (e.g., voting, weighted averaging, etc.) may be used to integrate the predictive results of the respective models. The obtained prediction probability may then be evaluated, and the performance of the model may be evaluated using the ROC curve, AUC values, etc. Meanwhile, decisions can be made or further analysis can be performed based on the predictive probabilities, depending on the traffic demands.

Specifically, the above steps may be implemented using a machine learning library in Python (e.g., scikit-learn, tensorFlow, pyTorch, etc.).

The predicted probability may be a probability that sales of the target goods to the store can be greater than a preset number in a certain period (for example, half a month) in the future. For example, store A has a probability of having a sales of more than K cell phones for 15 days in the future.

Step 209, determining a sales volume prediction label of the target goods according to the prediction probability output by each classification model and the probability threshold value corresponding to each classification model.

For example, if there are 5 classification models, cls_0, cls_1, cls_2, cls_3, cls_4, respectively, assume that the prediction probability of each classification model isThe probability threshold for each model is v _i I=0, 1,2,3,4, the sales prediction tag can be obtained by the following prediction function>

For example, ifIf the sales prediction flag is 5 in order to predict the probability of whether the sales is greater than 10, it is explained that the predicted sales is greater than 10, and the present invention is not limited thereto.

Step 210, determining a distribution mode of the target goods in the store based on the sales prediction tag.

It should be noted that, the specific implementation of step 210 may refer to the above embodiment, and will not be described herein.

In the embodiment of the disclosure, first, sales process information of a target commodity sold by a shop in a designated period is acquired, then, a target value interval associated with each classification model is acquired, then, the target value interval is divided to obtain a plurality of sub-intervals corresponding to the classification models, then, a plurality of super parameter combinations are determined according to the values of the classification models in the corresponding sub-intervals, then, model evaluation scores of initial joint prediction models when each super parameter combination is configured are acquired respectively, then, the target super parameter combinations and joint prediction models are determined according to the model evaluation scores, then, feature extraction is performed on the sales process information to obtain sales process features corresponding to the target commodity and the designated period and the shop, then, the sales process features are input into the joint prediction models which are pre-trained to obtain the prediction probability output by each classification model, finally, the sales quantity prediction labels of the target commodity are determined according to the prediction probability output by each classification model and the probability threshold corresponding to each classification model, and finally, the sales quantity prediction labels of the target commodity are determined according to the sales quantity prediction labels. Therefore, the size relation among the categories can be fully utilized, the scheme of globally searching the optimal threshold value is based on the grade attribute characteristics of the categories, the model deviation is reduced based on the mutual constraint relation of multiple models, the problems that the known space is too large and is greatly influenced by abnormal values are avoided, the prediction effect is relatively high, the stock days after the stock is separated in actual stock and replenishment scenes can be obviously optimized, the backorder is reduced, the turnover of the store is promoted, and the service value is created.

Fig. 5 is a flowchart of a method for determining a goods distribution mode according to a third embodiment of the present disclosure.

As shown in fig. 5, the method for determining the way of distributing goods may include the steps of:

step 301, acquiring sales process information of a shop selling a target commodity in a specified period.

And 302, extracting characteristics of the sales process information to obtain the sales process characteristics of the target goods corresponding to the designated time period and the store.

In step 303, the sales process features are input into the pre-trained joint prediction model to obtain the prediction probability output by each classification model.

It should be noted that, the specific implementation manner of steps 301 to 303 may refer to the above embodiment, and will not be described herein.

Step 304, it is determined whether the prediction probability output by each classification model is greater than the corresponding probability threshold.

It should be noted that, each of the two classification models has a corresponding probability threshold, and the probability thresholds corresponding to the different classification models may be the same or may be different, which is not limited herein.

Step 305, determining that the prediction result of the two classification models is a candidate label if the prediction probability is greater than the probability threshold, otherwise, determining that the prediction result is an empty label, wherein each classification model has a corresponding candidate label.

It will be appreciated that each classification model has a corresponding predicted sales. For example, if the predicted sales of the model of the classification model is R (R > 0) and the predicted result of the classification model is a candidate label, it is indicated that the sales K of the store to the target goods in the future time T satisfy the relationship: k is greater than R.

If the prediction result of the classification model is an empty label, the sales K of the store to the target goods in the future time T is described to satisfy the relation: k is less than or equal to R.

Wherein the candidate labels may be used to represent sales magnitudes predicted by the classification model. Since the predicted sales of different classification models are different, the corresponding candidate labels may also be different.

For example, if there are 5 classification models, model1, model2, model3, model4, model5, respectively, and the corresponding predicted sales are sequentially increased. The candidate labels corresponding to the model1, the model2, the model3, the model4 and the model5 are V1, V2, V3, V4 and V5 respectively.

In the embodiment of the present disclosure, the null tag may be denoted as V0, which is not limited herein.

As one example, V1 may represent a sales amount of 0 and model1 may be used to predict whether the sales of the store are greater than 0 during future time T.

V2 can represent sales K1 (K1 > 0), i.e., model2 can be used to predict whether store sales are greater than K1 at future time T.

V3 can represent sales K2 (K2 > K1), i.e., model3 can be used to predict whether store sales are greater than K3 at future time T.

V4 may represent sales K3 (K3 > K2), i.e., model4 may be used to predict whether store sales are greater than K3 at future time T.

V5 can represent sales K4 (K4 > K3), i.e., model3 can be used to predict whether store sales are greater than K4 at future time T.

For example, if the prediction probability of the model2 output is 0.8 and the probability threshold corresponding to the model2 is 0.9, the empty tag V0 may be used as the prediction result of the model2, that is, the sales K of the target item in the future time T satisfies the relationship: k is less than or equal to K1.

If the prediction probability of the model2 output is 0.95, V2 may be used as the prediction result of the model2, that is, the sales K of the target product in the future time T satisfies the relationship: k is greater than K1.

Step 306, a sequence number for each classification model is determined.

Wherein the serial number may be an identification number of each classification model.

It should be noted that, since the sales volumes predicted by different classification models are different, the sales volumes of the goods predicted by each classification model may be sorted according to the sales volumes of the goods predicted by each classification model, so as to determine the serial number corresponding to each classification model.

For example, if there are 5 sub-models, model1, model12, model3, model4, and model5, respectively, each of the sub-models has a corresponding predicted pin amount, model1, model12, model3, model4, and model5 have a corresponding predicted pin amount of k1, k2, k3, k4, and k5, respectively.

Taking model1 as an example, if the predicted sales amount corresponding to model1 is k1, it is explained that model1 is used to predict whether the sales amount of the store to the target goods can reach k1 in a certain time period T in the future, and the model output result of model1 is the predicted probability that the sales amount of the store to the target goods can reach k1 in a certain time period T in the future.

If k1 < k2 < k3 < k4 < k5, the serial numbers of model1, model12, model3, model4, and model5 may be assigned to 1, 2, 3, 4, and 5, respectively, the predicted pin number corresponding to model1 may be the smallest, the predicted pin number corresponding to model5 may be the highest, and the corresponding serial number may be the largest.

Alternatively, the serial numbers of model1, model12, model3, model4, and model5 may be assigned to 5, 4, 3, 2, and 1, respectively, the predicted sales corresponding to model1 may be the largest, the predicted sales corresponding to model5 may be the highest, and the corresponding serial numbers may be the smallest.

It should be noted that the above examples are only illustrative, and are not meant to limit the present disclosure.

And step 307, determining sales prediction labels of the target goods according to the serial numbers and the prediction results of each classification model.

Alternatively, in the case that the sequence number of the first two-classification model and the sequence number of the second two-classification model are adjacent sequence numbers, and the prediction result of the first two-classification model is a candidate tag and the prediction result of the second two-classification model is an empty tag, the candidate tag of the first two-classification model may be used as the sales prediction tag.

The first classification model and the second classification model may be two classification models with sequence numbers being adjacent sequence numbers.

For example, if the sequence number of the model x is n and the sequence number of the model y is n+1, it is explained that the sequence numbers of the model x and the model y are adjacent, the model x may be used as the first two-class model and the model y may be used as the second two-class model, which is not limited herein.

Wherein the candidate labels may be used to represent sales magnitudes predicted by the classification model. Since the predicted sales of different classification models are different, the corresponding candidate labels may also be different.

For example, if there are 5 classification models, model1, model2, model3, model4, model5, respectively, and the corresponding predicted sales are sequentially increased. The candidate labels corresponding to the model1, the model2, the model3, the model4 and the model5 are V1, V2, V3, V4 and V5 respectively.

If the model2 is the first two-class model and the model3 is the second two-class model, where the prediction result of the model2 is the candidate tag V2 and the prediction result of the model3 is the null tag V0, the prediction result of the model2, that is, the candidate tag V2, may be used as the sales prediction tag, that is, the sales K of the target article in the future time T may satisfy the relationship: k2 > K1.

Alternatively, in the case that the prediction results of the two classification models are candidate labels, the candidate label corresponding to the two classification model corresponding to the first serial number may be used as the sales prediction label.

The first sequence number may be the largest sequence number or may be the smallest sequence number.

The first serial number is the serial number corresponding to the classification model with the largest predicted sales.

Alternatively, if the predicted sales corresponding to the binary model with the largest sequence number is the largest, the largest sequence number may be used as the first sequence number, and if the predicted sales corresponding to the binary model with the smallest sequence number is the largest, the smallest sequence number may be used as the first sequence number.

Alternatively, in the case where the prediction results of the respective two classification models are empty tags, the empty tags may be used as sales prediction tags.

For example, the serial numbers of model1, model12, model3, model4, and model5 are assigned to 1, 2, 3, 4, and 5, respectively. Wherein, the corresponding predicted sales of model5 is the largest, and 5 is the first sequence number.

If the predicted results of the model1, the model12, the model3, the model4 and the model5 are V1, V2, V3, V4 and V5, respectively, since the serial number of the model5 is the largest, V5 can be used as the sales prediction label, that is, the sales K of the target article in the future time T satisfies the relationship: k is greater than K4.

If the predicted results of the model1, the model12, the model3, the model4 and the model5 are all blank labels V0, then V0 may be used as the sales prediction labels, that is, the sales K representing the target article in the future time T satisfies the relationship: k=0.

It should be noted that the above examples are only illustrative, and are not meant to limit the present disclosure.

Step 308, determining the distribution mode of the target goods in the store based on the sales prediction label.

It should be noted that, the specific implementation of step 308 may refer to the above embodiment, and will not be described herein.

In the embodiment of the disclosure, sales process information of a target commodity sold by a shop in a designated period is firstly obtained, then feature extraction is carried out on the sales process information to obtain sales process features of the target commodity corresponding to the designated period and the shop, then the sales process features are input into a combined prediction model which is pre-trained to obtain the prediction probability output by each classification model, whether the prediction probability output by each classification model is larger than a corresponding probability threshold value is judged, a prediction result of the classification model is determined to be a candidate label under the condition that the prediction probability is larger than the probability threshold value, otherwise the prediction result is an empty label, wherein each classification model is provided with a corresponding candidate label, then a serial number of each classification model is determined, then the sales quantity prediction label of the target commodity is determined according to the serial number and the prediction result of each classification model, and finally the sales quantity prediction label of the target commodity in the shop is determined based on the sales quantity prediction label. Therefore, the method can fully utilize the size relation among the categories, and based on the grade attribute characteristics of the categories, the scheme of globally searching the optimal threshold value reduces model deviation based on the mutual constraint relation of multiple models, avoids the problems of overlarge knowledge space and large influence by abnormal values, has higher prediction effect, solves the ordinal classification problem based on the method of integrating learning and globally searching the optimal threshold value, has lower requirement on training data, and is more stable in training.

Fig. 6 is a schematic structural view of a determining apparatus of an article dispensing mode according to an embodiment of the present disclosure.

As shown in fig. 6, the determination device 600 of the goods distribution mode includes:

an acquisition module 610 for acquiring sales process information of a target item sold by a store in a specified period;

a first determining module 620, configured to determine sales volume prediction labels of the target goods based on the sales process information and a pre-trained joint prediction model, where the joint prediction model includes a plurality of two-classification models 630;

and the second determining module is used for determining the goods distribution mode of the target goods in the store based on the sales quantity prediction label.

Optionally, the first determining module includes:

the feature extraction unit is used for extracting features of the sales process information to obtain sales process features of the target goods corresponding to the appointed period and the store;

the obtaining unit is used for inputting the sales process characteristics into the pre-trained joint prediction model so as to obtain the prediction probability output by each classification model;

the determining unit is used for determining sales volume prediction labels of the target goods according to the prediction probability output by each classification model and the probability threshold value corresponding to each classification model.

Optionally, the determining unit includes:

the judging subunit is used for judging whether the prediction probability output by each classification model is larger than the corresponding probability threshold value;

the first determining subunit is configured to determine that, when the prediction probability is greater than the probability threshold, the prediction result of the classification model is a candidate label, and if not, the prediction result is an empty label, where each classification model has a corresponding candidate label;

a second determining subunit, configured to determine a sequence number of each of the classification models;

and the third determination subunit is used for determining sales volume prediction labels of the target goods according to the serial numbers of the classification models and the prediction results.

Optionally, the first determining subunit is specifically configured to:

taking the candidate label of the first two-classification model as the sales prediction label under the condition that the serial number of the first two-classification model and the serial number of the second two-classification model are adjacent serial numbers, the prediction result of the first two-classification model is a candidate label and the prediction result of the second two-classification model is an empty label;

or,

under the condition that the prediction results of the two classification models are candidate labels, the candidate labels corresponding to the two classification models corresponding to the first serial numbers are used as sales prediction labels, wherein the first serial numbers are serial numbers corresponding to the classification models with the largest predicted sales;

Or,

and taking the empty label as the sales volume prediction label when the prediction results of the two classification models are empty labels.

Optionally, the first determining module further includes:

the first acquisition subunit is used for acquiring a target value interval associated with each classification model;

the dividing subunit is used for dividing the target value interval to obtain a plurality of subintervals corresponding to the classification model;

a fourth determining subunit, configured to determine a plurality of hyper-parameter combinations according to values of each of the two classification models in the corresponding multiple subintervals, where the hyper-parameter combinations contain values of each of the two classification models in any subinterval;

the second obtaining subunit is used for respectively obtaining the model evaluation scores of the initial joint prediction model when each super-parameter combination is configured;

a fifth determination subunit for determining the target hyper-parameter combinations and joint prediction models based on the respective model evaluation scores,

and the value of each classification model in the target hyper-parameter combination in any subinterval is a probability threshold corresponding to the classification model.

Optionally, the fifth determining subunit is specifically configured to:

taking the hyper-parameter combination corresponding to the highest model evaluation score in the model evaluation scores as the target hyper-parameter combination,

and taking the initial joint prediction model configuring the target hyper-parameter combination as the usable joint prediction model.

Optionally, the second obtaining subunit is specifically configured to:

acquiring a verification data set and an evaluation index for verifying the initial joint prediction model;

configuring the initial joint prediction model based on the plurality of super-parameter combinations to obtain initial joint prediction models configuring the super-parameter combinations;

based on the verification data set, verifying each initial joint prediction model configured with the super-parameter combination to obtain a model prediction result;

and evaluating the model prediction result based on the evaluation index to obtain a model evaluation score.

Optionally, the sales process information at least includes one or more of the following:

store information;

order information;

competing for goods information;

time information;

weather information;

promotional campaign information.

Optionally, the second determining module is specifically configured to:

determining a first quantity of items associated with the sales prediction tag and a second quantity of items currently held by the store;

a branch amount for the target item in the store is determined based on the first item amount and the second item amount.

In the embodiment of the disclosure, sales process information of a target commodity sold by a shop in a designated period is firstly obtained, then a sales volume prediction label of the target commodity is determined based on the sales process information and a pre-trained combined prediction model, wherein the combined prediction model comprises a plurality of classification models, and finally a commodity distribution mode of the target commodity in the shop is determined based on the sales volume prediction label. Therefore, the accuracy of predicting the sales volume of goods is improved, the problem that the model prediction deviation is increased due to data sparseness is solved through a combined prediction mode of a plurality of two-classification models, reasonable goods distribution of shops is facilitated, and out-of-sales and zero-sales are avoided.

In order to achieve the above embodiments, the present disclosure further proposes an electronic device including: the method for determining the goods distribution mode according to the foregoing embodiments of the present disclosure is implemented when the processor executes the program.

In order to implement the above-mentioned embodiments, the present disclosure also proposes a non-transitory computer-readable storage medium storing a computer program which, when executed by a processor, implements a method of determining a distribution manner of goods as proposed in the foregoing embodiments of the present disclosure.

In order to implement the above-mentioned embodiments, the present disclosure also proposes a computer program product which, when executed by an instruction processor in the computer program product, performs a method of determining a way of sorting goods as proposed by the foregoing embodiments of the present disclosure.

Fig. 7 illustrates a block diagram of an exemplary electronic device suitable for use in implementing embodiments of the present disclosure. The electronic device 12 shown in fig. 7 is merely an example and should not be construed as limiting the functionality and scope of use of the disclosed embodiments.

As shown in fig. 7, the electronic device 12 is in the form of a general purpose computing device. Components of the electronic device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, a bus 18 that connects the various system components, including the system memory 28 and the processing units 16.

Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, and a local bus using any of a variety of bus architectures. By way of example, and not limitation, such architectures include industry standard architecture (Industry Standard Architecture, ISA) bus, micro channel architecture (Micro Channel Architecture, MAC) bus, enhanced ISA bus, video electronics standards association (Video Electronics Standards Association, VESA) local bus, and peripheral component interconnect (Peripheral Component Interconnection, PCI) bus.

Electronic device 12 typically includes a variety of computer system readable media. Such media can be any available media that is accessible by electronic device 12 and includes both volatile and nonvolatile media, removable and non-removable media.

Memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (Random Access Memory, RAM) 30 and/or cache memory 32. The electronic device 12 may further include other removable/non-removable, volatile/nonvolatile computer system storage media. By way of example only, storage system 34 may be used to read from or write to non-removable, nonvolatile magnetic media (not shown in FIG. 7, commonly referred to as a "hard disk drive"). Although not shown in fig. 7, a disk drive for reading from and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading from or writing to a removable non-volatile optical disk such as, for example, optical disk read only memory (Compact Disc Read Only Memory, CD-ROM), digital versatile disk read only memory (Digital Video Disc Read Only Memory, DVD-ROM), or other optical media, may be provided. In such cases, each drive may be coupled to bus 18 through one or more data medium interfaces. Memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to carry out the functions of the various embodiments of the disclosure.

A program/utility 40 having a set (at least one) of program modules 42 may be stored in, for example, memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each or some combination of which may include an implementation of a network environment. Program modules 42 generally perform the functions and/or methods in the embodiments described in this disclosure.

The electronic device 12 may also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), one or more devices that enable a user to interact with the electronic device 12, and/or any devices (e.g., network card, modem, etc.) that enable the electronic device 12 to communicate with one or more other computing devices. Such communication may occur through an input/output (I/O) interface 22. Also, the electronic device 12 may communicate with one or more networks (e.g., local area network (Local Area Network, LAN), wide area network (Wide Area Network, WAN) and/or public network, such as the internet) via the network adapter 20. As shown, the network adapter 20 communicates with other modules of the electronic device 12 over the bus 18. It should be appreciated that although not shown, other hardware and/or software modules may be used in connection with electronic device 12, including, but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, data backup storage systems, and the like.

The processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, for example, implementing the methods mentioned in the foregoing embodiments.

In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.

Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present disclosure, the meaning of "a plurality" is at least two, such as two, three, etc., unless explicitly specified otherwise.

Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps of the process, and additional implementations are included within the scope of the preferred embodiment of the present disclosure in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the embodiments of the present disclosure.

Logic and/or steps represented in the flowcharts or otherwise described herein, e.g., a ordered listing of executable instructions for implementing logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program is printed, as the program may be electronically captured, via, for instance, optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

It should be understood that portions of the present disclosure may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. As with the other embodiments, if implemented in hardware, may be implemented using any one or combination of the following techniques, as is well known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.

Those of ordinary skill in the art will appreciate that all or a portion of the steps carried out in the method of the above-described embodiments may be implemented by a program to instruct related hardware, where the program may be stored in a computer readable storage medium, and where the program, when executed, includes one or a combination of the steps of the method embodiments.

Furthermore, each functional unit in the embodiments of the present disclosure may be integrated in one processing module, or each unit may exist alone physically, or two or more units may be integrated in one module. The integrated modules may be implemented in hardware or in software functional modules. The integrated modules may also be stored in a computer readable storage medium if implemented in the form of software functional modules and sold or used as a stand-alone product.

The above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, or the like. Although embodiments of the present disclosure have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the present disclosure, and that variations, modifications, alternatives, and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the present disclosure.

Claims (13)

1. A method of determining a manner of sorting goods, comprising:

acquiring sales process information of a shop selling target goods in a specified period;

determining sales volume prediction labels of the target goods based on the sales process information and a pre-trained combined prediction model, wherein the combined prediction model comprises a plurality of classification models;

and determining the goods distribution mode of the target goods in the store based on the sales volume prediction label.

2. The method of claim 1, wherein the determining sales volume prediction tags for the target good based on the sales process information and a pre-trained joint prediction model comprises:

extracting characteristics of the sales process information to obtain sales process characteristics of the target goods corresponding to the appointed period and the store;

Inputting the sales process characteristics into the pre-trained joint prediction model to obtain the prediction probability output by each classification model;

and determining sales prediction labels of the target goods according to the prediction probability output by each classification model and the probability threshold value corresponding to each classification model.

3. The method of claim 2, wherein determining the sales prediction tag of the target item according to the prediction probability corresponding to each of the classification models and the probability threshold corresponding to each of the classification models comprises:

judging whether the prediction probability output by each classification model is larger than the corresponding probability threshold value;

under the condition that the prediction probability is larger than the probability threshold, determining that a prediction result of the classification model is a candidate label, otherwise, a blank label is obtained, wherein each classification model has a corresponding candidate label;

determining the serial number of each classification model;

and determining sales volume prediction labels of the target goods according to the serial numbers of the classification models and the prediction results.

4. The method of claim 3, wherein said determining sales prediction tags for said target item based on a serial number of each of said classification models and said prediction results comprises:

Taking the candidate label of the first two-classification model as the sales prediction label under the condition that the serial number of the first two-classification model and the serial number of the second two-classification model are adjacent serial numbers, the prediction result of the first two-classification model is a candidate label and the prediction result of the second two-classification model is an empty label;

or,

under the condition that the prediction results of the two classification models are candidate labels, the candidate labels corresponding to the two classification models corresponding to the first serial numbers are used as sales prediction labels, wherein the first serial numbers are serial numbers corresponding to the classification models with the largest predicted sales;

or,

and taking the empty label as the sales volume prediction label when the prediction results of the two classification models are empty labels.

5. The method of claim 2, further comprising, prior to determining the sales prediction tags for the target item based on the prediction probabilities output by each of the classification models and the probability thresholds corresponding to each of the classification models:

acquiring a target value interval associated with each classification model;

dividing the target value interval to obtain a plurality of subintervals corresponding to the classification model;

Determining a plurality of super-parameter combinations according to the values of the two classification models in the corresponding sub-intervals, wherein the super-parameter combinations contain the values of each two classification model in any sub-interval;

respectively obtaining model evaluation scores of the initial joint prediction model when each super-parameter combination is configured;

determining the target hyper-parameter combination and a joint prediction model according to each model evaluation score,

and the value of each classification model in the target hyper-parameter combination in any subinterval is a probability threshold corresponding to the classification model.

6. The method of claim 5, wherein said determining said target hyper-parameter combinations and joint prediction models based on each of said model evaluation scores comprises:

taking the hyper-parameter combination corresponding to the highest model evaluation score in the model evaluation scores as the target hyper-parameter combination,

and taking the initial joint prediction model configuring the target hyper-parameter combination as the usable joint prediction model.

7. The method of claim 5, wherein separately obtaining model evaluation scores for the initial joint prediction model in configuring each super-parameter set comprises:

Acquiring a verification data set and an evaluation index for verifying the initial joint prediction model;

configuring the initial joint prediction model based on the plurality of super-parameter combinations to obtain initial joint prediction models configuring the super-parameter combinations;

based on the verification data set, verifying each initial joint prediction model configured with the super-parameter combination to obtain a model prediction result;

and evaluating the model prediction result based on the evaluation index to obtain a model evaluation score.

8. The method of claim 1, wherein the sales process information comprises at least one or more of:

store information;

order information;

competing for goods information;

time information;

weather information;

promotional campaign information.

9. The method of claim 1, wherein the determining a distribution of the target item at the store based on the sales volume prediction tag comprises:

determining a first quantity of items associated with the sales prediction tag and a second quantity of items currently held by the store;

a branch amount for the target item in the store is determined based on the first item amount and the second item amount.

10. A determination apparatus of a mode of sorting goods, characterized by comprising:

the acquisition module is used for acquiring sales process information of a shop selling target goods in a specified period;

the first determining module is used for determining sales volume prediction labels of the target goods based on the sales process information and a pre-trained combined prediction model, wherein the combined prediction model comprises a plurality of classification models;

and the second determining module is used for determining the goods distribution mode of the target goods in the store based on the sales quantity prediction label.

11. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing a method of determining the manner of dispensing items as claimed in any one of claims 1 to 9 when the program is executed by the processor.

12. A computer-readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements a method of determining a way of distributing goods according to any one of claims 1-9.

13. A computer program product comprising a computer program which, when executed by a processor, implements a method of determining a way of sorting goods according to any one of claims 1-9.