A regionalized sales forecasting system and method based on big data

Abstract

The application relates to the technical field of sales prediction, and particularly provides a regionalized sales prediction system and method based on big data, which comprises the following steps: acquiring a feature weight set of a to-be-predicted region and real-time multi-source heterogeneous data flow; acquiring a plurality of second sales impact features and a plurality of second feature combinations according to the real-time multi-source heterogeneous data flow; adjusting a first initial weight according to the difference between all the second sales impact features and corresponding first sales impact features to obtain a first prediction weight, and adjusting a second initial weight according to the difference between all the second feature combinations and corresponding first feature combinations to obtain a second prediction weight; and calculating a sales prediction result according to all the second sales impact features and corresponding first prediction weights and all the second feature combinations and corresponding second prediction weights; the method can adapt to the dynamic change of the contribution degree of the sales impact features and the feature combinations in the sales prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sales forecasting, in particular to a regionalized sales forecasting system and method based on big data. BACKGROUND

[0002] In modern enterprise operation, accurately predicting the market demand of each region in a future period of time is a key prerequisite for formulating production plans, optimizing inventory management, planning logistics distribution, and efficiently allocating marketing resources.

[0003] The existing sales forecasting method uses fixed weights for sales forecasting. Specifically, the process of the existing sales forecasting method is as follows: based on historical multi-source heterogeneous data, sales influence features (for example, product popularity features) and feature combinations (for example, the combination of product popularity features and promotion activity features) are obtained, each sales influence feature and each feature combination corresponds to a fixed weight; the sales forecasting result is calculated according to the sales influence features and their corresponding weights and the feature combinations and their corresponding weights. Since the contribution degree of the sales influence features in sales forecasting is dynamic, and the contribution degree of the feature combinations in sales forecasting is also dynamic, the existing technology cannot adapt to the dynamic changes in the contribution degree of the sales influence features and the feature combinations in sales forecasting due to the use of the fixed weight sales forecasting method, resulting in the problem that the sales forecasting result cannot accurately reflect the actual sales situation of the region, thereby resulting in low accuracy and reliability of sales forecasting.

[0004] At present, there is no effective technical solution to the above problems. It should be noted that the above information disclosed in this part is only used to understand the background of the present application concept, and therefore can contain information that does not constitute prior art. SUMMARY

[0005] The purpose of the present application is to provide a regionalized sales forecasting system and method based on big data, which can effectively solve the problem that the sales forecasting result cannot accurately reflect the actual sales situation of the region due to the use of the fixed weight sales forecasting method which cannot adapt to the dynamic changes in the contribution degree of the sales influence features and the feature combinations in sales forecasting.

[0006] In a first aspect, the present application provides a regionalized sales forecasting method based on big data, which includes the following steps:

[0007] S1, obtaining the feature weight set of the region to be predicted and the real-time multi-source heterogeneous data stream, and preprocessing the real-time multi-source heterogeneous data stream, the feature weight set including multiple sets of first sales influence features and their corresponding first initial weights and multiple sets of first feature combinations and their corresponding second initial weights;

[0008] S2, obtaining a plurality of second sales influence features and a plurality of second feature combinations according to the real-time multi-source heterogeneous data stream;

[0009] S3, adjusting a first initial weight corresponding to each second sales influence feature according to a difference between the second sales influence feature and a corresponding first sales influence feature to obtain a first predicted weight corresponding to the second sales influence feature, and adjusting a second initial weight corresponding to each second feature combination according to a difference between the second feature combination and a corresponding first feature combination to obtain a second predicted weight corresponding to the second feature combination;

[0010] S4, calculating a sales prediction result according to all the second sales influence features and the corresponding first predicted weights and all the second feature combinations and the corresponding second predicted weights.

[0011] The regional sales prediction method based on big data provided by the application can adapt the dynamic changes of the contribution degrees of the sales influence features and the feature combinations in the sales prediction by adjusting the second initial weight corresponding to each second feature combination according to the difference between the second feature combination and the corresponding first feature combination, and by adjusting the second initial weight corresponding to each second feature combination according to the difference between the second feature combination and the corresponding first feature combination, that is, the sales prediction method of the application can adapt the dynamic changes of the contribution degrees of the sales influence features and the feature combinations in the sales prediction, so the application can effectively solve the problem that the sales prediction result cannot accurately reflect the actual sales situation of the region due to the fact that the sales prediction method with fixed weights cannot adapt the dynamic changes of the contribution degrees of the sales influence features and the feature combinations in the sales prediction, thereby effectively improving the accuracy and reliability of the sales prediction.

[0012] In a second aspect, the application further provides a regional sales prediction system based on big data, comprising:

[0013] A data processing module is configured to obtain a feature weight set of a region to be predicted and a real-time multi-source heterogeneous data stream, and to pre-process the real-time multi-source heterogeneous data stream, wherein the feature weight set comprises a plurality of groups of first sales influence features and corresponding first initial weights, and a plurality of groups of first feature combinations and corresponding second initial weights;

[0014] A feature acquisition module is configured to obtain a plurality of second sales influence features and a plurality of second feature combinations according to the real-time multi-source heterogeneous data stream;

[0015] the weight confirmation module is configured to adjust the first initial weight corresponding to each second sales-influencing feature according to the difference between all the second sales-influencing features and the corresponding first sales-influencing features to obtain a first predicted weight corresponding to each second sales-influencing feature, and adjust the second initial weight corresponding to each second feature combination according to the difference between all the second feature combinations and the corresponding first feature combinations to obtain a second predicted weight corresponding to each second feature combination;

[0016] the sales prediction module is configured to calculate a sales prediction result according to all the second sales-influencing features and the corresponding first predicted weights, and all the second feature combinations and the corresponding second predicted weights.

[0017] The regional sales prediction system based on big data provided by the application can adapt the dynamic change of the contribution degree of the feature combination in the sales prediction by adjusting the second initial weight corresponding to each second feature combination according to the difference between all the second feature combinations and the corresponding first feature combinations, and adapt the dynamic change of the contribution degree of the feature combination in the sales prediction by adjusting the second initial weight corresponding to each second feature combination according to the difference between all the second feature combinations and the corresponding first feature combinations, that is, the sales prediction method of the application can adapt the dynamic change of the contribution degree of the sales-influencing feature and the feature combination in the sales prediction, so the application can effectively solve the problem that the sales prediction result cannot accurately reflect the actual sales situation of the region due to the fixed weight sales prediction method that cannot adapt the dynamic change of the contribution degree of the sales-influencing feature and the feature combination in the sales prediction, thereby effectively improving the accuracy and reliability of the sales prediction.

[0018] As can be seen from the above, the regional sales prediction system and method based on big data provided by the application can adapt the dynamic change of the contribution degree of the feature combination in the sales prediction by adjusting the second initial weight corresponding to each second feature combination according to the difference between all the second feature combinations and the corresponding first feature combinations, and adapt the dynamic change of the contribution degree of the feature combination in the sales prediction by adjusting the second initial weight corresponding to each second feature combination according to the difference between all the second feature combinations and the corresponding first feature combinations, that is, the sales prediction method of the application can adapt the dynamic change of the contribution degree of the sales-influencing feature and the feature combination in the sales prediction, so the application can effectively solve the problem that the sales prediction result cannot accurately reflect the actual sales situation of the region due to the fixed weight sales prediction method that cannot adapt the dynamic change of the contribution degree of the sales-influencing feature and the feature combination in the sales prediction, thereby effectively improving the accuracy and reliability of the sales prediction. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A flow chart of a regionalized sales prediction method based on big data provided by an embodiment of the present application.

[0020] Figure 2 A structural schematic diagram of a regionalized sales prediction system based on big data provided by an embodiment of the present application.

[0021] The reference signs: 1, data processing module; 2, feature acquisition module; 3, weight confirmation module; 4, sales prediction module. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0023] It should be noted that: similar reference signs and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0024] In a first aspect, as shown in the drawings, the present application provides a regionalized sales prediction method based on big data, which comprises the following steps: Figure 1

[0025] S1, acquiring a feature weight set of a to-be-predicted region and a real-time multi-source heterogeneous data stream, and pre-processing the real-time multi-source heterogeneous data stream, the feature weight set comprising multiple groups of first sales influence features and corresponding first initial weights and multiple groups of first feature combinations and corresponding second initial weights;

[0026] S2, acquiring multiple second sales influence features and multiple second feature combinations according to the real-time multi-source heterogeneous data stream;

[0027] ​S3, adjusting the first initial weight corresponding to each second sales impact feature according to the difference between all second sales impact features and the corresponding first sales impact features to obtain the first predicted weight corresponding to each second sales impact feature, and adjusting the second initial weight corresponding to each second feature combination according to the difference between all second feature combinations and the corresponding first feature combinations to obtain the second predicted weight corresponding to each second feature combination;

[0028] S4, calculating the sales prediction result according to all second sales impact features and the corresponding first predicted weights and all second feature combinations and the corresponding second predicted weights.

[0029] In the embodiment, the to-be-predicted region in step S1 is a region in need of sales prediction. In step S1, the feature weight set of the to-be-predicted region and the real-time multi-source heterogeneous data stream are acquired to obtain the basic data required for sales prediction. In this embodiment, the feature weight set of the to-be-predicted region and the real-time multi-source heterogeneous data stream can be acquired by reading data from a database, a file system or an external data source interface. For example, the feature weight set can be acquired by reading a database, and the real-time multi-source heterogeneous data stream can be acquired by reading data from a network platform and a sales system. In step S1, the real-time multi-source heterogeneous data stream is preprocessed, that is, the collected raw data is cleaned and converted. In this embodiment, the real-time multi-source heterogeneous data stream can be preprocessed by using an existing data cleaning algorithm to remove noise and outliers in the real-time multi-source heterogeneous data stream. In this way, the data quality of the real-time multi-source heterogeneous data stream can be improved. The feature weight set in this embodiment includes a plurality of groups of first sales impact features and the corresponding first initial weights and a plurality of groups of first feature combinations and the corresponding second initial weights. That is, the feature weight set can reflect the initial quantitative set of the influence degree of different sales impact factors and their combinations on sales. The feature weight set is preferably obtained based on historical data analysis. For example, based on the sales data and impact factors in the past year, the first sales impact features such as “product popularity” and “promotion intensity” and the initial weights (first initial weights) thereof and the first feature combinations such as “product popularity and promotion intensity combination” and the initial weights (second initial weights) thereof are determined. The real-time multi-source heterogeneous data stream in this embodiment is raw data (such as sales transaction data, duration activity data and weather data) related to sales from different sources, with different formats and continuously updated, which can provide real-time information reflecting the current market situation.

[0030] The second sales impact feature of step S2 is a specific factor extracted from the real-time multi-source heterogeneous data stream and capable of directly or indirectly affecting sales, such as product price, promotion intensity, and user discussion heat, etc., that is, this embodiment is equivalent to obtaining various independent factors currently affecting sales by obtaining the second sales impact feature. The feature combination of step S2 is a combined impact factor formed by the interaction between multiple second sales impact features, for example, the feature combination is the superposition effect of product price and promotion intensity. The multiple second sales impact features and multiple second feature combinations obtained from the real-time multi-source heterogeneous data stream of step S2 are the process of extracting factors and factor combinations currently affecting sales from the preprocessed real-time multi-source heterogeneous data stream, this embodiment can obtain the second sales impact feature by using a preset feature extraction algorithm to extract features from the real-time multi-source heterogeneous data stream, this embodiment can obtain the second feature combination by analyzing whether there is a specific association or combination relationship between the second sales impact features based on a preset feature combination identification rule, and the second sales impact features with specific association or combination relationship are taken as the feature combination, for example, the current product sales growth rate is extracted from the real-time sales data as the second sales impact feature, the user discussion heat is extracted from the social media data as the second sales impact feature, and the promotion activity intensity is extracted from the promotion activity data as the second sales intensity feature, and then the promotion activity intensity is combined with the user discussion heat as the second feature combination based on the preset feature combination identification rule.

[0031] The adjustment of the first initial weight corresponding to each second sales impact feature according to the difference between all second sales impact features and the corresponding first sales impact features in step S3 is a process of modifying the initial feature weight according to the different degrees between the real-time extracted sales impact features and the sales impact features extracted according to the historical data. In this embodiment, the adjustment of the first initial weight corresponding to each second sales impact feature according to the difference between all second sales impact features and the corresponding first sales impact features can be realized by using a weight adjustment formula based on the difference size or a lookup table method. For example, when the real-time extracted product heat (second sales impact feature) is much higher than the product heat extracted according to the historical data (first sales impact feature), the first initial weight corresponding to the product heat can be increased to obtain the first predicted weight. For another example, the mapping relationship table about the combination of impact feature differences and the combination of weight adjustment coefficients is first queried according to the difference between all second sales impact features and the corresponding first sales impact features to obtain the weight adjustment coefficient corresponding to each second sales impact feature, and then the first initial weight corresponding to the second sales impact feature is multiplied by the weight adjustment coefficient to obtain the first predicted weight. The adjustment of the second initial weight corresponding to each second feature combination according to the difference between all second feature combinations and the corresponding first feature combinations in step S3 is a process of modifying the initial combination weight according to the different degrees between the real-time extracted feature combinations and the feature combinations extracted according to the historical data. In this embodiment, the adjustment of the second initial weight corresponding to each second feature combination according to the difference between all second feature combinations and the corresponding first feature combinations is preferably similar to the adjustment of the first initial weight corresponding to each second sales impact feature according to the difference between all second sales impact features and the corresponding first sales impact features. For example, when the real-time extracted “promotion activity and product heat” combination (second feature combination) is different from the “promotion activity and product heat” combination (corresponding first feature combination) obtained by historical analysis, the second initial weight corresponding to the combination is adjusted based on the difference in the interaction intensity (for example, the sales promotion effect when appearing at the same time) of the combination to obtain the second predicted weight. It should be understood that this embodiment can adapt the first predicted weight to the dynamic change of the contribution degree of the sales impact feature in sales prediction by adjusting the first initial weight corresponding to each second sales impact feature according to the difference between all second sales impact features and the corresponding first sales impact features. This embodiment can adapt the second predicted weight to the dynamic change of the contribution degree of the feature combination in sales prediction by adjusting the second initial weight corresponding to each second feature combination according to the difference between all second feature combinations and the corresponding first feature combinations. That is, this embodiment is equivalent to dynamically adjusting the contribution degree of the sales impact feature and the feature combination in sales prediction according to the real-time market change.

[0032] The specific process of step S4 can be: multiplying the value of each second sales impact feature by its corresponding first prediction weight to obtain a plurality of weighted impact features; multiplying the value of each second feature combination by its corresponding second prediction weight to obtain a plurality of weighted feature combinations; inputting all weighted impact features and all weighted feature combinations into an existing sales prediction model to obtain a sales prediction result.

[0033] The core innovation of the present application is that the way of adjusting the second initial weight corresponding to each second feature combination according to the difference between the second feature combination and its corresponding first feature combination makes the second prediction weight adapt to the dynamic change of the contribution degree of the feature combination in sales prediction, and the way of adjusting the second initial weight corresponding to each second feature combination according to the difference between the second feature combination and its corresponding first feature combination makes the second prediction weight adapt to the dynamic change of the contribution degree of the feature combination in sales prediction, that is, the sales prediction method of the present application can adapt to the dynamic change of the contribution degree of the sales impact feature and the feature combination in sales prediction, so the present application can effectively solve the problem that the sales prediction result cannot accurately reflect the actual sales situation of the region due to the fixed weight sales prediction method cannot adapt to the dynamic change of the contribution degree of the sales impact feature and the feature combination in sales prediction, thereby effectively improving the accuracy and reliability of sales prediction.

[0034] Specifically, the method first acquires the feature weight set of the region to be predicted and the real-time multi-source heterogeneous data stream. The feature weight set provides initial weight information obtained based on historical data analysis, and the real-time data stream provides current market dynamic information. The real-time data stream is preprocessed to ensure data quality and lay a foundation for subsequent analysis. Then, the sales impact features and feature combinations reflecting the current market situation are extracted from the preprocessed real-time data stream. These real-time features and combinations represent the specific factors and their interactions that currently affect sales. Then, the real-time acquired sales impact features and feature combinations are compared with the corresponding sales impact features and feature combinations in the feature weight set, and the prediction weight reflecting the current real-time influence degree is obtained by adjusting the initial weight in the feature weight set according to the difference, so that the weight used in sales prediction is no longer fixed, but can adaptively change according to real-time conditions. Finally, the final sales prediction result is calculated using the real-time acquired sales impact features and their first prediction weights, and the real-time acquired feature combinations and their corresponding second prediction weights. The whole process forms a closed loop, and the prediction result can more accurately reflect the actual sales situation of the region through real-time data driven dynamic adjustment of the weight.

[0035] As a preferred embodiment, the scheme of the present application is implemented as follows: first, real-time multi-source heterogeneous data streams are obtained from enterprise internal sales databases, third-party market research platforms, social media platforms, etc. through a data interface, feature weight sets are obtained from a database storing sales impact features based on historical data analysis and their corresponding prediction weights as well as feature combinations and their corresponding prediction weights, and the real-time data streams are cleaned and de-duplicated. Next, discussion heat features are extracted from social media data using natural language processing algorithms, product heat features are extracted from sales data using statistical analysis methods, and promotion activity features are identified from promotion data using rule matching, and these extracted real-time features are combined and identified, for example, product heat features are combined with promotion activity features. Then, according to the corresponding sales impact features in the historical feature weight set, the first initial weight corresponding to the sales impact feature is adjusted, for example, the percentage change of the discussion heat is calculated, and the initial weight adjustment coefficient is obtained by querying the preset weight adjustment mapping table according to the percentage change. Similarly, the difference between the real-time value and the historical value of the combination of product heat features and promotion activity features is calculated, the corresponding adjustment coefficient is obtained based on the difference, and the second prediction weight is obtained by multiplying the second initial weight corresponding to the combination of product heat features and promotion activity features by the adjustment coefficient. Finally, all second sales impact features and their corresponding first prediction weights as well as all second feature combinations and their corresponding second prediction weights are input into an existing sales prediction model to obtain the sales prediction result of the area to be predicted.

[0036] Through the above scheme, the present application dynamically adjusts the weight of sales prediction according to the difference between the real-time obtained sales impact features and feature combinations and the historical corresponding items, which overcomes the disadvantage of the prior art that fixed weight cannot adapt to the dynamic change of sales impact factors, so that the sales prediction result can more accurately reflect the actual sales situation of the area, thereby improving the accuracy and reliability of sales prediction.

[0037] In some preferred embodiments, step S3 comprises:

[0038] S31, according to the difference between all second sales impact features and their corresponding first sales impact features, querying the pre-constructed mapping relationship table about impact feature difference combinations and weight adjustment coefficient combinations to obtain the first initial weight adjustment coefficient corresponding to each second sales impact feature, and according to the difference between all second feature combinations and their corresponding first feature combinations, querying the pre-constructed mapping relationship table about feature combination difference combinations and weight adjustment coefficient combinations to obtain the second initial weight adjustment coefficient corresponding to each second feature combination;

[0039] S32, obtaining the area attributes of the area to be predicted, the area attributes including market maturity and consumer preference;

[0040] S33, querying a pre-constructed mapping relationship table about the region attribute, the influence feature weight correction coefficient combination and the feature combination weight correction coefficient combination according to the region attribute to obtain a first correction coefficient corresponding to each first initial weight adjustment coefficient and a second correction coefficient corresponding to each second initial weight adjustment coefficient;

[0041] S34, calculating a first weight adjustment coefficient corresponding to each second sales influence feature according to the first initial weight adjustment coefficient and the first correction coefficient corresponding to the second sales influence feature, and calculating a second weight adjustment coefficient corresponding to each second feature combination according to the second initial weight adjustment coefficient and the second correction coefficient corresponding to the second feature combination;

[0042] S35, for each second sales influence feature, calculating a first prediction weight according to the first weight adjustment coefficient and the first initial weight corresponding thereto;

[0043] S36, for each second feature combination, calculating a second prediction weight according to the second weight adjustment coefficient and the second initial weight corresponding thereto.

[0044] The mapping relationship table of influencing feature difference combinations and weight adjustment coefficient combinations in this embodiment is a mapping relationship table that stores different influencing feature difference combinations and corresponding weight adjustment coefficient combinations. For example, when feature A increases significantly and feature B decreases slightly, the weight adjustment coefficient corresponding to feature A is 1.2, and the weight adjustment coefficient corresponding to feature B is 0.9. When feature A increases slightly and feature B decreases slightly, the weight adjustment coefficient corresponding to feature A is 1.05, and the weight adjustment coefficient corresponding to feature B is 0.95. The mapping relationship table of feature combination difference combinations and weight adjustment coefficient combinations in this embodiment is a data structure that stores weight adjustment coefficient combinations corresponding to different feature combination difference combinations. For example, when the current feature combination is "product popularity is high and there are promotional activities" and the historical feature combination is "product popularity is medium and there are no promotional activities", the weight adjustment coefficient corresponding to the feature combination difference is 1.5. The regional attributes of this embodiment are indicators used to describe the market characteristics and consumer behavior preferences of a specific geographical area. The regional attributes include market maturity and consumer preferences. The market maturity refers to the stage of regional market development. The market maturity can reflect factors affecting sales, such as the degree of market competition and consumer acceptance. The market maturity can be expressed as indicators such as the popularity rate and degree of competition of similar products in the region. The consumer preference is the inclination of consumers in the region towards specific products, brands or consumption methods. The consumer preference can be expressed as indicators such as sensitivity to specific product characteristics and prices and brand loyalty. This embodiment can obtain the regional attributes of the area to be predicted by analyzing market research data and consumer behavior reports. The mapping relationship table of regional attributes, influencing feature weight correction coefficient combinations, and feature combination weight correction coefficient combinations in this embodiment is a data structure that stores different regional attributes and corresponding weight correction coefficients for different sales influencing features and weight correction coefficients corresponding to different feature combinations. For example, when the regional attributes are high market maturity and medium price sensitivity, the weight correction coefficient corresponding to the product popularity feature is 1.1, the weight correction coefficient corresponding to the promotion activity feature is 1.08, and the weight correction coefficient corresponding to the product popularity and promotion activity combination feature is 1.05. Step S34 can calculate the first weight adjustment coefficient by multiplying the first initial weight adjustment coefficient with the corresponding first correction coefficient, and calculate the second weight adjustment coefficient by multiplying the second initial weight adjustment coefficient with the corresponding second correction coefficient. Step S35 can calculate the first predicted weight by multiplying the first weight adjustment coefficient with the corresponding first initial weight, and step S36 can calculate the second predicted weight by multiplying the second weight adjustment coefficient with the corresponding second initial weight.Since the embodiment takes into account the difference between the real-time data acquisition feature and combination and the historical feature and combination and the influence of the regional attribute of the to-be-predicted region when determining the prediction weight, i.e., the embodiment is equivalent to dynamically and accurately adjusting the initial weight based on the feature difference and the regional attribute, the embodiment can effectively improve the accuracy and reliability of the prediction weight, thereby further improving the accuracy of the sales prediction and effectively solving the problem of inaccurate weight adjustment caused by adjusting the weight based only on the feature / combination difference.

[0045] In some preferred embodiments, step S34 comprises:

[0046] S341, acquiring a sales feature set of a region adjacent to the to-be-predicted region, the sales feature set comprising a plurality of third sales quantity influence features and a plurality of third feature combinations;

[0047] S342, querying a pre-constructed mapping relationship table about sales quantity influence feature combinations, feature combination sets and weight correction coefficient combinations according to all the third sales quantity influence features and all the third feature combinations to obtain third correction coefficients corresponding to each second sales quantity influence feature and fourth correction coefficients corresponding to each second feature combination;

[0048] S343, calculating a first weight adjustment coefficient corresponding to each second sales quantity influence feature according to the first initial weight adjustment coefficient, the first correction coefficient and the third correction coefficient corresponding to each second sales quantity influence feature, and calculating a second weight adjustment coefficient corresponding to each second feature combination according to the second initial weight adjustment coefficient, the second correction coefficient and the fourth correction coefficient corresponding to each second feature combination.

[0049] The third sales impact feature of this embodiment is similar to the second sales impact feature of the above embodiment, and the third feature combination of this embodiment is similar to the second feature combination of the above embodiment, which will not be repeated here. The mapping relationship table of this embodiment about the sales impact feature combination, the feature combination set and the weight correction coefficient combination is a data structure for storing the correlation between the sales features of different adjacent regions (all combinations of the third sales impact feature and all combinations of the third feature) and the weight correction coefficient combination of the current region sales impact factor (the weight correction coefficient corresponding to all second sales impact features and the weight correction coefficient corresponding to all second feature combinations). For example, when the sales impact feature combination is that product X is highly popular in region B and region B is conducting a large-scale promotion activity, the weight correction coefficient of the product X popularity feature in region A is 0.9, and the weight correction coefficient of the "product X popularity and promotion activity combination" feature combination in region A is 0.8. This embodiment is equivalent to fully considering the influence of the sales features of the adjacent regions on the weight of the sales impact factor of the current region after obtaining the preliminary weight correction coefficient according to the attributes of the region to be predicted. This weight correction method which comprehensively considers the attributes of the region to be predicted and the sales features of the adjacent regions can make the calculated weight adjustment coefficient more reflect the actual situation, so that this embodiment can effectively improve the accuracy and reliability of the initial weight adjustment, thereby further improving the accuracy and reliability of the prediction weight, and further improving the accuracy and reliability of the sales prediction.

[0050] In some preferred embodiments, step S343 comprises:

[0051] A1, querying the pre-constructed mapping relationship table about the feature type combination and the impact feature weight correction coefficient combination according to the types of all second sales impact features to obtain a fifth correction coefficient corresponding to each second sales impact feature;

[0052] A2, querying the pre-constructed mapping relationship table about the combination type combination and the feature combination weight correction coefficient combination according to the types of all second feature combinations to obtain a sixth correction coefficient corresponding to each second feature combination;

[0053] A3, calculating a first weight adjustment coefficient corresponding to each second sales impact feature according to the first initial weight adjustment coefficient, the first correction coefficient, the third correction coefficient and the fifth correction coefficient corresponding to each second sales impact feature;

[0054] A4, calculating a second weight adjustment coefficient corresponding to each second feature combination according to the second initial weight adjustment coefficient, the second correction coefficient, the fourth correction coefficient and the sixth correction coefficient corresponding to each second feature combination.

[0055] The mapping relationship table of the embodiment about the feature type combination and the combination of the influence feature weight correction coefficients is a data structure for storing different feature type combinations (a set of types of the plurality of second sales influence features) and corresponding combinations of influence feature weight correction coefficients (a set of weight correction coefficients corresponding to the plurality of second sales influence features), for example, when the feature type combination is price, promotion and weather, the influence feature weight correction coefficient corresponding to the price is 1.05, the influence feature weight correction coefficient corresponding to the promotion is 1.1, and the influence feature weight correction coefficient corresponding to the weather is 0.98. The mapping relationship table of the embodiment about the combination type combination and the combination of the feature combination weight correction coefficients is a data structure for storing different combination type combinations (a set of types of the plurality of feature combinations) and corresponding combinations of feature combination weight correction coefficients (a set of weight correction coefficients corresponding to the plurality of feature combinations), for example, when the combination type combination is price and promotion and weather and social media, the feature combination weight correction coefficient corresponding to the price and promotion is 1.15, and the feature combination weight correction coefficient corresponding to the weather and social media is 1.02. The embodiment can more comprehensively and finely adjust the initial weight by introducing additional correction coefficients according to the types of the sales influence features and the types of the feature combinations, and combining and calculating the correction coefficients based on the differences, the regional attributes and the adjacent regional sales features, so that the embodiment can effectively improve the accuracy and reliability of the initial weight adjustment and avoid the situation that the adjustment accuracy of the initial weight adjustment is affected due to the failure to fully consider the inherent influence of the feature and combination types when adjusting the initial weight, thereby further improving the accuracy and reliability of the prediction weight, and further improving the accuracy and reliability of the sales prediction.

[0056] In some preferred embodiments, the real-time multi-source heterogeneous data stream includes a plurality of heterogeneous data streams, and step S2 includes:

[0057] S21, respectively extracting data quality features of each heterogeneous data stream to obtain quality features corresponding to each heterogeneous data stream;

[0058] S22, for each heterogeneous data stream, determining a sales influence feature extraction algorithm according to the corresponding quality features, and then extracting sales influence features from the heterogeneous data stream by using the sales influence feature extraction algorithm to obtain a plurality of second sales influence features;

[0059] S23, identifying feature combinations of all second sales influence features based on a preset feature combination identification rule to obtain a plurality of second feature combinations.

[0060] The data quality feature of this embodiment is an attribute for evaluating the reliability, availability and applicability of the data stream, which can include indicators such as integrity (whether there are missing data), accuracy (whether the data reflects the true situation), timeliness (whether the data is up-to-date) and consistency (whether the data conflicts between different sources). The sales impact feature extraction algorithm of this embodiment is an algorithm for identifying, quantifying and extracting factors or patterns related to sales changes from the data stream, which can be an existing statistical method (such as correlation analysis), a machine learning method (such as a feature selection algorithm), a natural language processing method (for text data), etc. The determination of the sales impact feature extraction algorithm according to the corresponding quality feature of this embodiment refers to the process of dynamically determining the feature extraction algorithm according to the quality status of the heterogeneous data stream. This embodiment can determine the feature extraction algorithm by querying a pre-constructed mapping relationship table between quality features and feature extraction algorithms according to the quality features, for example, the mapping relationship table is: for a heterogeneous data stream with low integrity, an algorithm insensitive to missing values or data imputation can be selected; for a heterogeneous data stream with questionable accuracy, a more robust feature extraction method or a lower weight for extracting features from the heterogeneous data stream can be used. The sales impact feature extraction of this embodiment refers to the process of extracting specific data items or indicators reflecting sales impact factors from the data stream using the determined algorithm. This embodiment can use data analysis, pattern matching, feature calculation, etc. to realize sales impact feature extraction, for example, extracting the "discount intensity" feature from the promotion data stream and the "historical sales" feature from the sales data stream. The preset feature combination identification rule of this embodiment is a rule set for determining which second sales impact features should be combined together to form a new feature combination, for example, the rule can specify that the "product popularity feature" should be combined with the "promotion activity feature", and the preset feature combination identification rule is pre-set by a person skilled in the art according to experience. Since this embodiment is equivalent to adaptively selecting or adjusting the feature extraction algorithm of the heterogeneous data stream according to the data quality of the heterogeneous data stream, this embodiment can effectively extract more reliable and more relevant second sales impact features from data of different qualities and avoid the negative impact of low-quality data on feature extraction, i.e. this embodiment can effectively improve the accuracy and reliability of second sales impact feature extraction to provide more reliable input for subsequent weight adjustment and sales prediction, thereby further improving the accuracy and reliability of sales prediction.

[0061] In some preferred embodiments, step S22 comprises:

[0062] S221, for each heterogeneous data stream, determining a sales impact feature extraction algorithm according to the type and corresponding quality feature of the heterogeneous data stream.

[0063] The type of the heterogeneous data stream of this embodiment is the category to which the heterogeneous data stream belongs (for example, text data, numerical data, image data, or time series data, etc.), and the type of the heterogeneous data stream can be identified by data source identification, metadata tagging, or content analysis. The sales impact feature extraction algorithm can be determined by querying a pre-constructed mapping relationship table of data stream types, quality characteristics, and feature extraction algorithms according to the type of the heterogeneous data stream and the corresponding quality characteristics. For example, for sales data with low integrity, an algorithm with low sensitivity to missing values can be selected; for social media data with high timeliness, an algorithm that can quickly process streaming data can be selected. The sales impact feature extraction algorithm can be more accurately matched to the data characteristics of the data stream and avoid feature extraction bias caused by mismatch between the algorithm and the data stream type by determining the sales impact feature extraction algorithm according to the type of the heterogeneous data stream and the corresponding quality characteristics. Therefore, the embodiment can effectively improve the accuracy and efficiency of extracting sales impact features from heterogeneous data of different sources and different forms, thereby further improving the accuracy and reliability of sales forecasting.

[0064] In some preferred embodiments, the real-time multi-source heterogeneous data stream includes a sales data stream, a promotion data stream, a weather data stream, and a social media data stream. The sales data stream of this embodiment can reflect the current sales situation, and is an important data source for sales forecasting. The sales data stream provides actual sales information in dimensions such as product, region, and time, and can be obtained from the enterprise's sales management system. Since promotion activities have a significant short-term stimulating effect on sales, the promotion data stream of this embodiment can reflect the impact of different promotion strategies, promotion intensity, and promotion time on sales, so that the sales forecasting method can consider the sales fluctuations caused by these external intervention factors. The promotion data stream can be obtained from the enterprise's marketing activity management platform. Since weather conditions are related to the sales of many goods, such as seasonal goods, outdoor supplies, beverages, etc., the weather data stream of this embodiment can reflect the impact of weather changes on consumer purchasing behavior and product demand, thereby enhancing the adaptability of the sales forecasting algorithm to environmental factors. The weather data stream can be obtained from a third-party weather data service interface. Since user discussions, reviews, hot topics, etc. on social media reflect public sentiment, product reputation, market trends, and potential demand, the social media data stream of this embodiment can reflect the potential impact of unstructured information on sales, such as the social media heat of a certain product indicating changes in its sales potential. The social media data stream can be obtained using a web crawler or a social media platform open interface.

[0065] In some preferred embodiments, the feature weight set is obtained based on analysis of historical sales data, a historical sales-influencing feature set and a historical feature combination set that are contemporaneous with the current time node. The current time node refers to a specific time point or time period for which the sales prediction is made, such as a specific date, week, month or quarter. The contemporaneous refers to a historical time period that has similarity or correspondence in time period with the current time node, for example, if the current time node is a certain quarter, the contemporaneous historical data can refer to the data of the same quarter in the past several years; if the current time node is during a certain holiday, the contemporaneous historical data can refer to the data during the same holiday in the past several years. The historical sales data refers to the sales data of a historical time period that has similarity or correspondence in time period with the current time node. The historical sales-influencing feature set refers to a set of various influencing factors related to sales activities in a historical time period that has similarity or correspondence in time period with the current time node. The historical feature combination set refers to the influence of the interaction between different influencing factors on sales in a historical time period that has similarity or correspondence in time period with the current time node, such as "sales of a certain product during a certain promotion" or "bad weather and online sales" and the like. The specific process of analysis based on the historical sales data, the historical sales-influencing feature set and the historical feature combination set that are contemporaneous with the current time node can be as follows: determine the time node for which the sales prediction is currently needed, such as a specific month of the next quarter; determine the historical time period that is contemporaneous with the month, such as the same month in the past three or five years; collect the sales data, the corresponding sales-influencing feature data (such as promotion activities, weather, competitor prices, etc. in the month) and the pre-identified or generated feature combination data in these contemporaneous historical time periods; train a time series analysis model or a regression model using these contemporaneous historical data, and the goal of the model is to learn the relationship between different features and feature combinations and sales results under these contemporaneous historical conditions; after the model is trained, extract the parameters (such as regression coefficients, feature importance scores, etc.) that reflect the importance or influence of the features in the model, and convert these parameters into initial weights to form the feature weight set. Since the feature weight set of this embodiment is obtained based on analysis of historical sales data, a historical sales-influencing feature set and a historical feature combination set that are contemporaneous with the current time node, this embodiment can effectively solve the problem that the first initial weight and the second initial weight are not accurate enough due to the fact that the feature weight set fails to fully reflect the historical regularity under the current prediction time node, so that the obtained feature weight set is more timely and targeted and can more accurately reflect the real influence of different features on sales in the current prediction period, thereby providing a more reliable basis for subsequent weight adjustment and sales prediction, and thus effectively improving the final sales prediction accuracy.

[0066] In some preferred embodiments, the preprocessing includes data cleaning, deduplication, format conversion and standardization. Data cleaning refers to identifying and correcting or removing errors, inconsistencies or incomplete information in the data, which can be achieved by techniques such as handling missing values, outliers, inconsistent records, etc. Deduplication refers to detecting and deleting duplicate records in the data stream, which can be achieved by methods based on unique identifiers or multi-field comparison. Format conversion refers to unifying data from different data sources with different formats into a consistent format, which can be achieved by techniques such as parsing, restructuring, mapping, etc. Standardization refers to normalizing the data to eliminate the dimensional differences between different characteristics, which can be achieved by methods such as min-max scaling, Z-score standardization, etc.

[0067] From the above, the present application provides a regional sales prediction method based on big data, which adjusts the second initial weight corresponding to each second characteristic combination according to the difference between all second characteristic combinations and their corresponding first characteristic combinations, so that the second prediction weight adapts to the dynamic change of the contribution degree of the characteristic combination in sales prediction. By adjusting the second initial weight corresponding to each second characteristic combination according to the difference between all second characteristic combinations and their corresponding first characteristic combinations, the second prediction weight adapts to the dynamic change of the contribution degree of the characteristic combination in sales prediction, i.e. the sales prediction method of the present application can adapt to the dynamic change of the contribution degree of the sales volume influencing characteristic and the characteristic combination in sales prediction. Therefore, the present application can effectively solve the problem that the sales prediction result cannot accurately reflect the actual sales situation of the region due to the fixed weight sales prediction method which cannot adapt to the dynamic change of the contribution degree of the sales volume influencing characteristic and the characteristic combination in sales prediction, thereby effectively improving the accuracy and reliability of sales prediction.

[0068] In a second aspect, as shown in Figure 2 The present application also provides a regional sales prediction system based on big data, which comprises:

[0069] A data processing module 1 is configured to obtain a feature weight set of a region to be predicted and a real-time multi-source heterogeneous data stream, and preprocess the real-time multi-source heterogeneous data stream. The feature weight set includes multiple groups of first sales volume influencing characteristics and their corresponding first initial weights, and multiple groups of first characteristic combinations and their corresponding second initial weights.

[0070] A feature acquisition module 2 is configured to obtain multiple second sales volume influencing characteristics and multiple second characteristic combinations according to the real-time multi-source heterogeneous data stream.

[0071] a weight confirmation module 3 configured to adjust the first initial weight corresponding to each second sales impact feature according to the difference between all the second sales impact features and the corresponding first sales impact features to obtain a first predicted weight corresponding to each second sales impact feature, and adjust the second initial weight corresponding to each second feature combination according to the difference between all the second feature combinations and the corresponding first feature combinations to obtain a second predicted weight corresponding to each second feature combination;

[0072] a sales prediction module 4 configured to calculate a sales prediction result according to all the second sales impact features and the corresponding first predicted weights and all the second feature combinations and the corresponding second predicted weights.

[0073] The application provides a regionalized sales prediction system based on big data, which comprises a data processing module 1, a feature acquisition module 2, a weight confirmation module 3 and a sales prediction module 4. The regionalized sales prediction system based on big data is used to execute the steps in the regionalized sales prediction method based on big data provided in the first aspect. The principle of the regionalized sales prediction system based on big data is the same as that of the regionalized sales prediction method based on big data provided in the first aspect, which will not be discussed here in detail.

[0074] As can be seen, the regionalized sales prediction system and method based on big data can adapt the second predicted weight to the dynamic change of the contribution degree of the feature combination in sales prediction by adjusting the second initial weight corresponding to each second feature combination according to the difference between all the second feature combinations and the corresponding first feature combinations. The regionalized sales prediction system and method based on big data can adapt the second predicted weight to the dynamic change of the contribution degree of the feature combination in sales prediction by adjusting the second initial weight corresponding to each second feature combination according to the difference between all the second feature combinations and the corresponding first feature combinations. That is, the sales prediction method can adapt to the dynamic change of the contribution degree of the sales impact feature and the feature combination in sales prediction. Therefore, the application can effectively solve the problem that the sales prediction result cannot accurately reflect the actual sales situation of the region due to the fact that the sales prediction method with fixed weight cannot adapt to the dynamic change of the contribution degree of the sales impact feature and the feature combination in sales prediction, thereby effectively improving the accuracy and reliability of sales prediction.

[0075] In the embodiments provided by the present application, it should be understood that the disclosed apparatus and method can be implemented in other manners. The embodiments described above are merely exemplary, for example, the division of the units is only a logical function division, and there can be another division manner in actual implementation; for example, a plurality of units or components can be combined or integrated into another machine, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different parts can be indirect couplings or communication connections through some interfaces, devices or units, and can be electric, mechanical or in other forms.

[0076] In addition, each functional module in each embodiment of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0077] In this article, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations.

[0078] The above is only an embodiment of the present application, and is not used to limit the protection scope of the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A regional sales forecasting method based on big data, characterized in that: The regional sales forecasting method based on big data includes the following steps: S1. Obtain a feature weight set and a real-time multi-source heterogeneous data stream for the area to be predicted, and preprocess the real-time multi-source heterogeneous data stream, wherein the feature weight set includes multiple groups of first sales volume influencing features and their corresponding first initial weights, and multiple groups of first feature combinations and their corresponding second initial weights; S2. Acquire multiple second sales volume influencing features and multiple second feature combinations based on the real-time multi-source heterogeneous data stream; S3. Adjusting the first initial weight corresponding to each second sales-influencing feature based on the difference between all second sales-influencing features and their corresponding first sales-influencing features to obtain a first prediction weight corresponding to each second sales-influencing feature, and adjusting the second initial weight corresponding to each second feature combination based on the difference between all second feature combinations and their corresponding first feature combinations to obtain a second prediction weight corresponding to each second feature combination; S4. Calculating a sales forecast result based on all of the second sales volume influencing features and their corresponding first prediction weights and all of the second feature combinations and their corresponding second prediction weights; Step S3 includes: S31. Based on the differences between all the second sales-influencing features and their corresponding first sales-influencing features, query a pre-constructed mapping table of difference combinations of influencing features and weight adjustment coefficient combinations to obtain a first initial weight adjustment coefficient corresponding to each of the second sales-influencing features. Based on the differences between all the second feature combinations and their corresponding first feature combinations, query a pre-constructed mapping table of difference combinations of feature combinations and weight adjustment coefficient combinations to obtain a second initial weight adjustment coefficient corresponding to each of the second feature combinations. S32. Acquire regional attributes of the area to be predicted, where the regional attributes include market maturity and consumer preferences; S33: querying a pre-built mapping relationship table of regional attributes, influencing feature weight correction coefficient combinations, and feature combination weight correction coefficient combinations based on the regional attributes to obtain a first correction coefficient corresponding to each of the first initial weight adjustment coefficients and a second correction coefficient corresponding to each of the second initial weight adjustment coefficients; S34. Calculate the first weight adjustment coefficient corresponding to each second sales-influencing feature based on the first initial weight adjustment coefficient and the first correction coefficient corresponding to each second sales-influencing feature, and calculate the second weight adjustment coefficient corresponding to each second feature combination based on the second initial weight adjustment coefficient and the second correction coefficient corresponding to each second feature combination; S35. For each of the second sales volume influencing features, calculate a first prediction weight based on the corresponding first weight adjustment coefficient and the first initial weight; S36. For each second feature combination, calculate a second prediction weight according to its corresponding second weight adjustment coefficient and second initial weight.

2. The regional sales forecasting method based on big data according to claim 1, characterized in that: Step S34 includes: S341. Obtain a sales feature set for an area adjacent to the area to be predicted, the sales feature set comprising a plurality of third sales volume influencing features and a plurality of third feature combinations; S342. Query a pre-built mapping relationship table of sales-influencing feature combinations, feature combination sets, and weight correction coefficient combinations based on all the third sales-influencing features and all the third feature combinations to obtain a third correction coefficient corresponding to each second sales-influencing feature and a fourth correction coefficient corresponding to each second feature combination. S343. Calculate the first weight adjustment coefficient corresponding to each of the second sales volume influencing characteristics based on the first initial weight adjustment coefficient, the first correction coefficient and the third correction coefficient corresponding to each of the second sales volume influencing characteristics, and calculate the second weight adjustment coefficient corresponding to each of the second feature combinations based on the second initial weight adjustment coefficient, the second correction coefficient and the fourth correction coefficient corresponding to each of the second feature combinations.

3. The regional sales forecasting method based on big data according to claim 2, characterized in that: Step S343 includes: A1. Querying a pre-built mapping table of feature type combinations and weight correction coefficient combinations of influencing features based on the types of all second sales volume influencing features to obtain a fifth correction coefficient corresponding to each second sales volume influencing feature; A2. Querying a pre-built mapping relationship table of combination type combinations and feature combination weight correction coefficient combinations based on the types of all second feature combinations to obtain a sixth correction coefficient corresponding to each second feature combination; A3. Calculate the first weight adjustment coefficient corresponding to each of the second sales volume influencing characteristics based on the first initial weight adjustment coefficient, the first correction coefficient, the third correction coefficient, and the fifth correction coefficient corresponding to each of the second sales volume influencing characteristics; A4. Calculate the second weight adjustment coefficient corresponding to each second feature combination according to the second initial weight adjustment coefficient, the second correction coefficient, the fourth correction coefficient, and the sixth correction coefficient corresponding to each second feature combination.

4. The regional sales forecasting method based on big data according to claim 1, characterized in that: The real-time multi-source heterogeneous data stream includes multiple heterogeneous data streams, and step S2 includes: S21, extracting data quality features for each of the heterogeneous data streams to obtain quality features corresponding to each of the heterogeneous data streams; S22. For each of the heterogeneous data streams, determine a sales impact feature extraction algorithm based on the corresponding quality feature, and then use the sales impact feature extraction algorithm to extract sales impact features from the heterogeneous data stream to obtain a plurality of second sales impact features. S23. Perform feature combination identification on all the second sales volume influencing features based on preset feature combination identification rules to obtain multiple second feature combinations.

5. The regional sales forecasting method based on big data according to claim 4 is characterized in that: Step S22 includes: S221. For each of the heterogeneous data streams, determine a sales impact feature extraction algorithm according to the type of the heterogeneous data stream and the corresponding quality characteristics.

6. The regional sales forecasting method based on big data according to claim 1, characterized in that: The real-time multi-source heterogeneous data streams include sales data streams, promotion data streams, weather data streams and social media data streams.

7. The regional sales forecasting method based on big data according to claim 1 is characterized in that ,The feature weight set is obtained based on the analysis of ,historical sales data for the same period as the current time node, the ,historical sales impact feature set and the historical feature ,combination set.

8. The regional sales forecasting method based on big data according to claim 1 is characterized in that ,The preprocessing includes data cleaning, deduplication, format conversion and ,standardization.

9. A regional sales forecasting system based on big data, characterized by: The regional sales forecasting system based on big data includes: a data processing module, configured to obtain a feature weight set and a real-time multi-source heterogeneous data stream for the area to be predicted, and preprocess the real-time multi-source heterogeneous data stream, wherein the feature weight set includes a plurality of groups of first sales volume influencing features and their corresponding first initial weights, and a plurality of groups of first feature combinations and their corresponding second initial weights; a feature acquisition module, configured to acquire a plurality of second sales volume influencing features and a plurality of second feature combinations based on the real-time multi-source heterogeneous data stream; a weight confirmation module, configured to adjust the first initial weight corresponding to each second sales-influencing feature based on the difference between all the second sales-influencing features and their corresponding first sales-influencing features, so as to obtain a first prediction weight corresponding to each second sales-influencing feature, and to adjust the second initial weight corresponding to each second feature combination based on the difference between all the second feature combinations and their corresponding first feature combinations, so as to obtain a second prediction weight corresponding to each second feature combination; a sales forecasting module, configured to calculate a sales forecast result based on all of the second sales volume influencing features and their corresponding first forecast weights and all of the second feature combinations and their corresponding second forecast weights; The process of adjusting the first initial weight corresponding to each second sales-influencing feature according to the difference between all the second sales-influencing features and their corresponding first sales-influencing features to obtain the first prediction weight corresponding to each second sales-influencing feature, and adjusting the second initial weight corresponding to each second feature combination according to the difference between all the second feature combinations and their corresponding first feature combinations to obtain the second prediction weight corresponding to each second feature combination includes: S31. Based on the differences between all the second sales-influencing features and their corresponding first sales-influencing features, query a pre-constructed mapping table of difference combinations of influencing features and weight adjustment coefficient combinations to obtain a first initial weight adjustment coefficient corresponding to each of the second sales-influencing features. Based on the differences between all the second feature combinations and their corresponding first feature combinations, query a pre-constructed mapping table of difference combinations of feature combinations and weight adjustment coefficient combinations to obtain a second initial weight adjustment coefficient corresponding to each of the second feature combinations. S32. Acquire regional attributes of the area to be predicted, where the regional attributes include market maturity and consumer preferences; S33: querying a pre-built mapping relationship table of regional attributes, influencing feature weight correction coefficient combinations, and feature combination weight correction coefficient combinations based on the regional attributes to obtain a first correction coefficient corresponding to each of the first initial weight adjustment coefficients and a second correction coefficient corresponding to each of the second initial weight adjustment coefficients; S34. Calculate the first weight adjustment coefficient corresponding to each second sales-influencing feature based on the first initial weight adjustment coefficient and the first correction coefficient corresponding to each second sales-influencing feature, and calculate the second weight adjustment coefficient corresponding to each second feature combination based on the second initial weight adjustment coefficient and the second correction coefficient corresponding to each second feature combination; S35. For each of the second sales volume influencing features, calculate a first prediction weight based on the corresponding first weight adjustment coefficient and the first initial weight; S36. For each second feature combination, calculate a second prediction weight according to its corresponding second weight adjustment coefficient and second initial weight.

Images