Model training method, product sales prediction method and product sales analysis method

Abstract

The embodiment of the invention provides a model training method, a product sales prediction method and a product sales analysis method.The model training method comprises the steps that sample product information of a plurality of sample products is obtained, and the sample product information comprises product sales of the sample products and element parameters of sales elements; a target element parameter of the target sales volume element is determined based on a factor parameter of a target element factor, the target sales volume element is at least one of the sales volume elements, and the target element factor is at least one element factor influencing the element strength of the target sales volume element; and training the initial analysis model according to the sample product information to obtain a target analysis model. The trained target analysis model not only can accurately predict the product sales volume, but also can accurately analyze the internal relation among the product sales volume, the sales volume elements and the element factors, and improves the depth of fine attribution of the product sales volume.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of artificial intelligence technology, and in particular to model training methods, product sales prediction methods, and product sales analysis methods. Background Technology

[0002] In insurance product management, sales forecasting and attribution analysis are two core capabilities supporting product decision-making: forecasting helps anticipate market reactions and guides pricing, marketing resource allocation, and capacity planning; attribution reveals the driving logic behind sales, clarifying which product design elements truly influence consumer choices. Combining these two approaches enables a shift from "passive response" to "proactive design." Therefore, accurately identifying and quantifying the impact of different design elements on sales has become crucial for enhancing product competitiveness.

[0003] However, traditional models either overfit to noise or lack business interpretability, failing to effectively guide product optimization. Therefore, there is an urgent need for a model training method that can simultaneously improve model prediction accuracy and attribution depth. Summary of the Invention

[0004] In view of the above, embodiments of this specification provide a model training method. One or more embodiments of this specification also relate to a product sales forecasting method, a product sales analysis method, a model training device, a product sales forecasting device, a product sales analysis device, a computing device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0005] According to a first aspect of the embodiments of this specification, a model training method is provided, comprising: The sample product information of multiple sample products is obtained respectively. The sample product information includes the product sales volume and the element parameters of the sales volume element. The target element parameters of the target sales volume element are determined based on the factor parameters of the target element factor. The target sales volume element is at least one of the sales volume elements, and the target element factor is at least one element factor that affects the element strength of the target sales volume element. Based on the sample product information, the initial analysis model is trained to obtain the target analysis model.

[0006] According to a second aspect of the embodiments of this specification, a product sales forecasting method is provided, comprising: Obtain target product information, including sales volume factor parameters; The target analysis model is invoked to process the element parameters and obtain the target predicted sales volume of the target product. The target analysis model is trained based on the model training method.

[0007] According to a third aspect of the embodiments of this specification, a product sales analysis method is provided, comprising: Obtain target product information, including target product sales volume and sales volume factor parameters; The target analysis model is invoked to analyze the sales volume and factor parameters of the target product, and to obtain the influence index of the sales volume factor. The influence index is used to describe the degree of influence of the sales volume factor on the sales volume of the target product. The target analysis model is trained based on the model training method.

[0008] According to a fourth aspect of the embodiments of this specification, a model training apparatus is provided, comprising: The first acquisition module is configured to acquire sample product information of multiple sample products respectively. The sample product information includes the product sales volume and the element parameters of the sales volume element of the sample product. The target element parameters of the target sales volume element are determined based on the factor parameters of the target element factor. The target sales volume element is at least one of the sales volume elements, and the target element factor is at least one element factor that affects the element strength of the target sales volume element. The training module is configured to train the initial analysis model based on the sample product information to obtain the target analysis model.

[0009] According to a fifth aspect of the embodiments of this specification, a product sales forecasting device is provided, comprising: The second acquisition module is configured to acquire target product information of the target product, wherein the target product information includes the element parameters of the sales element; The processing module is configured to call the target analysis model, process the element parameters, and obtain the target predicted sales of the target product. The target analysis model is trained based on the model training method.

[0010] According to a sixth aspect of the embodiments of this specification, a product sales analysis device is provided, comprising: The third acquisition module is configured to acquire target product information of the target product, wherein the target product information includes target product sales volume and element parameters of sales volume factors; The analysis module is configured to call the target analysis model to analyze the sales volume and factor parameters of the target product and obtain the influence index of the sales volume factor. The influence index is used to describe the degree of influence of the sales volume factor on the sales volume of the target product. The target analysis model is trained based on the model training method.

[0011] According to a seventh aspect of the embodiments of this specification, a computing device is provided, comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the methods provided in the first, second, or third aspects described above.

[0012] According to an eighth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores a computer program / instructions that, when executed by a processor, implement the steps of the methods provided in the first, second, or third aspects described above.

[0013] According to a ninth aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the methods provided in the first, second, or third aspects described above.

[0014] The model training method provided in one embodiment of this specification effectively solves the problem that traditional sales regression models struggle to characterize the impact of factor factors (such as coverage amount, deductible, and reimbursement ratio) on sales through a two-round hierarchical modeling paradigm of "product sales volume - sales volume factors" and "target factor parameters - target factor factors." First, in the second round of modeling, target factor factors (such as coverage amount under hospitalization liability) are aggregated into target factor parameters reflecting "factor strength," transforming abstract target factor factors into quantifiable and comparable numerical features. Then, in the first round of modeling, the initial analysis model is trained jointly based on the target factor parameters and the factor parameters of other sales volume factors (such as marketing expenses and premiums). This accurately captures the transmission effect of underlying factor factors on product sales volume through factor strength while preserving high-level business semantics. The resulting target analysis model can not only accurately predict product sales volume but also precisely analyze the intrinsic relationships between product sales volume, sales volume factors, and factor factors, improving the depth of refined attribution of product sales volume. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a model training method provided in one embodiment of this specification; Figure 2 This is a schematic diagram illustrating product information for an insurance product provided in one embodiment of this specification; Figure 3 This is a schematic diagram illustrating the processing procedure of a model training method provided in one embodiment of this specification; Figure 4 This is a flowchart illustrating the processing procedure of a model training method provided in one embodiment of this specification. Figure 5 This is a flowchart illustrating a product sales forecasting method provided in one embodiment of this specification; Figure 6This is a flowchart illustrating a product sales analysis method provided in one embodiment of this specification; Figure 7 This is an architecture diagram of a task processing system provided in one embodiment of this specification; Figure 8 This is a schematic diagram of the structure of a model training device provided in one embodiment of this specification; Figure 9 This is a schematic diagram of the structure of a product sales forecasting device provided in one embodiment of this specification; Figure 10 This is a schematic diagram of the structure of a product sales analysis device provided in one embodiment of this specification; Figure 11 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0016] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0017] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items. The term “at least one” as used in one or more embodiments of this specification means “one or more,” and “a plurality of” means “two or more.” The term “comprising” is an open-ended description and should be understood as “including but not limiting,” and may include other content in addition to what has been described.

[0018] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0019] Furthermore, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in one or more embodiments of this specification are obtained through open-source datasets or public datasets that comply with their license agreements, or are obtained with full authorization from the relevant parties. Moreover, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0020] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0021] A regression model is a mathematical model used in statistics and machine learning to study the relationship between variables. Its core objective is to predict or explain the value of a dependent variable (response variable) based on one or more independent variables (explanatory variables).

[0022] Elastic Net Regression (UPR) is a regularized linear regression method that combines the advantages of Ridge Regression and Lasso Regression. It achieves feature selection through L1 regularization and alleviates multicollinearity through L2 regularization, making it particularly suitable for modeling scenarios with high dimensionality and variable correlation. In the context of insurance product sales analysis, when faced with numerous potential influencing factors and highly correlated features, UPR can provide more stable and interpretable model results than Lasso or Ridge Regression alone.

[0023] L1 regularization (Lasso regularization) is a constraint that adds the sum of the absolute values ​​of the weight parameters as a penalty term to the loss function of a linear model. Its core function is to automatically select features and sparsify the model by forcing the weights of unimportant features to shrink to exactly zero. It is particularly suitable for identifying key driving variables in high-dimensional data.

[0024] L2 regularization (Ridge regularization) adds the sum of squares of the weight parameters as a penalty term to the loss function of a linear model. It does not compress the weights to zero, but rather reduces the magnitude of all weights as a whole, effectively suppressing overfitting, improving model stability, and still providing robust coefficient estimates when multicollinearity exists in the features.

[0025] Discretization is a technique in feature engineering that transforms continuous numerical variables into discrete (or categorical) variables. This serves various purposes, including simplifying models, handling nonlinear relationships, reducing the impact of outliers, and improving the performance of certain algorithms. In the embodiments described in this specification, discretization methods include equal-frequency binning and binning based on business rules or domain knowledge.

[0026] Feature selection refers to selecting a subset of variables that are useful for predicting the target variable from a large number of original input variables (features), while eliminating redundant or irrelevant variables.

[0027] Multicollinearity refers to a highly linear correlation between two or more independent variables in a regression model. Feature selection is about "picking out the truly important variables," while multicollinearity is about "variables being too similar, leading to model confusion." L1 regularization excels at feature selection, L2 regularization (Ridge) excels at multicollinearity, and Elastic Net combines the advantages of both.

[0028] Equal Frequency Binning, also known as Quantile Binning, is a data preprocessing method that divides continuous variables into several discrete intervals. Its core idea is to ensure that each bin (interval) contains approximately the same number of samples. Suppose there is a continuous variable (such as an insurance customer's annual income) with N=10000 samples. If it needs to be divided into k=5 bins, equal frequency binning can find the 20th, 40th, 60th, and 80th quantiles as split points. The result is that each bin contains approximately 10000 / 5=2000 samples.

[0029] K-means clustering is a classic unsupervised machine learning algorithm that aims to divide a set of unlabeled data points into K non-overlapping clusters, such that the data points within each cluster are as similar as possible (usually measured by Euclidean distance), while the data points between different clusters are as different as possible. The algorithm iteratively optimizes to minimize the sum of the squared distances from all points within each cluster to the cluster center (i.e., the "mean" or "centroid").

[0030] A DataFrame is a two-dimensional tabular data structure with row indexes and column names. Each column can be of a different data type (such as integers, floating-point numbers, strings, etc.), supporting efficient data cleaning, filtering, aggregation, and analysis operations.

[0031] This specification provides a model training method, and also relates to a product sales forecasting method, a product sales analysis method, a model training device, a product sales forecasting device, a product sales analysis device, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0032] See Figure 1 , Figure 1 This specification shows a flowchart of a model training method provided in one embodiment, which specifically includes the following steps: Step 102: Obtain sample product information for multiple sample products. The sample product information includes the product sales volume and the element parameters of the sales volume factor. The target element parameter of the target sales volume factor is determined based on the factor parameters of the target element factor. The target sales volume factor is at least one of the sales volume factors, and the target element factor is at least one element factor that affects the element strength of the target sales volume factor.

[0033] It should be noted that sample products refer to specific products used for model training, or observational instances of products within specific time, region, channel, version, and other dimensions. Sample products can be products from different scenarios, such as insurance products in an insurance scenario, retail products in an e-commerce retail scenario, financial products in a fintech scenario, and so on. For example, an insurance product could be a million-dollar medical insurance plan (Type A) sold online by an insurance company in a certain region. Similarly, a financial product could be a wealth management product offered by a bank targeting young customers.

[0034] Sample product information refers to a structured dataset describing sample products. This information is essential for model training and includes both target and independent variables. It provides the initial analytical model with a basis for learning "how product design affects sales." Taking insurance products as an example, the sample product information includes 3200 units sold, and the key parameters for sales factors include premium of 350, inpatient liability score of 9, and outpatient / emergency liability score of 5.

[0035] Product sales volume refers to the actual number of sample products sold within a specific period. It can be used as a target variable in supervised learning to reflect the market's acceptance of the sample product.

[0036] Sales volume factors refer to product attributes or external factors that directly affect product sales, typically stemming from the product itself, market environment, consumer characteristics, and sales channels. Sales volume factors are the first-level input features of the initial processing model, representing business dimensions that have a potential impact on product sales. They can serve as core independent variables in regression models to establish a statistical relationship with sales volume. Taking insurance products as an example, sales volume factors can include two categories: commercial factors and liability factors. Commercial factors include premiums and marketing expenses. Liability factors can be understood as coverage elements, such as inpatient coverage, outpatient coverage, emergency coverage, and bed fees. Premiums are the fees that policyholders pay to the insurance company periodically or in a lump sum to obtain insurance coverage. Marketing expenses are the costs incurred by the insurance company in promoting and selling the insurance product, such as channel commissions, advertising fees, and operational support fees. Liability refers to the specific items for which the insurance company assumes compensation or payment obligations, as clearly listed in the insurance contract—in other words, "what is covered."

[0037] Element parameters refer to the continuous / discrete values ​​obtained by quantifying abstract or non-numerical sales factors. Different types of sales factors can be determined in different ways. Taking insurance products as an example, the element parameters of commercial factors can be directly taken as their original values ​​or standardized values. For example, assuming marketing expenses are 350 yuan, then the element parameter for marketing expenses is 350. Liability factors can be calculated using a scoring function; for example, the element parameter for hospitalization liability is calculated to be 9.

[0038] Target sales volume factors refer to higher-order factors with an internal hierarchical structure among multiple sales volume factors. For example, a target sales volume factor can be a responsibility factor. Target sales volume factors are the focus of attribution analysis and are typically the core responsibilities of the product. Target sales volume factors can guide the initial analysis model to focus on key business issues and support refined strategy development.

[0039] It's worth noting that one possible implementation is to treat all high-level elements with internal hierarchical structures as independent target sales elements and determine their parameters accordingly. Another possible implementation, given the potentially large number of target sales elements, can categorize them based on target element factors, with elements of the same category sharing a single target element parameter. For example, if two target sales elements (such as 100% and 80% reimbursement for outpatient and emergency room visits) are identical in core elements like coverage and applicable scenarios, differing only in parameters of the three factor elements: "coverage amount," "deductible," and "reimbursement ratio," then these two target sales elements can be considered to belong to the same category.

[0040] Target factor refers to the specific underlying elements that constitute the target sales element, directly affecting its strength. Target factor factors are typically fine-grained design variables in product terms, collectively defining the actual coverage level of the target sales element. For example, target factor factors for liability elements include the sum insured, deductible, and reimbursement ratio. The sum insured refers to the maximum payout limit, such as a maximum annual cumulative reimbursement of 4 million yuan for medical insurance. The deductible is the portion that the insurance company does not cover and that the customer must bear; amounts exceeding the deductible are reimbursable. The reimbursement ratio refers to the percentage of coverage exceeded by the insurance company, such as 100% reimbursement through social security settlement.

[0041] Factor parameters refer to the original values ​​or pre-processed numerical forms of target factor elements. Factor parameters can carry product design details and are the starting point for realizing the transmission chain "from factor to sales". Taking insurance products as an example, the factor parameter of the target factor "insured amount" can be 200, the factor parameter of the target factor "deductible" can be 0, and the factor parameter of the target factor "reimbursement ratio" can be 1 (i.e., 100%).

[0042] Target element parameters refer to the quantified values ​​calculated using a scoring function based on the factor parameters of the target element factors. Scoring functions include, but are not limited to, label encoding, one-hot encoding, K-means clustering, and methods based on statistical testing. For example, a scoring function could be: score(liability) = g(coverage amount) + g(deductible) + g(reimbursement ratio). Here, the score function scores the liability element, converting it into a numerical value. If liability is empty, score = 0. The g function discretizes a continuous number. Target element parameters reflect the attribute strength (e.g., coverage strength) of the target sales element and are the aggregation result from factors to elements.

[0043] See Figure 2 , Figure 2 This specification illustrates a product information diagram of an insurance product provided in one embodiment. The insurance product is divided into "Accident Insurance," "Medical Insurance," and "Life Insurance." "Accident Insurance" includes liability elements such as "Outpatient Liability," "Emergency Liability," and "Bed Fee." Based on the factor "Sum Insured," "Outpatient Liability" can be divided into three types: "Outpatient (Sum Insured 100)," "Outpatient (Sum Insured 200)," and "Outpatient (Sum Insured 500)."

[0044] In practical applications, there are various ways to obtain sample product information for multiple sample products, and the specific method chosen depends on the actual situation. This specification does not impose any limitations on this approach. One possible implementation of this specification involves receiving sample product information for each of the multiple sample products sent by the user through a client. Another possible implementation involves reading the sample product information for each of the multiple sample products from other data acquisition devices or databases.

[0045] In one optional embodiment of this specification, the above-mentioned acquisition of sample product information of multiple sample products may include the following steps: For the first sample product, obtain the product sales volume, the element parameters of the first sales element, the factor parameters of the target element factor, and the auxiliary parameters of the target element factor. The first sample product is any one of multiple sample products, and the first sales element is the sales element other than the target sales element. Based on the equal-frequency binning strategy, the auxiliary parameters are divided to obtain multiple parameter intervals and the corresponding auxiliary element parameters of the parameter intervals; Based on the factor parameters, the target parameter range is determined from multiple parameter ranges, and the auxiliary element parameters corresponding to the target parameter range are determined as the target element parameters of the target sales volume element.

[0046] It should be noted that auxiliary parameters refer to the historical value set of the same target factor from all sample products. They are used as reference data to construct binning rules, reflecting the distribution characteristics of the target factor in the overall product pool, and ensuring that the discretization results are statistically representative and comparable. For example, the auxiliary parameter is a list of values ​​for "insurance amount" in all sample products, such as [50, 100, 150, 200, 300, 400, 600].

[0047] A parameter interval refers to a range of continuous values ​​divided by applying equal-frequency binning to auxiliary parameters. Each parameter interval contains approximately the same number of samples, enabling data-driven grouping. Parameter intervals can transform continuous factors into ordered discrete levels, facilitating standardized scoring. For example, the above value list can be divided into three equal-frequency parameter intervals: a low interval [50-150], a medium interval (150-300) and a high interval (300-600).

[0048] Auxiliary element parameters refer to the discretized level labels (such as 1, 2, 3) corresponding to the parameter range. Auxiliary element parameters can represent the relative strength of factors within the corresponding parameter range and are usually set by business rules or default strategies. For example, the auxiliary element parameter corresponding to the low range is 1, the auxiliary element parameter corresponding to the medium range is 2, and the auxiliary element parameter corresponding to the high range is 3. Optionally, if a target sales volume element is missing, its target element parameter is assigned a default value of 2 (medium).

[0049] The target parameter range refers to the parameter range to which the factor parameters of the target element factor belong. By finding a matching target parameter range, the relative position of the product within the group can be determined, realizing the mapping from individual values ​​to general levels and ensuring cross-product comparability. For example, if the insured amount of an insurance product is 3 million yuan, falling into the middle range (150-300), then its target element parameter is 2.

[0050] Target element parameters refer to the auxiliary element parameters corresponding to the target parameter range. Target element parameters are the quantified results of discretized target element factors, and can be the sum of multiple factors to form a total element score, reflecting the strength of its attribute. For example, if the range determines g(insured amount) = 2, g(deductible) = 3, and g(reimbursement ratio) = 3, then the score(liability) = 8.

[0051] In one optional embodiment of this specification, in addition to dividing the auxiliary parameters based on the equal-frequency binning strategy to obtain multiple parameter intervals and the corresponding auxiliary element parameters of the parameter intervals, business experts can also be allowed to perform manual scoring "based on binning of business rules or domain knowledge".

[0052] The solution implemented in this specification, through an equal-frequency binning mechanism based on all auxiliary element parameters, transforms target element factors into discrete levels with statistical significance and business semantics, and dynamically generates target element parameters for target sales elements accordingly. This effectively solves the problem of difficulty in uniformly quantifying and horizontally comparing coverage details in insurance products. At the same time, other sales elements (such as premiums and marketing expenses) are retained as regular features, constructing a complete and hierarchical input space. This allows the trained target analysis model to accurately predict sales and precisely attribute the results to the strength of liability design, significantly improving the scientific rigor, fairness, and implementation efficiency of intelligent analysis and iteration of insurance products.

[0053] Step 104: Train the initial analysis model based on the sample product information to obtain the target analysis model.

[0054] It should be noted that the initial analysis model refers to the untrained model. The target analysis model refers to the final model trained using sample product information. The target analysis model can be used to predict the sales volume of new products and can also output the influence indicators of various sales factors (such as regression coefficients). The initial analysis model can be different regression models, such as elastic network regression, support vector regression (SVR), logistic regression, polynomial regression, stepwise regression, decision tree regression, random forest regression, etc., or a composite model composed of multiple regression models. The effectiveness of the composite model can be verified through cross-validation.

[0055] The target analysis model trained by applying the solutions in the embodiments of this specification can not only accurately predict product sales, but also accurately analyze the intrinsic relationship between product sales, sales factors, and factor factors, thereby improving the depth of refined attribution of product sales.

[0056] In one optional embodiment of this specification, the sample product information further includes the category of the sample product; the above-mentioned training of the initial analysis model based on the sample product information to obtain the target analysis model may include the following steps: Based on the category, the sample product information is classified to obtain at least one sample information set, wherein the sample product information in the sample information set corresponds to the same category. Based on at least one set of sample information, an initial analysis model is trained to obtain at least one target analysis model, wherein the target analysis model corresponds one-to-one with the set of sample information.

[0057] It should be noted that the category refers to a label used to distinguish the business type or product line to which the sample products belong, and can reflect the essential differences in the sample products in terms of coverage, customer positioning, or regulatory attributes. Taking insurance products as an example, insurance product categories include, but are not limited to, life insurance, health insurance, and accident insurance.

[0058] A sample information set refers to a subset of data consisting of multiple sample product information items of the same category. For example, product information for all life insurance products (such as term life insurance, whole life insurance, and endowment insurance) constitutes a sample information set.

[0059] A target analysis model is an analytical model trained on a specific set of sample information. This model is specifically designed to predict and attribute the sales volume of a particular product category, possessing domain adaptability. For example, target analysis models specifically designed for life insurance, health insurance, and accident insurance can be trained.

[0060] The solution implemented in this specification, by introducing a product category dimension, divides the full sample product information into multiple homogeneous sample information sets according to categories, and independently trains a dedicated target analysis model for each sample information set. This effectively solves the fundamental differences in consumer decision-making logic, factor sensitivity, and market patterns among different product lines. For example, "reimbursement ratio" has a significant impact on the sales of short-term medical insurance, but almost no effect on long-term savings products. Modeling them together would lead to signal dilution and attribution distortion. Through classification modeling, not only is the prediction accuracy and interpretive consistency of each target analysis model significantly improved, but also refined product management and intelligent decision-making closed loops can be achieved.

[0061] In one optional embodiment of this specification, training the initial analysis model based on sample product information to obtain the target analysis model may include the following steps: Input the factor parameters into the initial analysis model to obtain the predicted sales volume of the sample products; Based on the predicted sales volume and product sales volume, the initial analysis model is trained to obtain the target analysis model.

[0062] It should be noted that the predicted sales volume refers to the product sales estimate output by the initial analysis model based on the factor parameters. It is the initial analysis model's current "guess" about the market performance of the sample products and is used to compare with the actual product sales volume to calculate the loss value. The loss value can be calculated using methods including, but not limited to, mean squared error, root mean square error, and cross-entropy.

[0063] For example, taking an insurance product as an example, the process of generating predicted sales from the initial analysis model can be: Predicted sales = f(marketing expenses, premium, score(liability 1), score(liability 2)...score(liability n)). Where f represents the initial analysis model.

[0064] By applying the solutions in the embodiments of this specification and through the supervised learning process described in these embodiments, the element parameters are input into the initial analysis model to generate predicted sales. The model parameters are continuously adjusted based on the error between the predicted sales and the actual product sales, ultimately obtaining a high-precision target analysis model. This achieves end-to-end quantitative modeling from product design features to market performance. In the insurance scenario, the target analysis model can not only accurately predict the sales of new products, but also output the regression coefficients of various sales elements (such as hospitalization liability scores and premiums), clearly revealing "which element details can truly drive sales." Furthermore, since the element parameters themselves have integrated business rules (such as scoring based on target element factors), the model results possess both data-driven objectivity and business-interpretable actionability.

[0065] In one optional embodiment of this specification, before inputting the element parameters into the initial analysis model to obtain the predicted sales volume of the sample products, the following steps may also be included: The feature parameters are standardized to obtain standardized feature parameters, wherein the standardization process includes at least one of centralization and scaling. Inputting the factor parameters into the initial analysis model to obtain the predicted sales volume of the sample products can include the following steps: Standardized element parameters are input into the initial analysis model to obtain the predicted sales volume of sample products.

[0066] It should be noted that standardization refers to data preprocessing operations that center (subtract the mean) and / or scale (divide by the standard deviation or range) the feature parameters. Standardization eliminates differences in units and orders of magnitude between different feature parameters, making model training more stable and convergent faster.

[0067] Centralization refers to subtracting the mean of each sales factor's parameter from the mean of its components across multiple sample products, resulting in a new mean of 0 for that sales factor. Centralization eliminates systematic biases in the parameter values, making it easier for the initial processing model to learn the intercept term and providing a stable foundation for subsequent scaling or regularization. When using regularized linear models such as elastic network regression, centralization prevents excessively large feature means from interfering with weight estimation. For example, if the average "marketing expenses" for a batch of medical insurance products is 1000 yuan, and the actual investment for a certain insurance product is 1200 yuan, then the centralized value would be 1200 yuan. 1000 = 200; the average value of the entire "Marketing Expenses" column after centralization is 0.

[0068] Scaling refers to dividing the parameter of each sales element by its standard deviation across multiple sample products, making the standard deviation of that sales element 1 (i.e., unit variance). Scaling unifies the fluctuation range of different element parameters, bringing all element parameters to a similar numerical scale. This ensures fair penalties for each sales element in the initial processing model, preventing situations where "marketing expenses (in the thousands of yuan)" are misjudged as "more important" due to their large value, while "responsibility score (in single digits)" is ignored. It also accelerates the convergence of gradient descent algorithms. For example, if the standard deviation of "marketing expenses" is 400 yuan, and the centralized value of an insurance product is 200, then after scaling, it becomes 200 / 400 = 0.5.

[0069] Standardized feature parameters refer to feature parameters that have undergone standardization and possess statistical characteristics such as zero mean, unit variance (or fixed interval).

[0070] For example, the code for standardizing feature parameters to obtain standardized feature parameters is shown below: scaler = StandardScaler() # Creates a standardizer object to transform feature values ​​into a distribution with a mean of 0 and a standard deviation of 1; x_train_scaled=scaler.fit_transform(x_train) # Calculate the mean and standard deviation based on the training set, and use them to standardize the training set; x_test_scaled=scaler.transform(x_test) # Standardize the test set using the mean and standard deviation learned from the training set, ensuring that the training and test data use the same scaling parameters.

[0071] It's worth noting that StandardScaler() ensures comparability between different features, especially when they differ in units or scales. For regularized linear models like resilient network regression, standardization helps prevent certain features from disproportionately influencing the model due to their large numerical range.

[0072] The solution implemented in this specification introduces standardization processing before model training, converting the element parameters into standardized element parameters with consistent scale and stable distribution. This effectively solves the problem of large differences in the dimensions and wide range of values ​​among product sales elements. It not only significantly improves the convergence speed and training stability of the initial analysis model, but also ensures that the regularization mechanism can be applied fairly to all sales elements, preventing high-level variables (such as marketing expenses measured in "yuan") from masking low-level but critical guarantee indicators (such as responsibility scores).

[0073] In one optional embodiment of this specification, training the initial analysis model based on sample product information to obtain the target analysis model may include the following steps: The sample product information is divided into training product information and test product information; Based on the training product information, the initial analysis model is trained to obtain an intermediate analysis model; Using the test product information, the intermediate analysis model is tested, and if the test results are satisfactory, the intermediate analysis model is determined as the target analysis model.

[0074] It should be noted that training product information refers to a subset of the sample product information used to train the initial analysis model. The initial analysis model can learn how factor parameters affect sales and fit internal weights through the training product information. Test product information refers to an independent subset of the sample product information that was not used in the initial analysis model training.

[0075] Test product information is used to simulate the performance of the initial analysis model on unknown data, evaluate its generalization ability, prevent overfitting, and ensure that the initial analysis model learns general patterns rather than training set noise. For example, assuming there are 800 insurance products, 80% of the sample product information of these 800 insurance products is used as training product information to train the regression coefficients of the elastic network regression model. The remaining 20% ​​of the sample product information is used as test product information. The specific proportion of the sample product information is set according to the actual situation, and this specification does not impose any limitations on it in the embodiments.

[0076] Intermediate analysis models refer to initial analysis models that have completed parameter learning based on training product information. While these models possess predictive capabilities, they have not yet undergone independent validation, and their reliability remains to be confirmed.

[0077] Test results refer to the test results of the intermediate analysis model on the tested product information, such as pass or fail. Test results objectively measure the predictive accuracy and stability of the intermediate analysis model on unseen data, ensuring the model has practical application value.

[0078] For example, suppose for each product, we have a set of data (if the current product does not have Responsibility 1, then x3=0): y=product sales; x1=marketing expenses; x2=insurance premium; x3=score(Responsibility 1); x4=score(Responsibility 2); x5=score(Responsibility 3). The code for dividing the sample product information to obtain training product information and test product information is shown below: # Example: Creating a simple DataFrame data={ 'x1':[1,2,3,4,5], 'x2':[2,3,4,5,6], 'x3':[3,4,5,6,7], 'x4':[3,4,5,6,7], 'x5':[3,4,5,6,7], 'y':[10,20,30,40,50] } df = pd.DataFrame(data) # Converts the dictionary data into a pandas DataFrame for easier manipulation of structured data; x=df[['x1','x2','x3','x4','x5']]# Extracts all feature columns (x1 to x5) from the DataFrame as input variable x; y=df['y']# Extracts the target column y from the DataFrame as the output variable (label); x_train,x_test,y_train,y_test=train_test_split(x,y,test_size=0.2,random_state=42) # This randomly splits the dataset into training and test sets in an 8:2 ratio, and fixes the random seed to ensure reproducible results.

[0079] The scheme implemented in this specification uses a training-test partitioning mechanism to split sample product information into a training set (including training product information) and a test set (including test product information). First, the training product information is used to fit an intermediate analysis model, and then the generalization ability is objectively verified using independent test product information. Only when the test results meet the standards is the model established as the target analysis model. This effectively avoids the risk of model overfitting and ensures that the final model not only performs well on historical data but also accurately predicts the sales volume of new products.

[0080] In one optional embodiment of this specification, the initial analysis model includes a resilient network regression model; training the initial analysis model based on sample product information to obtain the target analysis model may include the following steps: Obtain the target model parameters of the elastic network regression model, wherein the target model parameters include at least one of the regularization strength parameter and the regularization ratio parameter; Based on the target model parameters and sample product information, the elastic network regression model is trained to obtain the target analysis model.

[0081] It should be noted that the Elastic Network Regression (ELR) model is a regularized linear regression model, and its loss function includes both L1 and L2 regularization terms. The ERL model can automatically filter out sales factors that significantly impact sales (such as eliminating ineffective marketing channels) and stably estimate the joint effect of related sales factors (such as insurance coverage and reimbursement rates). By balancing sparsity and stability, the ERL model outperforms single regularization methods in high-dimensional or related feature scenarios.

[0082] Target model parameters are the key hyperparameters used to configure the resilient regression model, determining its regularization behavior and further affecting the model's generalization ability, sparsity, and prediction accuracy. Target model parameters are not learned directly from training data but must be set through validation or prior knowledge to control the complexity of the resilient regression model. For example, target model parameters might include a regularization strength parameter (alpha) = 1.0 and a regularization ratio parameter (l1_ratio) = 0.5.

[0083] The regularization strength parameter is a hyperparameter that controls the overall regularization penalty in an elastic network. The larger the regularization strength parameter, the more the model coefficients are compressed, and the simpler the model becomes; when the regularization strength parameter = 0, it degenerates into ordinary linear regression.

[0084] The regularization ratio parameter is a hyperparameter used to control the mixing ratio of L1 and L2 regularization terms, with a value ranging from [0,1]. When l1_ratio=1, it is pure Lasso (which can produce sparse solutions), and when l1_ratio=0, it is pure Ridge (preserving all features). The regularization ratio parameter can be used to adjust the trade-off between feature selection and coefficient stability in the elastic regression model. For example, l1_ratio=0.5 means that the L1 and L2 regularization terms each account for half, which can both eliminate weakly correlated sales factors and maintain a reasonable distribution of coefficients when "insurance amount" and "reimbursement ratio" are highly correlated.

[0085] In practical applications, there are various ways to obtain the target model parameters of an elastic network regression model. The specific method chosen depends on the actual situation, and the embodiments in this specification do not impose any limitations on this. In one possible implementation, the target model parameters can be obtained by summarizing prior knowledge. In another possible implementation, the target model parameters can be selected from multiple candidate model parameters using a grid search strategy and a cross-validation strategy.

[0086] For example, the code for training an elastic network regression model based on the target model parameters and sample product information to obtain the target analysis model is shown below: elastic_net = ElasticNet(alpha=1.0, l1_ratio=0.5) # Initialize the ElasticNet model using the target model parameters (assuming alpha=1.0, l1_ratio=0.5). elastic_net.fit(x_train_scaled, y_train) # Train the model. Use the standardized training features x_train_scaled and the corresponding target values ​​y_train to train the Elastic Net model so that it learns the mapping relationship between features and sales.

[0087] By applying the scheme of the embodiments in this specification, the target model parameters of the elastic network regression model are set. Among them, alpha controls the overall complexity to avoid overfitting, and l1_ratio balances feature selection and the stability of related factors. This allows the initial analysis model to automatically identify key driving factors (such as eliminating invalid liability items) and reasonably allocate the contributions of highly related factor factors (such as coverage amount and reimbursement ratio). The final target analysis model not only performs robustly on the test set, but its regression coefficients can also be directly converted into an actionable insight: "For every 1 unit increase in factor strength, sales change by X units". This provides a scientific and quantifiable decision-making basis for product refinement design and dynamic optimization.

[0088] In one optional embodiment of this specification, obtaining the target model parameters of the elastic network regression model may include the following steps: Obtain multiple candidate model parameters for the elastic network regression model, wherein the candidate model parameters include at least one of the candidate regularization strength parameter and the candidate regularization ratio parameter; Based on grid search and cross-validation strategies, target model parameters are selected from multiple candidate model parameters.

[0089] It should be noted that candidate model parameters refer to a set of predefined hyperparameter combinations used to define the candidate parameter combination space. Candidate model parameters can cover the possible values ​​of different regularization strength parameters and regularization ratio parameters, forming candidate schemes to be evaluated. This provides a basis for automatically selecting target model parameters and avoids manual trial and error. For example, candidate model parameters include candidate regularization strength parameters [0.1, 1.0, 10.0] and candidate regularization ratio parameters [0.1, 0.5, 0.9], for a total of 9 combinations.

[0090] The grid search strategy refers to an approach that exhaustively searches for all possible configurations within a predefined space of candidate parameter combinations and evaluates their performance. Specifically, it can traverse the parameter grid formed by candidate model parameters to select combinations that perform exceptionally well on validation metrics, achieving automated and unbiased selection of hyperparameters and improving model robustness. In addition to the grid search strategy, an optional embodiment of this specification can also use a random search strategy combined with cross-validation to select target model parameters from multiple candidate model parameters.

[0091] Cross-validation is a method that divides the training product information into multiple subsets, using a subset for training and the rest for validation in turn, to evaluate the stability of the model. Cross-validation strategies typically employ k-fold cross-validation (such as 5-fold) to reduce evaluation bias caused by a single partition.

[0092] For example, the code for selecting target model parameters from multiple candidate model parameters based on a grid search strategy and a cross-validation strategy is shown below: from sklearn.model_selection import GridSearchCV # Imports the grid search cross-validation tool for hyperparameter tuning; param_grid = {'alpha':[0.1,1.0,10.0],'l1_ratio':[0.1,0.5,0.9]} # Defines the combination of hyperparameters to be tuned for the Elastic Net model, including candidate model parameters for alpha and l1_ratio; grid_search = GridSearchCV(ElasticNet(), param_grid, cv=5) # Creates a grid search object that will iterate through all parameter combinations and evaluate the performance of each parameter combination using 5-fold cross-validation; grid_search.fit(x_train_scaled, y_train) # Performs a grid search on the standardized training data, automatically trains and validates all parameter combinations, and finds the parameters of the target model; print("Best parameters found:", grid_search.best_params_)# Outputs the hyperparameter combination that performed very well in cross-validation as the target model parameters (e.g., {'alpha':1.0,'l1_ratio':0.5}).

[0093] The solution implemented in this specification combines a grid search strategy with a cross-validation strategy to automatically select target model parameters from multiple candidate model parameters. This effectively solves the overfitting or underfitting problem caused by improper hyperparameter settings in sales analysis using elastic network regression models. The target analysis model trained in this way not only exhibits high stability in cross-validation, but its regression coefficients can also be directly converted into actionable insights, significantly improving the scientific nature, automation level, and business implementation value of product intelligent analysis.

[0094] In one optional embodiment of this specification, after training the initial analysis model based on sample product information to obtain the target analysis model, the following steps may be further included: Output the regression coefficients of the target analysis model. The regression coefficients correspond one-to-one with the sales factors and are used to reflect the degree of influence of the corresponding sales factors on product sales.

[0095] It's important to note that regression coefficients are the weighted parameters corresponding to each sales element in the target analysis model. They represent the expected change in product sales when the sales element changes by one unit. In a regression model, regression coefficients directly reflect the direction and strength of the marginal impact of a feature on the target variable. Regression coefficients can serve as core attribution indicators, revealing which sales elements positively / negatively drive sales and the degree of this positive / negative driving force. For example, the regression coefficient for hospitalization liability is +210, indicating that for every 1-point increase in the element parameter of this sales element, sales are expected to increase by 210 units; the coefficient for premium is -80, indicating that for every 1-yuan increase in the element parameter of this sales element, sales will decrease by 80 units.

[0096] For example, the code for outputting the regression coefficients of the target analysis model is shown below: print("Coefficients:", elastic_net.coef_)# Outputs the regression coefficients (weights) corresponding to each feature in the Elastic Net model, which are used to explain the direction of influence, contribution, and relative importance of each sales factor on the prediction target (product sales).

[0097] By applying the solutions described in this specification, after obtaining the target analysis model, regression coefficients corresponding one-to-one with sales factors are further output. This transforms the complex model results into intuitive and quantifiable business insights: each regression coefficient clearly indicates the direction (positive / negative) and strength (absolute value) of the marginal impact of the corresponding sales factor on product sales. This transparent and interpretable output not only verifies the rationality of the target analysis model but also enables product managers to accurately identify high-value optimization points, achieving an efficient closed loop from "data modeling" to "strategic action," significantly improving the scientific nature of product iteration and the efficiency of resource utilization.

[0098] In one optional embodiment of this specification, after training the initial analysis model based on sample product information to obtain the target analysis model, the following steps may be further included: The target analysis model is evaluated to obtain model evaluation indicators; If the model evaluation indicators do not meet the model evaluation conditions, the target analysis model is trained until the model evaluation indicators meet the model evaluation conditions, and a target analysis model that has passed the evaluation is obtained.

[0099] It should be noted that model evaluation metrics are statistical measures used to quantify the predictive performance of a target analysis model, such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²). These metrics objectively measure the goodness of fit and generalization ability of the target analysis model on test or validation data, serving as a basis for judging whether the model is "qualified" and guiding whether further training or adjustments are necessary.

[0100] Model evaluation criteria refer to pre-defined performance standards for a model, used to determine whether the evaluation indicators are acceptable. These criteria are typically determined by business requirements or historical benchmarks, reflecting the minimum accuracy requirements for the model, preventing the deployment of inefficient or unreliable models, and ensuring decision-making quality. For example, a model evaluation criterion might require R² ≥ 0.75; otherwise, it is considered unmet.

[0101] A target analysis model that has passed evaluation refers to a final, usable model whose evaluation indicators meet the model evaluation conditions after repeated training and evaluation. A target analysis model that has passed evaluation possesses both good fitting ability and business practicality, and can be put into practical application for new product sales forecasting, factor attribution analysis, and strategy simulation.

[0102] For example, the code for evaluating the target analysis model and obtaining model evaluation metrics is shown below: y_pred = elastic_net.predict(x_test_scaled) # Use the trained Elastic Net model to predict sales based on the standardized test set features, and get the predicted sales y_pred; mse = mean_squared_error(y_test, y_pred) # Calculates the mean squared error (MSE) between the predicted sales volume and the actual product sales volume. It measures the mean squared error of the prediction. The smaller the value, the more accurate the prediction. r2 = r2_score(y_test, y_pred) # Calculate the coefficient of determination R², which reflects the proportion of the variance of the target variable in the test set explained by the target analysis model. The value range is usually from 0 to 1. The closer to 1, the better the fit. print(f"Mean Squared Error: {mse}") # Outputs the specific value of the mean squared error, which is helpful for evaluating the model's accuracy; print(f"R^2 Score: {r2}") # Outputs the specific numerical value of the R² score, used to comprehensively judge the model's interpretability and generalization performance.

[0103] The solution implemented in this specification allows for immediate performance evaluation after the initial training of the target analysis model. If the model evaluation metrics fail to meet the evaluation conditions, retraining is automatically triggered (e.g., adjusting target model parameters, feature engineering, or data partitioning) until a target analysis model that meets business accuracy requirements is produced. This effectively avoids the model failure risk associated with "one-time training and deployment," providing robust, reliable, and implementable intelligent decision support for product pricing, responsibility optimization, and resource allocation, significantly improving the maturity and effectiveness of data-driven product management.

[0104] See Figure 3 , Figure 3 This document illustrates a schematic diagram of the processing flow of a model training method provided in one embodiment of this specification. Specifically, the process includes: starting from "beginning" and entering the closed-loop model training process. First, through "responsibility scoring and data preparation," the factor parameters of the target element factors are transformed into quantifiable target element parameters using business rules, obtaining sample product information (including product sales volume and sales volume factor parameters) for model training. Next, the sample product information is divided into "test and training datasets," and the element parameters are "standardized" to eliminate dimensional differences. Subsequently, the core modeling stage is entered, sequentially executing "model parameter tuning" (e.g., determining target model parameters through grid search and cross-validation strategies), "model training" (fitting an elastic network regression model using the training dataset), and "model evaluation" (validating model evaluation metrics on the test dataset). If the model evaluation metrics do not meet the model evaluation conditions, the model is trained until the model evaluation metrics meet the model evaluation conditions, obtaining a model that passes the evaluation, and finally "ending" the closed-loop model training process.

[0105] The following is in conjunction with the appendix Figure 4 Taking the application of the model training method provided in this specification in an insurance scenario as an example, the model training method will be further explained. Among other things, Figure 4The flowchart of a model training method according to an embodiment of this specification is shown, which specifically includes the following steps: Step 402: Obtain sample product information for multiple sample insurance products. The sample product information includes the product sales volume and the element parameters of the sales volume factors. The sales volume factors include marketing expenses, premiums, and liabilities. The liabilities include the sum insured, deductible, and reimbursement ratio.

[0106] Step 404: Divide the sample product information to obtain training product information and test product information.

[0107] Step 406: Obtain multiple candidate model parameters for the elastic network regression model, wherein the candidate model parameters include at least one of the candidate regularization strength parameter and the candidate regularization ratio parameter; based on the grid search strategy and the cross-validation strategy, select the target model parameter from the multiple candidate model parameters.

[0108] Step 408: Train the elastic network regression model based on the target model parameters and training product information to obtain the intermediate analysis model.

[0109] Step 410: Use the test product information to test the intermediate analysis model, and if the test results are satisfactory, determine the intermediate analysis model as the target analysis model.

[0110] See Figure 5 , Figure 5 This specification illustrates a flowchart of a product sales forecasting method according to an embodiment, which specifically includes the following steps: Step 502: Obtain target product information, which includes the element parameters of sales volume.

[0111] Step 504: Call the target analysis model to process the element parameters and obtain the target predicted sales volume of the target product. The target analysis model is trained based on the model training method.

[0112] It's important to note that the target product refers to the product whose sales volume is to be predicted. The target product is the object of analysis during the model inference phase and does not yet have actual sales data or is in the pre-market stage. The target product can be a product in different scenarios, such as insurance products in an insurance scenario, retail products in an e-commerce retail scenario, financial products in a fintech scenario, etc. Taking insurance products as an example, the target product could be a high-end, comprehensive medical insurance plan that an insurance company plans to launch in 2026. Target product information refers to a structured set of data describing the target product.

[0113] Targeted sales forecast refers to the estimated sales value output by a target analysis model based on the factor parameters of the target product. Targeted sales forecast is a quantitative prediction of the future market performance of a target product, usually presented as monthly or first-year sales, and can provide data support for product pricing, resource allocation, and channel strategies. For example, the target analysis model predicts that the first-month sales of "2026 version of high-end million-dollar medical insurance" will be 3450 units. For example, target predicted sales = f(marketing expenses, premium, score(liability 1), score(liability 2)...score(liability n)). Here, f represents the target analysis model.

[0114] The descriptions of "sales factors," "factor parameters," and "target analysis model" can be found in the descriptions of the model training methods described above. These will not be repeated in the embodiments described in this manual.

[0115] By applying the solutions in the embodiments of this specification, and by calling the trained target analysis model, only the element parameters of the target product need to be input to efficiently generate the target sales forecast, thereby achieving a rapid and objective assessment of the market potential of new products. In the insurance scenario, the product team can accurately predict the impact of different design schemes on sales before the insurance product is officially launched, avoiding reliance on subjective experience or high-cost pilot projects. At the same time, since the target analysis model itself integrates hierarchical modeling logic and regularization mechanisms, the target sales forecast is not only accurate but also has strong business interpretability.

[0116] See Figure 6 , Figure 6 This specification shows a flowchart of a product sales analysis method according to an embodiment, which specifically includes the following steps: Step 602: Obtain target product information, which includes target product sales volume and element parameters of sales volume factors.

[0117] Step 604: Call the target analysis model to analyze the sales volume and factor parameters of the target product and obtain the influence index of the sales volume factor. The influence index is used to describe the degree of influence of the sales volume factor on the sales volume of the target product. The target analysis model is trained based on the model training method.

[0118] It should be noted that target product sales volume refers to the actual number of target products sold within a specific period. Target product sales volume can serve as a benchmark for attribution analysis, reflecting the market's ultimate acceptance of the product.

[0119] Impact metrics are numerical indicators used to quantify the contribution of each sales factor to the sales volume of a target product. Impact metrics can be regression coefficients of the target analysis model, reflecting "how much the sales factor increases / decreases sales volume".

[0120] By applying the solutions described in this specification, a rigorously trained target analysis model is invoked, combined with the target product's sales volume and factor parameters, to calculate the impact indicators of each sales volume factor. This enables a refined attribution of the causes of sold product sales. These impact indicators not only reflect whether the target product is "selling well" but also precisely pinpoint "why it's selling well" or "where it's lagging behind," allowing the product team to specifically strengthen strengths and correct weaknesses. Furthermore, because the target analysis model itself integrates hierarchical element construction and regularized training mechanisms, its impact indicators possess both statistical reliability and business interpretability, effectively supporting product review and iterative decision-making from experience-driven to data-driven approaches, significantly improving the scientific nature of product management.

[0121] Considering the large number of model parameters in the target analysis model and the limited computing resources of the client, the product sales forecasting method and / or product sales analysis method proposed in the embodiments of this specification can be applied to, for example... Figure 7 The task processing system shown is not limited to this. See also Figure 7 , Figure 7 This specification illustrates an architecture diagram of a task processing system according to an embodiment of the present specification. The task processing system may include a client 702 and a server 704. In one optional embodiment of this specification, client 702 is used to send target product information of the target product to server 704, wherein the target product information includes element parameters of sales volume factors; server 704 is used to call the target analysis model, process the element parameters, and obtain the target predicted sales volume of the target product, wherein the target analysis model is trained based on the model training method; and send the target predicted sales volume of the target product to client 702; client 702 is also used to receive the target predicted sales volume of the target product sent by server 704.

[0122] In another optional embodiment of this specification, client 702 is used to send target product information of the target product to server 704, wherein the target product information includes target product sales volume and element parameters of sales volume factors; server 704 is used to call the target analysis model to analyze the target product sales volume and element parameters to obtain the influence index of sales volume factors, wherein the influence index is used to describe the degree of influence of sales volume factors on target product sales volume, and the target analysis model is trained based on the model training method; and sends the influence index of sales volume factors to client 702; client 702 is also used to receive the influence index of sales volume factors sent by server 704.

[0123] like Figure 7As shown, the target analysis model is deployed in server 704. Server 704 can connect to one or more clients 702 via a local area network (LAN), wide area network (WAN), internet connection, or other types of data network. Client 702 may include, but is not limited to, smartphones, tablets, laptops, PDAs, personal computers (PCs), smart home devices, and in-vehicle devices. Client 702 can also interact with users through a graphical user interface to invoke the target analysis model, thereby implementing the product sales forecasting method and / or product sales analysis method provided in the embodiments of this specification.

[0124] It is worth noting that the product sales forecasting and / or product sales analysis methods provided in the embodiments of this specification are generally executed by the server. However, in other embodiments of this specification, if the client's runtime resources can meet the deployment and runtime conditions of the target analysis model, the client may also have similar functions to the server, thereby executing the product sales forecasting and / or product sales analysis methods provided in the embodiments of this specification. In other embodiments, the product sales forecasting and / or product sales analysis methods provided in the embodiments of this specification may also be executed jointly by the client and the server.

[0125] Corresponding to the above-described model training method embodiments, this specification also provides embodiments of model training apparatus. Figure 8 A schematic diagram of a model training apparatus provided in one embodiment of this specification is shown. Figure 8 As shown, the device includes: The first acquisition module 802 is configured to acquire sample product information of multiple sample products respectively. The sample product information includes the product sales volume and the element parameters of the sales volume element of the sample product. The target element parameters of the target sales volume element are determined based on the factor parameters of the target element factor. The target sales volume element is at least one of the sales volume elements, and the target element factor is at least one element factor that affects the element strength of the target sales volume element. Training module 804 is configured to train the initial analysis model based on sample product information to obtain the target analysis model.

[0126] Optionally, the initial analysis model includes an elastic network regression model; the training module 804 is further configured to obtain the target model parameters of the elastic network regression model, wherein the target model parameters include at least one of a regularization strength parameter and a regularization ratio parameter; and to train the elastic network regression model based on the target model parameters and sample product information to obtain the target analysis model.

[0127] Optionally, the training module 804 is further configured to obtain multiple candidate model parameters of the resilient network regression model, wherein the candidate model parameters include at least one of candidate regularization strength parameters and candidate regularization ratio parameters; and to select target model parameters from the multiple candidate model parameters based on a grid search strategy and a cross-validation strategy.

[0128] Optionally, the first acquisition module 802 is further configured to acquire, for the first sample product, the product sales volume, the element parameters of the first sales volume element, the factor parameters of the target element factor, and the auxiliary parameters of the target element factor, wherein the first sample product is any one of multiple sample products, and the first sales volume element is the sales volume element other than the target sales volume element; based on the equal-frequency binning strategy, the auxiliary parameters are divided to obtain multiple parameter intervals and the auxiliary element parameters corresponding to the parameter intervals; based on the factor parameters, the target parameter interval is determined from the multiple parameter intervals, and the auxiliary element parameters corresponding to the target parameter interval are determined as the target element parameters of the target sales volume element.

[0129] Optionally, the training module 804 is further configured to input the element parameters into the initial analysis model to obtain the predicted sales volume of the sample products; and to train the initial analysis model based on the predicted sales volume and product sales volume to obtain the target analysis model.

[0130] Optionally, the device further includes: a standard module configured to standardize the feature parameters to obtain standardized feature parameters, wherein the standardization process includes at least one of centering and scaling; and a training module 804 further configured to input the standardized feature parameters into an initial analysis model to obtain the predicted sales volume of the sample products.

[0131] Optionally, the sample product information also includes the category of the sample product; the training module 804 is further configured to classify the sample product information according to the category to obtain at least one sample information set, wherein the sample product information in the sample information set corresponds to the same category; and to train the initial analysis model according to the at least one sample information set to obtain at least one target analysis model, wherein the target analysis model corresponds one-to-one with the sample information set.

[0132] Optionally, the training module 804 is further configured to divide the sample product information to obtain training product information and test product information; train the initial analysis model based on the training product information to obtain an intermediate analysis model; test the intermediate analysis model using the test product information, and determine the intermediate analysis model as the target analysis model if the test results are satisfactory.

[0133] Optionally, the device further includes an output module configured to output regression coefficients of the target analysis model, wherein the regression coefficients correspond one-to-one with the sales factors, and the regression coefficients are used to reflect the degree of influence of the corresponding sales factors on product sales.

[0134] Optionally, the device further includes: an evaluation module configured to evaluate the target analysis model and obtain model evaluation indicators; if the model evaluation indicators do not meet the model evaluation conditions, to train the target analysis model until the model evaluation indicators meet the model evaluation conditions and obtain a target analysis model that has passed the evaluation.

[0135] The target analysis model trained by the model training method proposed in the embodiments of this specification can not only accurately predict product sales, but also accurately analyze the intrinsic relationship between product sales, sales factors, and factor factors, thereby improving the depth of refined attribution of product sales.

[0136] The above is an illustrative scheme of a model training device according to this embodiment. It should be noted that the technical solution of this model training device and the technical solution of the model training method described above belong to the same concept. For details not described in detail in the technical solution of the model training device, please refer to the description of the technical solution of the model training method described above.

[0137] Corresponding to the above-described product sales forecasting method embodiments, this specification also provides embodiments of a product sales forecasting device. Figure 9 A schematic diagram of a product sales forecasting device according to one embodiment of this specification is shown. Figure 9 As shown, the device includes: The second acquisition module 902 is configured to acquire target product information of the target product, wherein the target product information includes the element parameters of the sales element; Processing module 904 is configured to call the target analysis model, process the element parameters, and obtain the target predicted sales of the target product. The target analysis model is trained based on the model training method.

[0138] By applying the solutions in the embodiments of this specification, and by calling the trained target analysis model, only the element parameters of the target product need to be input to efficiently generate the target sales forecast, thereby achieving a rapid and objective assessment of the market potential of the new product. Since the target analysis model itself integrates hierarchical modeling logic and regularization mechanism, the target sales forecast is not only accurate, but also has strong business interpretability.

[0139] The above is an illustrative scheme of a product sales forecasting device according to this embodiment. It should be noted that the technical solution of this product sales forecasting device and the technical solution of the product sales forecasting method described above belong to the same concept. For details not described in detail in the technical solution of the product sales forecasting device, please refer to the description of the technical solution of the product sales forecasting method described above.

[0140] Corresponding to the above-described product sales analysis method embodiments, this specification also provides embodiments of a product sales analysis device. Figure 10 A schematic diagram of a product sales analysis device according to one embodiment of this specification is shown. Figure 10 As shown, the device includes: The third acquisition module 1002 is configured to acquire target product information of the target product, wherein the target product information includes target product sales volume and element parameters of sales volume factors. Analysis module 1004 is configured to call the target analysis model to analyze the sales volume and factor parameters of the target product and obtain the influence index of the sales volume factor. The influence index is used to describe the degree of influence of the sales volume factor on the sales volume of the target product. The target analysis model is trained based on the model training method.

[0141] By applying the solutions described in this specification, a rigorously trained target analysis model is invoked, combined with the target product's sales volume and factor parameters, to calculate the impact indicators of each sales volume factor. This enables a refined attribution of the causes of sold product sales. These impact indicators not only reflect whether the target product is "selling well" but also precisely pinpoint "why it's selling well" or "where it's lagging behind," allowing the product team to specifically strengthen strengths and correct weaknesses. Furthermore, because the target analysis model itself integrates hierarchical element construction and regularized training mechanisms, its impact indicators possess both statistical reliability and business interpretability, effectively supporting product review and iterative decision-making from experience-driven to data-driven approaches, significantly improving the scientific nature of product management.

[0142] The above is a schematic diagram of a product sales analysis device according to this embodiment. It should be noted that the technical solution of this product sales analysis device and the technical solution of the product sales analysis method described above belong to the same concept. Details not described in detail in the technical solution of the product sales analysis device can be found in the description of the technical solution of the product sales analysis method described above.

[0143] Figure 11 A structural block diagram of a computing device according to one embodiment of this specification is shown. The components of the computing device 1100 include, but are not limited to, a memory 1110 and a processor 1120. The processor 1120 is connected to the memory 1110 via a bus 1130, and a database 1150 is used to store data.

[0144] The computing device 1100 also includes an access device 1140, which enables the computing device 1100 to communicate via one or more networks 1160. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 1140 may include one or more of any type of wired or wireless network interface (e.g., Network Interface Card (NIC)), such as an IEEE 802.11 Wireless Local Area Networks (WLAN) interface, a Wi-MAX (World Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0145] In one embodiment of this specification, the aforementioned components of the computing device 1100 and Figure 11 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 11 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0146] The computing device 1100 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers. The computing device 1100 can also be a mobile or stationary server.

[0147] The processor 1120 is used to execute computer programs / instructions, which, when executed by the processor, implement the steps of the above-mentioned model training method, product sales prediction method, or product sales analysis method.

[0148] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device belongs to the same concept as the technical solutions of the model training method, product sales prediction method, and product sales analysis method described above. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solutions of the model training method, product sales prediction method, or product sales analysis method described above.

[0149] An embodiment of this specification also provides a computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the above-described model training method, product sales prediction method, or product sales analysis method.

[0150] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the technical solutions of the aforementioned model training method, product sales forecasting method, and product sales analysis method. Details not described in detail in the technical solution of the storage medium can be found in the descriptions of the technical solutions of the aforementioned model training method, product sales forecasting method, or product sales analysis method.

[0151] An embodiment of this specification also provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the above-described model training method, product sales prediction method, or product sales analysis method.

[0152] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product belongs to the same concept as the technical solutions of the above-mentioned model training method, product sales prediction method, and product sales analysis method. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solutions of the above-mentioned model training method, product sales prediction method, or product sales analysis method.

[0153] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0154] Computer instructions include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in computer-readable media can be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0155] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0156] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0157] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A model training method, characterized in that, The method comprises: obtaining sample product information of a plurality of sample products respectively, wherein the sample product information comprises product sales and element parameters of sales elements of the sample products, target element parameters of target sales elements are determined based on factor parameters of target element factors, the target sales elements are at least one of the sales elements, and the target element factors are at least one element factor affecting element strength of the target sales elements; training an initial analysis model based on the sample product information to obtain a target analysis model.

2. The method of claim 1, wherein, The initial analysis model comprises an elastic network regression model. The training of the initial analysis model based on the sample product information to obtain the target analysis model comprises: obtaining target model parameters of the elastic network regression model, wherein the target model parameters comprise at least one of a regularization strength parameter and a regularization proportion parameter; training the elastic network regression model based on the target model parameters and the sample product information to obtain the target analysis model.

3. The method of claim 2, wherein, The obtaining of the target model parameters of the elastic network regression model comprises: obtaining a plurality of candidate model parameters of the elastic network regression model, wherein the candidate model parameters comprise at least one of a candidate regularization strength parameter and a candidate regularization proportion parameter; screening the target model parameters from the plurality of candidate model parameters based on a grid search strategy and a cross-validation strategy.

4. The method of claim 1, wherein, The obtaining of the sample product information of the plurality of sample products comprises: for a first sample product, obtaining the product sales, element parameters of a first sales element, factor parameters of the target element factors, and auxiliary parameters of the target element factors of the first sample product, wherein the first sample product is any one of the plurality of sample products, and the first sales element is any one of the sales elements except the target sales elements; dividing the auxiliary parameters based on an equal frequency binning strategy to obtain a plurality of parameter intervals and auxiliary element parameters corresponding to the parameter intervals; determining a target parameter interval from the plurality of parameter intervals based on the factor parameters, and determining the auxiliary element parameters corresponding to the target parameter interval as the target element parameters of the target sales elements.

5. The method of claim 1, wherein, The training of the initial analysis model based on the sample product information to obtain the target analysis model comprises: inputting the element parameters into the initial analysis model to obtain predicted sales of the sample products; training the initial analysis model based on the predicted sales and the product sales to obtain the target analysis model.

6. The method of claim 5, wherein, Before the inputting of the element parameters into the initial analysis model to obtain the predicted sales of the sample products, the method further comprises: performing standardization processing on the element parameters to obtain standardized element parameters, wherein the standardization processing comprises at least one of centering processing and scaling processing; the inputting of the element parameters into the initial analysis model to obtain the predicted sales of the sample products comprises: Input the standardized element parameter into the initial analysis model to obtain a predicted sales volume of the sample product.

7. The method of claim 1, wherein, The sample product information further comprises a category of the sample product; The method of training the initial analysis model according to the sample product information to obtain a target analysis model comprises: Classify the sample product information according to the category to obtain at least one sample information set, wherein the sample product information in the sample information set corresponds to the same category; Train the initial analysis model according to the at least one sample information set to obtain at least one target analysis model, wherein the target analysis model corresponds to the sample information set one by one.

8. The method of claim 1, wherein, The method of training the initial analysis model according to the sample product information to obtain a target analysis model comprises: Divide the sample product information to obtain training product information and test product information; Train the initial analysis model according to the training product information to obtain an intermediate analysis model; Test the intermediate analysis model by using the test product information, and determine the intermediate analysis model as the target analysis model if the test result is passed.

9. The method according to any one of claims 1 to 8, characterized in that, After the initial analysis model is trained according to the sample product information to obtain a target analysis model, the method further comprises: Output a regression coefficient of the target analysis model, wherein the regression coefficient corresponds to the sales element one by one, and the regression coefficient is used to reflect the influence degree of the corresponding sales element on the product sales volume.

10. The method according to any one of claims 1 to 8, characterized in that, After the initial analysis model is trained according to the sample product information to obtain a target analysis model, the method further comprises: Evaluate the target analysis model to obtain a model evaluation index; If the model evaluation index does not satisfy a model evaluation condition, train the target analysis model until the model evaluation index satisfies the model evaluation condition to obtain an evaluation-passed target analysis model.

11. A product sales forecasting method characterized by, The method comprises: Obtain target product information of a target product, wherein the target product information comprises element parameters of sales elements; Call a target analysis model to process the element parameters to obtain a target predicted sales volume of the target product, wherein the target analysis model is trained based on the method according to any one of claims 1 to 10.

12. A product sales analysis method characterized by comprising: The method comprises: Obtain target product information of a target product, wherein the target product information comprises a target product sales volume and element parameters of sales elements; Call a target analysis model to analyze the target product sales volume and the element parameters to obtain an influence index of the sales elements, wherein the influence index is used to describe the influence degree of the sales elements on the target product sales volume, and the target analysis model is trained based on the method according to any one of claims 1 to 10.

13. A computing device, comprising: The method comprises: A memory and a processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which realize the steps of the method according to any one of claims 1 to 12 when executed by the processor.

14. A computer-readable storage medium, characterized in that, having computer programs / instructions stored therein, which computer programs / instructions, when executed by a processor, implement the steps of the method of any one of claims 1 to 12.

15. A computer program product, characterised in that, comprising computer programs / instructions, which computer programs / instructions, when executed by a processor, implement the steps of the method of any one of claims 1 to 12.

Images