A poultry farming project marketing method and marketing system based on customer portrait

Abstract

The present invention discloses a poultry breeding project marketing method and marketing system based on customer portraits, including: customer ordering intention identification through modeling of customer historical ordering data, using machine learning models to accurately predict customer ordering intentions for four-level companies. For customers with ordering intentions, predict their ordering probability for sales varieties with specific prices, ages, and average weights, and estimate the customer's price elasticity coefficient for sales varieties. For sales varieties under certain specifications and prices, predict the customer's order quantity range, and also estimate the customer's price elasticity coefficient for sales varieties of the same grade. The present invention constructs a marketing platform based on business knowledge and price sensitivity to predict customer purchasing intentions, product ordering preferences, and order quantities, thereby achieving accurate predictions and reasonable recommendations for poultry breeding projects.

Description

Technical Field

[0001] The present invention relates to the field of smart marketing technology, and more specifically, to a poultry farming project marketing method and marketing system based on customer portraits. Background Art

[0002] For poultry breeding projects, the existing marketing strategy is to simply conduct a surface analysis of the customer's historical ordering data manually to subjectively judge the customer's ordering intention for each fourth-level company, which has problems such as low prediction accuracy and high time and labor costs. In the existing marketing systems on the market, the customer modeling cannot well meet the customization needs of the business side, and does not deeply analyze the characteristics of different customer groups. Moreover, the recommendations made only consider the customer's preference for the product, without considering the impact of the price, making it difficult to grasp the customer's true ordering intention and order quantity. In addition, many intelligent marketing systems on the market do not consider the impact of factors such as product specifications and quality on customer ordering intention and order quantity. Therefore, there is an urgent need for an intelligent poultry breeding precision marketing platform that can realize customer reserve requirement reduction marketing and match people and goods. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a poultry farming project marketing method and marketing system based on customer portraits, which aims to extract features by modeling the attendance time series patterns of different customer groups, and use the task decomposition method to accurately predict the ordering intention of specific groups of people, so as to meet business needs while complying with the actual production environment and market laws.

[0004] The first aspect of the present invention provides a poultry farming project marketing method based on customer portraits, comprising:

[0005] Obtain historical interaction data between customers and the poultry project company to build a data set and perform preprocessing. Extract customer portrait features through customers' historical ordering data and extract customer attendance context features through customers' attendance data.

[0006] Using a machine learning model to predict customer ordering intention based on the customer portrait features and attendance context features, and dividing the customer groups according to the customer ordering intention;

[0007] Obtain the item representation of the poultry farming project and the customer representation of the customer, use the multi-layer perceptron to obtain the customer's natural preference for sales varieties of different specifications through the item representation and customer representation, and combine the customer's price preference to generate the customer's total preference for sales varieties of the poultry farming project;

[0008] An order quantity prediction model is constructed, and the total preference is used for model training. The order quantity range of customers for sales products of different specifications is predicted through the original features. At the same time, the price elasticity coefficient of customers for sales products is estimated to obtain the final order quantity range.

[0009] In this solution, a machine learning model is used to predict customer ordering intention based on the customer portrait features and attendance context features, and customer groups are divided according to the customer ordering intention, specifically:

[0010] The LightGBM model is used to build an order intention prediction model. The customer portrait features and attendance context features are used as model inputs. The forward distribution algorithm is used to learn and build the corresponding decision tree through negative gradient fitting residuals in each iteration.

[0011] Through training with the constructed decision tree, the customer's ordering intention is obtained, and the customer's ordering intention is imported into the decision module. In the decision module, three customer groups, low-frequency, medium-frequency and high-frequency, are divided through data analysis, and target customers are screened according to the customer groups that different customers fall into.

[0012] In this solution, the item representation of the poultry farming project and the customer representation of the customer are obtained, specifically:

[0013] Obtaining original features of the poultry breeding project, including sales varieties, customers, age in days, and average weight, and using a multi-layer perceptron to generate item representations of the poultry breeding project based on the original features;

[0014] The customer's interaction frequency with the poultry farming project is extracted based on the historical interaction data in the data set, and the frequency-sensitive vector is obtained using a linear layer. The customer ID, sales area, preference level, and attendance context features corresponding to the customer are read using the historical interaction data, and the customer's explicit representation is obtained through a multi-layer perceptron.

[0015] The poultry farming items in the historical interaction data are represented by the attention mechanism, the shopping basket information of customers in different time periods is obtained, and the shopping basket information is used to generate the customer's historical interest representation using the long short-term memory network;

[0016] The frequency-sensitive vector, the customer explicit representation and the customer historical interest representation are used to construct a customer representation.

[0017] In this solution, a multi-layer perceptron is used to obtain the customer's natural preference for sales varieties of different specifications through item representation and customer representation, and the customer's total preference for sales varieties of poultry projects is generated by combining the customer's price preference, specifically:

[0018] The customer ordering preferences are trained in two stages. In the first stage, a multi-layer perceptron is used to obtain the customer's natural preference for different specifications of sales products through item representation and customer representation. It is expressed as: , For customers, For sales varieties, For average weight, For days, is the characterization function;

[0019] In the second stage, the model parameters of the first stage are fixed, and the total price-related preference is trained based on the natural preference in the first stage. The correlation coefficient of the price elasticity coefficient is The price-related preference is obtained by combining the natural preference and the price-related preference to obtain the total preference of the customer for the sales varieties of the poultry project. , expressed as:

[0020] ;

[0021] in, is the activation function, is the price of the product being sold, It is the average price of all varieties sold by the same poultry project company.

[0022] In this solution, dynamic negative sampling is used to train customer ordering preferences. The sales records of other customers are aggregated according to the granularity of date, sales company, and sales variety. The sales volume is summed up, and the price and age are weighted averaged according to the sales volume to construct a negative sample pool.

[0023] According to the number of sales records purchased by the customer on the same day, the sales volume is used as the weight according to the preset ratio to randomly select sales items that have not been purchased from the negative sample pool as negative samples. The positive samples are lowered to generate pseudo positive samples, and the negative samples are raised to generate pseudo negative samples.

[0024] By comparing the correlation coefficient between customers' total preference for sales varieties and price elasticity coefficient , get the price elasticity coefficient of customers for different sales varieties , .

[0025] In this solution, an order quantity prediction model is constructed, and the total preference is used for model training, specifically:

[0026] The order quantity prediction model is constructed through the neural factor decomposition mechanism. The binning standard is preset in the model, and the historical order quantity of customers is counted. The order quantity range of the first 75% is divided into equal widths, and the order quantity range of the first 75% to 95% is divided into equal frequencies. The order quantity range of the last 5% is separated into an independent interval, and the obtained order quantity intervals are numbered;

[0027] The model training is carried out in two stages using the total preferences of the customers. The first stage uses the original features consisting of customers, sales varieties, average weight, and age to predict the probability of the target customers ordering the preset specifications of the sales varieties. , , are model parameters;

[0028] The second part uses the original features consisting of customers, sales product grades, and attendance context. Based on the natural order quantity interval number with the largest probability in the first part, the price elasticity coefficient of customers for different sales product grades is added. , treat the interval number as a continuous value and slide it to get the order interval prediction result , , is the price of the product being sold, It is the average price of all varieties sold by the same poultry project company.

[0029] In this solution, set the limit range according to the order quantity interval number range 1 to N. , For truncation operations, limit the results to 1 to N;

[0030] Integerize the order quantity interval prediction results and select the closest integer value as the final order quantity interval number , , This is a rounding operation.

[0031] The second aspect of the present invention also provides a poultry farming project marketing system based on customer portraits, including a data input and preprocessing module, a customer ordering intention module, a customer ordering preference prediction module and a customer ordering quantity prediction module;

[0032] The data input and preprocessing module constructs a data set at the granularity of customer-poultry company, and establishes different features through different dimensional combinations;

[0033] The customer ordering intention module uses customer profile features and attendance context features as input to construct a decision tree, obtains the customer's ordering intention, and finally outputs the target customer through the decision strategy to deliver to the downstream task;

[0034] The customer order preference prediction module uses the natural preference and price-related preference of the multi-layer perceptron, and combines it with the generated total preference, and adopts a multi-stage training method for training. In the first stage, the customer's preference attributes for the sales variety itself are characterized by the original features to generate natural preferences. In the second stage, the model parameters of the first stage are fixed, and the total preference related to price is trained based on the natural preference of the first stage.

[0035] The customer order quantity prediction module divides the order quantity intervals for prediction, performs training based on the customer's total preference, constructs an order quantity prediction model, predicts the customer's order quantity for sales products of different specifications, and predicts the final order quantity interval based on the sales product order quantity, taking into account the grade and price.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] In view of the significant differences among different customer groups, we extract features by modeling the attendance time series of different customer groups, and use the task decomposition method to accurately predict the ordering intention of specific groups, meeting business needs while complying with the actual production environment and market rules;

[0038] In order to solve the problems of large fluctuation range of order quantity and difficulty in prediction, insufficient price diversity of the same specification of the same customer, etc., the model predicts the order quantity interval more accurately by introducing standardization of order quantity intervals, multi-stage training combining classification and regression problems, and price sensitivity at the customer-sales product level in the customer order quantity prediction module.

[0039] The data-driven poultry precision marketing system has achieved the company's goal of using data-driven intelligence to achieve customer precision marketing and matching people and goods. It has solved problems such as the subjectivity of potential customer positioning and preference identification, the cumbersome and inefficient operations of sales staff in positioning potential customers, the inaccurate grasp of customer preferences through manual methods, and the difficulty in accumulating experience effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flow chart of the poultry farming project marketing method based on customer portraits of the present invention is shown;

[0041] Figure 2 A block diagram of the poultry farming project marketing method based on customer portraits of the present invention is shown;

[0042] Figure 3 A schematic diagram showing a process of predicting a customer's ordering intention according to the present invention;

[0043] Figure 4 A flow chart showing the generation of customer representations according to the present invention is shown;

[0044] Figure 5 A flow chart of constructing an order quantity prediction model according to the present invention is shown;

[0045] Figure 6 A block diagram of a poultry farming project marketing system based on customer portraits of the present invention is shown. DETAILED DESCRIPTION

[0046] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0047] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0048] Figure 1 , 2 The flowchart and block diagram of the poultry farming project marketing method based on customer portraits of the present invention are shown.

[0049] like Figure 1 As shown, the first aspect of the embodiment of the present invention provides a poultry farming project marketing method based on customer portraits, comprising:

[0050] S102, obtaining historical interaction data between the customer and the poultry project company to construct a data set, and performing preprocessing, extracting customer portrait features through the customer's historical order data, and extracting customer attendance context features through the customer's attendance data;

[0051] S104, using a machine learning model to predict the customer's ordering intention based on the customer portrait features and attendance context features, and dividing the customer groups according to the customer's ordering intention;

[0052] S106, obtaining the item representation of the poultry breeding project and the customer representation of the customer, using a multi-layer perceptron to obtain the customer's natural preference for sales varieties of different specifications through the item representation and the customer representation, and combining the customer's price preference to generate the customer's total preference for sales varieties of the poultry breeding project;

[0053] S108, constructing an order quantity prediction model, using the total preference to train the model, predicting the customer's order quantity range for different specifications of sales products through original features, and estimating the customer's price elasticity coefficient for the sales products to obtain the final order quantity range.

[0054] It should be noted that the data used is constructed with the customer-poultry project company as the granularity to construct a data set. Different features can be established through the combination of different dimensions to alleviate the problem of data sparsity. Customer portrait features are extracted based on the customer's historical order data, and the customer's attendance data is mined to obtain the customer's attendance characteristics. In view of the large number of branches, sub-branches and business outlets in each poultry project company, the negative sampling strategy is changed from daily sampling to multi-day sampling at a daily granularity, thereby reducing the number of negative samples.

[0055] Figure 3 A schematic diagram of a process for predicting a customer's ordering intention according to the present invention is shown.

[0056] It should be noted that the lightgbm model in the machine learning model is based on the GBDT tree generation principle. The forward distribution algorithm is used. In each iteration, the residual is fitted by negative gradient to learn a decision tree. In the process of generating a decision tree, when performing feature selection node splitting, it is necessary to sort the feature values, traverse all possible division points, and then calculate the information gain to select the optimal split point. The lightgbm model is used to build an order intention prediction model, and the customer portrait features and attendance context features are used as model inputs. The forward distribution algorithm is used. In each iteration, the residual is fitted by negative gradient to learn and build the corresponding decision tree;

[0057] Through training with the constructed decision tree, the customer's ordering intention is obtained, and the customer's ordering intention is imported into the decision module. In the decision module, three customer groups, low-frequency, medium-frequency, and high-frequency, are divided through data analysis, and target customers are screened according to the customer groups that different customers fall into. This can not only test the effectiveness of the model but also meet the needs of the business side, aiming to focus on the indicator effects of medium- and high-frequency customers. This processing method can better fit the actual market rules.

[0058] It should be noted that the original features of the poultry farming project, including sales varieties, customers, age in days, and average weight, are obtained, and a multi-layer perceptron is used to generate an item representation of the poultry farming project based on the original features; the customer's interaction frequency with the poultry farming project is extracted according to the historical interaction data in the data set, and the frequency-sensitive vector is obtained using a linear layer, and the customer ID, sales area, preference level, and attendance context features corresponding to the customer are read using the historical interaction data, and the customer's explicit representation is obtained through a multi-layer perceptron; the poultry farming project in the historical interaction data is represented by an attention mechanism, and the customer's shopping basket information in different time periods is obtained. The shopping basket information is used to generate a customer historical interest representation using a long short-term memory network. The process of generating a customer representation is as follows: Figure 4 The frequency sensitive vector, the customer explicit representation and the customer historical interest representation are used to construct the customer representation.

[0059] It should be noted that the introduction of natural preference , price-related preferences and total preferences The concept of is used to describe the composition of customer preferences for sales varieties from different interactions. Specifically, the total preference is divided into two parts and modeled separately. The customer ordering preference is trained in two stages. In the first stage, a multi-layer perceptron is used to obtain the customer's natural preference for sales varieties of different specifications through item representation and customer representation. For example, a customer prefers the sales variety of Datu 2 Gong because of its delicious meat and good taste. This preference has nothing to do with price and is a natural attribute of agricultural products. The natural preference It is expressed as: , For customers, For sales varieties, For average weight, For days, is the characterization function;

[0060] The second stage is price-related preference, which is the offset of customer ordering preference due to factors such as sales price, which is the total preference As a part of the process, the model parameters of the first stage are fixed, and the total price-related preference is trained based on the natural preference of the first stage, which is determined by the correlation coefficient of the price elasticity coefficient. The price-related preference is obtained by combining the natural preference and the price-related preference to obtain the total preference of the customer for the sales varieties of the poultry project. , expressed as:

[0061] ;

[0062] in, is the activation function. In order to alleviate the situation where the price difference of different sales varieties affects the model performance, the price is standardized. is the price of the product being sold, It is the average price of all sales varieties under the same poultry project company, limiting the price fluctuation range to a smaller range.

[0063] When predicting sales varieties and specifications, fine-grained negative samples are required for training. Dynamic negative sampling is used to train customer ordering preferences. The sales records of other customers are aggregated according to the granularity of date, sales company, and sales variety. The sales volume is summed, and the price and age are weighted averaged according to the sales volume to construct a negative sample pool. According to the number of sales records purchased by the customer on that day, sales volume is used as the weight at a ratio of 1:5 to randomly select sales varieties that have not been purchased from the negative sample pool as negative samples.

[0064] When learning the impact of price on ordering preferences, a certain amount of negative samples are needed for training. Generally speaking, for the same specification of sales products, if customers are willing to buy them, then lowering the price will increase their willingness to buy them. On the contrary, if customers have no willingness to buy them, once the price is increased, their willingness to buy them will decrease. Pseudo-positive samples are generated by appropriately lowering the price of positive samples, and pseudo-negative samples are generated by appropriately raising the price of negative samples.

[0065] By comparing the correlation coefficient between customers' total preference for sales varieties and price elasticity coefficient , get the price elasticity coefficient of customers for different sales varieties , This coefficient describes the customer's sensitivity to price fluctuations of sales products. The larger its absolute value, the more sensitive the customer is to the price of the sales product, and the greater the fluctuation of his / her purchase intention. This can be used for scenarios such as targeted promotions.

[0066] Figure 5 A flow chart of constructing an order quantity prediction model according to the present invention is shown.

[0067] According to an embodiment of the present invention, an order quantity prediction model is constructed, and the total preference is used to perform model training, specifically:

[0068] S502, constructing an order quantity prediction model through a neural factor decomposition mechanism, presetting a binning standard in the model, counting the historical order quantities of customers, dividing the order quantity range of the first 75% into equal widths, dividing the order quantity range of the first 75% to 95% into equal frequency, and dividing the order quantity range of the last 5% into an independent interval, and numbering the obtained order quantity intervals;

[0069] S504, using the total preferences of the customers to train the model in two stages. The first stage uses the original features consisting of customers, sales varieties, average weight, and age to predict the probability of the target customers ordering quantity intervals for sales varieties with preset specifications. , , are model parameters;

[0070] S506, the second part uses the original features consisting of customers, sales product grades, and attendance context, and adds the price elasticity coefficients of customers for different sales product grades based on the natural order quantity interval number with the largest probability in the first part. , treat the interval number as a continuous value and slide it to get the order interval prediction result , , is the price of the product being sold, It is the average price of all varieties sold by the same poultry project company.

[0071] The model is trained in stages. In the first stage, the customer's order quantity interval for a certain specification (age, average weight) of the sales product is predicted. In the second stage, based on the order quantity interval predicted in the first stage, the order quantity interval is floated by considering factors such as grade and price, and the final order quantity interval is predicted. At the same time, the price elasticity coefficient of the customer to the grade of the sales product is estimated according to the principle of economics. The training samples in the first stage are trained with the purchase records of the sales products of specific specifications by the customer, and the grade and price information corresponding to the sales product is used for training in the second stage. In the first stage of training, in order to improve the prediction performance, this model adopts a double loss function, and the mean square error between the predicted interval ID and the true interval ID is used as regularization; in the second stage, the concept of price elasticity coefficient is added with reference to microeconomics, and the interval number is treated as a continuous value for sliding to obtain the order interval prediction result.

[0072] Set the limit range according to the order quantity interval number range 1 to N , For truncation operations, limit the results to 1 to N;

[0073] Integerize the order quantity interval prediction results and select the closest integer value as the final order quantity interval number , , This is a rounding operation.

[0074] Figure 6 A block diagram of a poultry farming project marketing system based on customer portraits of the present invention is shown.

[0075] The second aspect of the present invention further provides a poultry project marketing system 6 based on customer portraits, including a data input and preprocessing module 61, a customer ordering intention module 62, a customer ordering preference prediction module 63 and a customer ordering quantity prediction module 64;

[0076] The data input and preprocessing module 61 constructs a data set at the granularity of customer-poultry company, and establishes different features through different latitude combinations;

[0077] The customer ordering intention module 62 uses the customer profile features and attendance context features as input to construct a decision tree, obtains the customer's ordering intention, and finally outputs the target customer through the decision strategy to deliver to the downstream task;

[0078] The customer order preference prediction module 63 uses the natural preference and price-related preference of the multi-layer perceptron, and combines the generated total preference to adopt a multi-stage training method for training. In the first stage, the customer's preference attributes for the sales variety itself are characterized by the original features to generate the natural preference. In the second stage, the model parameters of the first stage are fixed, and the total preference related to the price is trained based on the natural preference of the first stage.

[0079] The customer order quantity prediction module 64 divides the order quantity intervals for prediction, performs training based on the customer's total preference, constructs an order quantity prediction model, predicts the customer's order quantity for sales products of different specifications, and predicts the final order quantity interval based on the sales product order quantity, taking into account the grade and price.

[0080] The third aspect of the present invention also provides a computer-readable storage medium, which includes a poultry farming project marketing method program based on customer portrait. When the poultry farming project marketing method program based on customer portrait is executed by a processor, the steps of the poultry farming project marketing method based on customer portrait as described in any one of the above items are implemented.

[0081] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0082] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0083] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0084] Those skilled in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.

[0085] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0086] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A poultry farming project marketing method based on customer portrait, characterized in that: The following steps are involved: Obtain historical interaction data between customers and the poultry project company to build a data set and perform preprocessing. Extract customer portrait features through customers' historical ordering data and extract customer attendance context features through customers' attendance data. Using a machine learning model to predict customer ordering intention based on the customer portrait features and attendance context features, and dividing the customer groups according to the customer ordering intention; Obtain the item representation of the poultry farming project and the customer representation of the customer, use the multi-layer perceptron to obtain the customer's natural preference for sales varieties of different specifications through the item representation and customer representation, and combine the customer's price preference to generate the customer's total preference for sales varieties of the poultry farming project; Construct an order quantity prediction model, use the total preference to train the model, predict the customer's order quantity range for different specifications of sales products through the original features, and estimate the customer's price elasticity coefficient for the sales products to obtain the final order quantity range; The multi-layer perceptron is used to obtain the customer's natural preference for sales varieties of different specifications through item representation and customer representation, and the customer's total preference for sales varieties of poultry projects is generated by combining the customer's price preference, which is as follows: The customer ordering preferences are trained in two stages. In the first stage, a multi-layer perceptron is used to obtain the customer's natural preference for different specifications of sales products through item representation and customer representation. It is expressed as: , For customers, For sales varieties, For average weight, For days, is the characterization function; In the second stage, the model parameters of the first stage are fixed, and the total price-related preference is trained based on the natural preference in the first stage. The correlation coefficient of the price elasticity coefficient is The price-related preference is obtained by combining the natural preference and the price-related preference to obtain the total preference of the customer for the sales varieties of the poultry project. , expressed as: ， in, is the activation function, is the sales price of the product, It is the average price of all varieties sold by the same poultry project company; Dynamic negative sampling is used to train customer ordering preferences. Sales records of other customers are aggregated according to the granularity of date, sales company, and sales product. The sales volume is summed up, and the price and age are weighted averaged according to the sales volume to construct a negative sample pool. According to the number of sales records purchased by the customer on the same day, the sales volume is used as the weight according to the preset ratio to randomly select sales items that have not been purchased from the negative sample pool as negative samples. The positive samples are lowered to generate pseudo positive samples, and the negative samples are raised to generate pseudo negative samples. By comparing the correlation coefficient between customers' total preference for sales varieties and price elasticity coefficient , get the price elasticity coefficient of customers for different sales varieties , .

2. A poultry farming project marketing method based on customer portrait according to claim 1, characterized in that: A machine learning model is used to predict the customer's ordering intention based on the customer portrait features and attendance context features, and customer groups are divided according to the customer's ordering intention, specifically: The LightGBM model is used to build an order intention prediction model. The customer portrait features and attendance context features are used as model inputs. The forward distribution algorithm is used to learn and build the corresponding decision tree through negative gradient fitting residuals in each iteration. Through training with the constructed decision tree, the customer's ordering intention is obtained, and the customer's ordering intention is imported into the decision module. In the decision module, three customer groups, low-frequency, medium-frequency and high-frequency, are divided through data analysis, and target customers are screened according to the customer groups that different customers fall into.

3. A poultry farming project marketing method based on customer portrait according to claim 1, characterized in that: Obtain the item representation of the poultry farming project and the customer representation of the customer, specifically: Obtaining original features of the poultry breeding project, including sales varieties, customers, age in days, and average weight, and using a multi-layer perceptron to generate item representations of the poultry breeding project based on the original features; The customer's interaction frequency with the poultry farming project is extracted based on the historical interaction data in the data set, and the frequency-sensitive vector is obtained using a linear layer. The customer ID, sales area, preference level, and attendance context features corresponding to the customer are read using the historical interaction data, and the customer's explicit representation is obtained through a multi-layer perceptron. The poultry farming items in the historical interaction data are represented by the attention mechanism, the shopping basket information of customers in different time periods is obtained, and the shopping basket information is used to generate the customer's historical interest representation using the long short-term memory network; The frequency-sensitive vector, the customer explicit representation and the customer historical interest representation are used to construct a customer representation.

4. The poultry farming project marketing method based on customer portrait according to claim 1, characterized in that: Construct an order quantity prediction model and use the total preference to train the model, specifically: The order quantity prediction model is constructed through the neural factor decomposition mechanism. The binning standard is preset in the model, and the historical order quantity of customers is counted. The order quantity range of the first 75% is divided into equal widths, and the order quantity range of the first 75% to 95% is divided into equal frequencies. The order quantity range of the last 5% is separated into an independent interval, and the obtained order quantity intervals are numbered; The model training is carried out in two stages using the total preferences of the customers. The first stage uses the original features consisting of customers, sales varieties, average weight, and age to predict the probability of the target customers ordering the preset specifications of the sales varieties. , , are model parameters; The second part uses the original features consisting of customers, sales product grades, and attendance context. Based on the natural order quantity interval number with the largest probability in the first part, the price elasticity coefficient of customers for different sales product grades is added. , treat the interval number as a continuous value and slide it to get the order interval prediction result , , is the price of the product being sold, It is the average price of all varieties sold by the same poultry project company.

5. A poultry farming project marketing method based on customer portrait according to claim 4, characterized in that: Set the limit range according to the order quantity interval number range 1 to N , For truncation operations, limit the results to 1 to N; Integerize the order quantity interval prediction results and select the closest integer value as the final order quantity interval number , , This is a rounding operation.

6. A poultry farming project marketing system based on customer portraits, characterized in that: Implementing the poultry farming project marketing method based on customer portrait as described in claims 1-5, including a data input and preprocessing module, a customer ordering intention module, a customer ordering preference prediction module and a customer ordering quantity prediction module; The data input and preprocessing module constructs a data set at the granularity of customer-poultry company, and establishes different features through different dimension combinations; The customer ordering intention module uses customer profile features and attendance context features as input to construct a decision tree, obtains the customer's ordering intention, and finally outputs the target customer through the decision strategy to deliver to the downstream task; The customer order preference prediction module uses the natural preference and price-related preference of the multi-layer perceptron, and combines it with the generated total preference, and adopts a multi-stage training method for training. In the first stage, the customer's preference attributes for the sales variety itself are characterized by the original features to generate natural preferences. In the second stage, the model parameters of the first stage are fixed, and the total preference related to price is trained based on the natural preference of the first stage. The customer order quantity prediction module divides the order quantity intervals for prediction, performs training based on the customer's total preference, constructs an order quantity prediction model, predicts the customer's order quantity for sales products of different specifications, and predicts the final order quantity interval based on the sales product order quantity, taking into account the grade and price.

Images