Customer group determination method and device, electronic equipment and storage medium

Abstract

Embodiments of the invention provide a customer group determination method and apparatus, an electronic device and a storage medium. The method comprises the steps of obtaining a customer screening demand; performing intention classification on the customer screening demand according to a customer screening model, and determining a user intention; determining a trial space algorithm corresponding to the customer screening demand according to the user intention; key parameters are extracted from the customer screening requirements; the key parameters comprise at least one of a service scene, a target scale, a screening condition and a label combination; and inputting the key parameters into the customer screening model, calling the trial space algorithm through the customer screening model, and determining a target customer group according to the key parameters. Through the client screening model, intention classification is automatically performed on user demands, a user screening target is rapidly and accurately identified, a trial calculation space algorithm is automatically matched according to the user intention, adaptation of the algorithm and a scene is realized, and the efficiency can be improved while manpower can be saved.

Description

Technical Field

[0001] The present invention relates to the technical field of large models, and particularly to a method for determining a customer group, a device for determining a customer group, an electronic device, and a computer-readable storage medium. Background Art

[0002] Today, with the rapid development of digital marketing, consumers are bombarded with a vast amount of advertising and promotional information, leading to the gradual ineffectiveness of traditional marketing methods and users' fatigue with the same old promotion strategies. As the diversification of marketing methods causes users to experience marketing fatigue, it has become particularly important to improve the ability of precision marketing and personalized services. At the same time, enterprises' demand for precision marketing and personalized services is increasing day by day. How to efficiently identify high-value customer groups and formulate targeted strategies has become the key to improving marketing conversion rates.

[0003] Traditional marketing methods mainly rely on the screening of experts' business experience and it is difficult to screen out customers with real purchase intentions from a large number of customers. It requires a lot of manpower and time for trial calculations, and for large-scale data sets or frequent marketing activities, the efficiency is relatively low. Summary of the Invention

[0004] In view of the above problems, embodiments of the present invention are proposed to provide a method for determining a customer group, a device for determining a customer group, an electronic device, and a computer-readable storage medium that overcome the above problems or at least partially solve the above problems.

[0005] To solve the above problems, in a first aspect of embodiments of the present invention, a method for determining a customer group is provided. The method includes:

[0006] Obtain customer screening requirements;

[0007] Classify the intention of the customer screening requirements according to a customer screening model to determine the user intention;

[0008] Determine the corresponding trial calculation space algorithm for the customer screening requirements according to the user intention;

[0009] Extract key parameters from the customer screening requirements; the key parameters include at least one of a business scenario, a target scale, screening conditions, and a label combination;

[0010] Input the key parameters into the customer screening model, and call the trial calculation space algorithm through the customer screening model to determine the target customer group according to the key parameters.

[0011] Optionally, the trial calculation space algorithm includes an optimal trial calculation algorithm for a specified scale; the step of calling the trial calculation space algorithm through the customer screening model to determine the target customer group according to the key parameters includes:

[0012] Obtain customer data and the conversion probability value corresponding to the customer data;

[0013] Sort the customer data in descending order according to the conversion probability value to obtain a sorting result;

[0014] Determine a target customer group according to the sorting result and the target scale.

[0015] Optionally, the trial calculation space algorithm includes a specified condition optimal trial calculation algorithm; the step of calling the trial calculation space algorithm through the customer screening model to determine the target customer group according to the key parameters includes:

[0016] Convert the screening conditions into structured conditional statements;

[0017] Obtain customer data and the conversion probability value corresponding to the customer data;

[0018] Screen the customer data based on the structured conditional statements to obtain a candidate customer group;

[0019] Sort the candidate customer group in descending order according to the conversion probability value to obtain a sorting result;

[0020] Determine a target customer group according to the sorting result and the target scale.

[0021] Optionally, the trial calculation space algorithm includes a label optimal threshold trial calculation algorithm; the step of calling the trial calculation space algorithm through the customer screening model to determine the target customer group according to the key parameters includes:

[0022] Obtain customer data and the conversion probability value corresponding to the customer data;

[0023] Construct a decision tree model based on the label combination and the conversion probability value;

[0024] Perform feature threshold division based on the gain rate of the Gini coefficient, and allocate the customer data divided by the feature threshold to the smallest unit cluster in the decision tree model;

[0025] Sort the customer data in the smallest unit cluster in descending order to obtain a sorting result;

[0026] Screen the customer data according to the sorting result and the target scale to determine the target customer group, and calculate the estimated conversion probability value corresponding to the target customer group; the estimated conversion probability value is the expected probability that the user purchases the recommended product.

[0027] Optionally, the trial space algorithm includes an optimal label combination trial algorithm; the step of determining the target customer group according to the key parameters by invoking the trial space algorithm through the customer screening model includes:

[0028] Extracting customer data related to the business scenario from a preset database;

[0029] Determining a label combination rule and a screening scale according to the business scenario;

[0030] Screening the customer data according to the label combination rule and the screening scale to determine a target customer group, and calculating an estimated conversion probability value corresponding to the target customer group.

[0031] Optionally, the step of obtaining the conversion probability value corresponding to the customer data includes:

[0032] Obtaining marketing voice data;

[0033] Extracting customer data related to the business scenario from a preset database;

[0034] Obtaining, through an integration algorithm, a quantization of the marketing voice data and the customer data to obtain the conversion probability value corresponding to the customer data; the conversion probability value is the probability of a user's purchase intention for a product.

[0035] Optionally, the customer screening model is trained in the following manner:

[0036] Obtaining customer screening requirements, and extracting key parameters from the customer screening requirements; the key parameters include at least one of a business scenario, a target scale, screening conditions, and label combinations;

[0037] Obtaining a marketing voice data set;

[0038] Extracting a customer data set related to the business scenario from a preset database;

[0039] Obtaining, through an integration algorithm, a quantization of the marketing voice data set and the customer data set to obtain a set of conversion probability values;

[0040] Constructing a basic data set according to the set of conversion probability values;

[0041] Inputting the basic data set into a preset large model for model fine-tuning;

[0042] Determining a user intention based on the customer screening requirements through the fine-tuned preset large model, and determining the corresponding trial space algorithm for the customer screening requirements according to the user intention;

[0043] Input the key parameters into the fine-tuned pre-set large model, call the trial calculation space algorithm to train the pre-set large model, and obtain the customer screening model.

[0044] Optionally, constructing the basic data set according to the conversion probability value set includes:

[0045] Convert the marketing voice data into text data and preprocess the text data;

[0046] Configure a set of user pain point tags for the preprocessed text data;

[0047] Analyze the relationship between the conversion probability value, the user pain point tag and the product by constructing a Bayesian network;

[0048] Construct a basic data set according to the analysis result of the Bayesian network.

[0049] Optionally, inputting the basic data set into a pre-set large model for model fine-tuning includes:

[0050] Construct a query matrix, a key matrix, a value matrix and a weight matrix;

[0051] Decompose the query matrix, the key matrix, the value matrix and the weight matrix into at least one low-rank matrix;

[0052] Convert the elements in the low-rank matrix to generate a new weight matrix;

[0053] Overlay the new weight matrix on the low-rank matrix to determine the fine-tuned weight;

[0054] Fine-tune the pre-set large model based on the fine-tuned weight.

[0055] In the second aspect of the embodiments of the present invention, a customer group determination device is provided, characterized in that the device includes:

[0056] A demand acquisition module, configured to acquire customer screening demands;

[0057] An intention determination module, configured to classify the intention of the customer screening demand according to the customer screening model to determine the user intention;

[0058] An algorithm determination module, configured to determine the corresponding trial calculation space algorithm for the customer screening demand according to the user intention;

[0059] A parameter extraction module, configured to extract key parameters from the customer screening demand; the key parameters include at least one of a business scenario, a target scale, screening conditions, and a tag combination;

[0060] A customer determination module is configured to input the key parameters into the customer screening model, and call the trial calculation space algorithm according to the key parameters through the customer screening model to determine the target customer group.

[0061] According to a third aspect of the present invention, there is provided an electronic device, comprising: a processor, a memory, and a computer program stored on the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of a customer group determination method as described above are implemented.

[0062] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of a customer group determination method as described above are implemented.

[0063] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0064] Embodiments of the present invention provide a method, apparatus, electronic device, and storage medium for determining a customer group. The method includes: obtaining customer screening requirements; classifying the intent of the customer screening requirements according to a customer screening model to determine the user intent; determining a corresponding trial calculation space algorithm for the customer screening requirements according to the user intent; extracting key parameters from the customer screening requirements, where the key parameters include at least one of a business scenario, a target scale, screening conditions, and a tag combination; inputting the key parameters into the customer screening model, and calling the trial calculation space algorithm according to the key parameters through the customer screening model to determine the target customer group. By automatically classifying the intent of the user requirements through the customer screening model, quickly and accurately identifying the user screening target, automatically matching the trial calculation space algorithm according to the user intent, realizing the adaptation of the algorithm to the scenario, being able to adapt to the business screening of various business scenarios, and at the same time saving manpower consumption and screening out the target customer group faster to improve efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is a flowchart of the steps of a method for determining a customer group provided by an embodiment of the present invention;

[0066] Figure 2 is a schematic flowchart of determining a customer group by trial calculation of the optimal threshold of tags in a method for determining a customer group provided by an embodiment of the present invention;

[0067] Figure 3 is a flowchart of the training steps of a customer screening model in a method for determining a customer group provided by an embodiment of the present invention;

[0068] Figure 4 is a schematic flowchart of model fine-tuning in a method for determining a customer group provided by an embodiment of the present invention;

[0069] Figure 5 This is a structural block diagram of a customer group determination device provided by an embodiment of the present invention. Specific Embodiments

[0070] To make the above objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0071] Traditional marketing methods mainly rely on the screening of experts' business experience, making it difficult to screen out truly potential customers from a large number of customers. It requires a lot of manpower and time for trial calculations. For large-scale data sets or frequent marketing activities, the efficiency is relatively low. One of the core concepts of the embodiments of the present invention is to obtain customer screening requirements, determine user intentions, and use a customer screening model to call a trial calculation space algorithm to determine the target customer group according to key parameters. It can automatically classify user intentions, adapt to business screening in various business scenarios, and at the same time save manpower consumption and quickly screen out the target customer group to improve efficiency.

[0072] Refer to Figure 1 , which shows a flowchart of the steps of a customer group determination method provided by an embodiment of the present invention. The method may specifically include the following steps:

[0073] The customer group determination method of the embodiments of the present invention can be widely applied to industry scenarios that require accurate customer grouping, especially in data-intensive fields. It is applicable to large customer scales and changing business scenarios, and can shorten the traditional customer analysis that takes several weeks to the hour level, improving the efficiency of customer group determination.

[0074] Step 101, obtain customer screening requirements;

[0075] Customer screening requirements are screening criteria formulated to more accurately identify, classify, and select target customer groups. Text data or voice data is input by the user. In this embodiment, the user inputs the required customer group. For example, the user inputs "Help me screen a batch of target users for upgrading to gigabit", etc. The customer screening model determines the screening rules and scale according to the content input by the user.

[0076] Step 102, classify the customer screening requirements according to the customer screening model to determine the user intention;

[0077] The customer screening model is a systematic method for stratifying, scoring, and prioritizing potential customers through data analysis and algorithms, and can be obtained through machine learning models or large model training. Intention classification is to identify the true needs and purchase intentions of customers by analyzing customer behavior, language, and context. The user intention can be determined according to the intention classification results.

[0078] To classify the intent of user requirements according to the customer screening model, it is necessary to obtain the text data of customer screening requirements, label the intent for the text data, and determine the user intent through the customer screening model to screen out 20,000 target users for handling video memberships based on probability.

[0079] Step 103: Determine the corresponding trial space algorithm for the customer screening requirements according to the user intent.

[0080] The trial space algorithm (Trial Space Algorithm) is usually used to solve optimization, search, or parameter estimation problems. Its core idea is to generate, evaluate, and iterate candidate solutions within a predefined "trial space" to gradually approach the optimal solution. In the embodiments of the present invention, the trial space algorithm includes the specified scale optimal trial algorithm, the specified condition optimal trial algorithm, the label optimal threshold trial algorithm, and the label optimal combination trial algorithm. The specified scale optimal trial algorithm screens the target customer group according to the probability requirement; the specified condition optimal trial algorithm outputs the target customer group according to the label rule; the label optimal threshold trial algorithm screens the target customer group according to the label trial optimal rule and the scale requirement; the label optimal combination trial algorithm allows the user not to specify the label and scale, but only to input the required scenario, and the model automatically selects the label and tries to calculate the optimal rule and scale.

[0081] Classify the intent of the customer screening requirements according to the customer screening model to determine the user intent, and thus determine the corresponding trial space algorithm for the customer screening requirements. For example: when the user inputs the customer screening requirement "Output high-probability users with a scale of 20,000 for handling video memberships according to probability", and through the customer screening model, it is determined that the user intent is to screen out 20,000 target users for handling video memberships based on probability, then the corresponding trial space algorithm for the customer screening requirements of the customer screening model is the specified scale optimal trial algorithm.

[0082] Step 104: Extract key parameters from the customer screening requirements; the key parameters include at least one of business scenario, target scale, screening condition, and tag combination.

[0083] Business Scenario refers to a specific situation or problem, usually involving specific goals, processes, participants, and constraints. Target Scale is the target scale of the customer group to be actually screened, usually measured by quantifiable indicators. Filter Criteria refers to the limiting rules or parameters used to narrow down the scope and accurately locate the target object from a large amount of data or options according to specific rules. Tag Combination refers to associating multiple feature tags (Tags) of users / products / content through logical rules to form a more accurate screening or grouping strategy for refined operation, personalized recommendation, or data analysis.

[0084] According to the text data of the customer screening requirements input by the user, extract the features of the text data corresponding to the customer screening requirements. According to four core dimensions of business scenario, target scale, screening conditions, and label combination, extract the key parameters in the text data, so that the trial space algorithm determines the target customer group according to the key parameters.

[0085] Step 105, input the key parameters into the customer screening model, and call the trial space algorithm through the customer screening model to determine the target customer group according to the key parameters.

[0086] The customer screening model extracts structured key parameters from the text data of the customer screening requirements input by the user. After being verified and supplemented by the customer screening model, it calls the adapted trial space algorithm according to the parameter type, generates candidate solutions in the defined parameter space and evaluates the objective function, and determines the optimal parameter combination through iterative optimization, and finally outputs the target customer group.

[0087] In some embodiments, the trial space algorithm includes the specified scale optimal trial algorithm; the specified scale optimal trial algorithm (Optimal Scaling Trial Algorithm) is a calculation method that, under a preset scale constraint (such as the target customer scale), searches for the best parameter combination or decision-making plan through intelligent trial and iterative optimization. The step 105 includes the following sub-steps:

[0088] Sub-step S11, obtain customer data and the conversion probability value corresponding to the customer data;

[0089] The conversion probability value refers to the estimated possibility that a user completes the target behavior in a specific business scenario. The customer data is list data containing user portrait labels, and the list data includes the static information of the user (such as age, gender, geographical location, historical behavior, etc.) and the labeled user portrait (such as interest preferences, consumption habits, etc.).

[0090] The customer data is stored in the database of the server, and the customer data can be directly obtained from the database. The conversion probability value corresponding to the customer data is obtained and quantified through an integrated algorithm model. First, parse the marketing call recording through an automatic speech recognition algorithm to convert the speech into structured text data; then associate the user portrait label list data to construct a multi-dimensional feature matrix; finally, use the integrated algorithm model to quantify the user conversion probability and output an analysis result including the probability value and the contribution degree of key features.

[0091] Sub-step S12, perform a descending order sorting on the customer data according to the conversion probability value to obtain a sorting result;

[0092] Based on the user conversion probability values output by the integrated algorithm model, the customer data obtained from the database is sorted in descending order according to the probability values, generating a list of high-potential customers for priority outreach, providing a basis for the subsequent precise marketing strategy to execute according to priority.

[0093] Sub-step S13, determining the target customer group according to the sorting result and the target scale.

[0094] Based on the customer list sorted in descending order, combined with a preset target scale (such as selecting the top 20% or a fixed number threshold), by setting a number threshold or directly intercepting a certain number of customers, the final target customer group is determined to ensure that the screening result meets both the requirements of high conversion probability and business volume.

[0095] In the embodiment of the present invention, the specified-scale optimal trial calculation algorithm screens the target customer group required according to the probability. By the user inputting the customer screening requirement "screen 20,000 gigabit target user groups according to the probability". For example, the user inputs: "Screen 20,000 gigabit target user groups according to the probability." The customer screening model analyzes the user input and determines that the user's intention is to screen out 20,000 target customers who may upgrade to gigabit broadband based on the probability. The customer screening model extracts the following key parameters from the customer screening requirements input by the user: {"scenario": "upgrade to gigabit broadband", "required_size": 20000}. The customer screening model inputs the key parameters into the trial calculation space algorithm, and the trial calculation space algorithm processes and screens the target customer group. Through the above steps, after the customer screening model understands the user's intention, it transmits the key information to the trial calculation space algorithm, and the algorithm screens and outputs the 20,000 target customer groups with the highest probability according to the predicted probability of the user handling gigabit broadband.

[0096] In some embodiments, the trial calculation space algorithm includes a specified-condition optimal trial calculation algorithm; the specified-condition optimal trial calculation algorithm is an algorithm for finding the optimal solution under given constraint conditions. The step 105 further includes the following sub-steps:

[0097] Sub-step S21, converting the screening conditions into structured conditional statements;

[0098] The structured conditional statement is a decision-making system constructed by combining logical rules and hierarchical judgments, used to realize the automatic judgment of complex business scenarios. The screening conditions described in natural language are converted into machine-executable structured conditional statements through a rule engine or semantic parsing, realizing a lossless mapping from business requirements to code logic, and supporting automatic query and dynamic parameter injection.

[0099] Sub-step S22, obtaining the customer data and the corresponding conversion probability values of the customer data;

[0100] Customer data is stored in the server's database, and customer data can be directly retrieved from the database. The conversion probability value corresponding to the customer data is obtained and quantified through an integrated algorithm model. First, the marketing call recordings are parsed by an automatic speech recognition algorithm to convert the speech into structured text data. Finally, the integrated algorithm model is used to quantify the user conversion probability, and an analysis result including the probability value and the contribution degree of key features is output.

[0101] Sub-step S23, screening the customer data based on the structured conditional statement to obtain a candidate customer group;

[0102] Based on a structured conditional statement (such as the WHERE clause of SQL or the query condition of DataFrame), a candidate customer group that exactly matches the conditions is screened from the customer database to ensure that the data precisely matches the business rules, while supporting batch query optimization and real-time dynamic filtering, and finally an objective customer set that meets all constraint conditions is output.

[0103] Sub-step S24, sorting the candidate customer group in descending order according to the conversion probability value to obtain a sorting result;

[0104] Based on the user conversion probability value output by the integrated algorithm model, the customer data obtained from the database is sorted in descending order according to the probability value, and a customer list sorted in descending order of conversion potential is generated.

[0105] Sub-step S25, determining the target customer group based on the sorting result and the target scale.

[0106] According to the sorted candidate customer list, combined with a preset target scale (such as the top 1000 people or the top 20% percentile threshold), the final target customer group is accurately locked through threshold truncation or dynamic ratio allocation.

[0107] In the embodiment of the present invention, the specified condition optimal trial calculation algorithm outputs the target customer group according to the label rule. For example, the user inputs the customer screening requirement "want to screen 20,000 gigabit target user groups in xx cities according to the probability". For example, the user inputs: "Screen 20,000 gigabit target user groups in the city according to the probability." The customer screening model first analyzes the sentence input by the user and understands that the user's intention is to screen out 20,000 high-probability users with the potential to upgrade to gigabit broadband in the city. The customer screening model extracts the key parameters from the user input: {"city": "city", "user_scale": 20000, "business_demand": "upgrade to gigabit broadband"}, and the customer screening model inputs the key parameters into the trial calculation space algorithm. The algorithm executes the SQL statement to screen out the user group that meets the conditions from the full amount of user data. Sort in descending order according to the predicted probability of the user's upgrade to gigabit broadband. Automatically output the top 20,000 users with the highest probability and calculate the estimated conversion rate.

[0108] Refer to Figure 2 , which shows a schematic flowchart of determining a customer group by trial calculation of the optimal threshold of tags for a customer group determination method provided by an embodiment of the present invention;

[0109] In some embodiments, the trial calculation space algorithm includes a tag optimal threshold trial calculation algorithm; the tag optimal threshold trial calculation algorithm is a calculation method for determining the best classification threshold, according to the tag trial calculation optimal rules and scale. The step 105 further includes the following sub-steps:

[0110] Sub-step S31, obtaining customer data and the conversion probability value corresponding to the customer data;

[0111] The customer data is stored in the database of the server, and the customer data can be directly obtained from the database. The conversion probability value corresponding to the customer data is obtained and quantified through an integrated algorithm model. First, the marketing call recording is parsed by an automatic speech recognition algorithm to convert the speech into structured text data; finally, the integrated algorithm model is used to quantify the user conversion probability, and an analysis result including the probability value and the key feature contribution degree is output.

[0112] Sub-step S32, constructing a decision tree model based on the tag combination and the conversion probability value;

[0113] A decision tree is a supervised learning algorithm based on a tree structure, which can be used for classification and regression tasks. Using the user tag combination as the feature variable and the conversion probability value as the target variable, a prediction model is constructed through the decision tree algorithm, automatically generating interpretable rule branches, quantifying the feature importance and outputting a visual tree structure to train the tree model.

[0114] Sub-step S33, performing feature threshold division based on the gain ratio of the Gini coefficient, and allocating the customer data divided by the feature threshold to the smallest unit cluster in the decision tree model;

[0115] The Gini coefficient is a key indicator in the decision tree algorithm for measuring data impurity, and is widely used especially in the classification and regression tree algorithms. The gain ratio is an important indicator for feature selection in the decision tree algorithm, used to improve the preference problem of information gain for multi-valued features. Feature threshold division is a core step in decision trees and rule learning, used to determine the best split point for continuous features or ordered discrete features. The smallest unit cluster is a refined data grouping method, aiming to divide the data set into the smallest-scale sub-groups that meet specific constraint conditions.

[0116] The information gain rate of each feature is calculated based on the Gini coefficient, and the optimal partitioning threshold is determined by maximizing the gain rate. The customer data is recursively divided into the smallest leaf node unit of the decision tree to form homogeneous clusters, that is, the customer data divided by the feature threshold is assigned to the smallest unit cluster in the decision tree model.

[0117] Sub-step S34, sorting the customer data in the minimum unit cluster in descending order to obtain a sorting result;

[0118] For the smallest unit cluster generated by the leaf node of the decision tree model, the conversion probability values of the customers in the cluster are used as the sorting basis, and they are arranged in descending order to ensure that the customers in each segment group are strictly sorted according to their conversion potential, providing a precise priority execution sequence for subsequent differentiated marketing strategies.

[0119] Sub-step S35, screening the customer data based on the sorting result and the target scale to determine the target customer group, and calculating the estimated conversion probability value corresponding to the target customer group; the estimated conversion probability value is the expected probability of the user purchasing the recommended product.

[0120] Based on the sorted customer data and the preset target scale, the final target customer group is screened out through dynamic interception or quantile threshold, and the average estimated conversion probability value of the group is aggregated and calculated, and the target customer group and the estimated conversion probability value are output at the same time.

[0121] The embodiment of the present invention is a label optimal threshold trial calculation algorithm. According to the label trial optimal rules and scale, the user inputs the customer screening demand "Filter 20,000 gigabit target users based on network access time, user consumption and broadband traffic." For example, the user inputs: "Filter 20,000 gigabit target users based on network access time, user consumption and broadband traffic." The customer screening model analyzes the user input and determines that the user's intention is to filter out 20,000 target customers who may be upgraded to gigabit broadband based on the three labels of network access time, user consumption and broadband traffic. The customer screening model extracts the following key parameters from the user input and inputs the key parameters into the trial calculation space algorithm:

[0122] The trial space algorithm divides customer data into virtual positive and negative samples, uses the labels involved in the key parameters of the customer screening model input as features, and uses the positive and negative samples as the classification model based on the conversion probability value to train the tree model. The minimum unit cluster rule with the smallest error in the specified group size is taken out. The ratio of positive and negative samples in the minimum unit cluster in the analytical model is the ratio of the statistics of this minimum unit cluster. The higher the ratio, the higher the conversion rate is, and the output scale is more consistent with the input scale.

[0123] Filter out the smallest unit clusters that meet the quantity, perform feature threshold division based on the gain rate of the Gini coefficient. After traversing the features and dividing, allocate all samples to different smallest unit clusters:

[0124] Gini coefficient:

[0125] p i Represents the probability or proportion of the i-th category. It is the proportion of the i-th category in a certain dataset or subset.

[0126] Conditional Gini coefficient:

[0127]

[0128] Among them, the penalty coefficient λ is added i , depth is the depth of the tree, A is the dataset of the current node, that is, all samples of the current node in the decision tree, and B i is the subset obtained by dividing the dataset A according to a certain feature or condition.

[0129] Gini coefficient gain:

[0130] Gini(A,B) = Gini(A) - Gini(A|B)

[0131] Gini coefficient gain rate:

[0132]

[0133] Take out the customer group with the highest probability among the divided smallest unit clusters. The ratio of the smallest unit cluster and the positive and negative samples inside is the ratio counted by this smallest unit cluster. The higher the ratio, the higher the conversion rate closer to this rule. However, the one with the highest ratio is not necessarily the best leaf node sought. This node needs to meet: the highest conversion ratio, the largest number of conversions, and the fewest splitting times.

[0134] Through p ration 、p num 、p split Quantify the above three conditions, and take the weight vector β = [α1, α2, α2], and ∑ i α i = 1, which represent the weight sizes of the three conditions respectively, and can be dynamically adjusted according to the actual scenario. Take Purpose con as the final evaluation criterion:

[0135]

[0136] Among them, the conversion rate proportion score is represented by p ration :

[0137]

[0138] num(leaf_all): The total number of samples in this minimum unit cluster. n refers to the number of positive samples in the minimum unit cluster (leaf node).

[0139] Among them, the conversion quantity score p num represents:

[0140]

[0141] num(all_sample): The total number of samples of this model, num(exp_sample): The expected scale quantity. Among them, the split times score p split represents:

[0142] p split (x) = α3(1 - log max_dept x)

[0143] max_dept: The maximum depth. Finally, obtain the Purpose con The minimum unit cluster with the maximum value is the optimal minimum unit cluster.

[0144] The trial calculation space algorithm obtains the optimal minimum unit group, calculates the optimal screening rule, screens out 20,000 target customer groups, and calculates the estimated conversion rate at the same time.

[0145] In some embodiments, the trial calculation space algorithm includes a label optimal combination trial calculation algorithm; the label optimal combination trial calculation algorithm is an optimization method for determining the optimal label combination. Users do not need to specify labels and scales, but only need to input the required scenarios. The step 105 further includes the following sub-steps:

[0146] Sub-step S41, extracting customer data related to the business scenario from a preset database;

[0147] The customer data is stored in the database of the server, and the customer data can be directly obtained from the database. By extracting customer data related to the business scenario from the preset database.

[0148] Sub-step S42, determining the label combination rule and the screening scale according to the business scenario;

[0149] Based on the business scenario requirements, clarify the key label combination rules and set the target screening scale, and generate executable screening conditions through dynamic rules.

[0150] Sub-step S43, screening the customer data according to the label combination rule and the screening scale to determine the target customer group, and calculating the estimated conversion probability value corresponding to the target customer group.

[0151] Based on the label combination rules and screening scale determined by the business scenario, accurately screen the target customer group from the customer data through structured queries, and calculate the average estimated conversion probability value of this group.

[0152] The embodiment of the present invention is an algorithm for trial calculation of the optimal label combination. Users do not need to specify labels and scales. They only need to input the required business scenario, and the model automatically selects labels and tries to calculate the optimal rules and scales. For example, when a user inputs "Help me screen a batch of target users for gigabit upgrade", after the customer screening model understands the intention, it inputs {scenario} to the trial calculation space algorithm. The algorithm conducts trial calculations based on the full amount of users in the scenario and all portrait labels, and returns {the number of customer groups, screening label rules, estimated conversion rate} after calculating the optimal customer group, and gives the result to the large model as the output answer. Such as "According to the trial calculation results, users with a consumption between 129 and 169, monthly traffic usage greater than 36G, and currently in a gigabit community but not using gigabit broadband have a relatively high probability of gigabit upgrade, with a scale of approximately 40,000 and an estimated conversion rate of 10.6%".

[0153] In some embodiments, any one of the sub-steps S11, sub-step S22, and sub-step S31 includes the following sub-steps: including:

[0154] Sub-step S111, obtain marketing voice data;

[0155] Select the voice data channel according to the business scenario. Common sources include call recordings in the customer service system, user voice inputs in the self-service menu, AI outbound call systems, etc., and collect marketing voice data from channels such as customer service calls and sales product recordings.

[0156] Sub-step S112, extract customer data related to the business scenario from the preset database;

[0157] The customer data is stored in the database of the server, and the customer data can be directly obtained from the database. Based on the requirements of the business scenario, extract associated customer data from the preset database.

[0158] Sub-step S113, obtain the conversion probability value corresponding to the customer data by quantifying the marketing voice data and the customer data through an integrated algorithm; the conversion probability value is the probability of the user's purchase intention for the product.

[0159] Based on the marketing voice data, convert it into semantic feature data after text conversion through an automatic speech recognition algorithm, and combine it with the customer data. Through the integrated algorithm model, conduct multi-modal feature joint training and quantitatively output the conversion probability value (i.e., the purchase intention probability) of each customer.

[0160] The result output by the customer screening model in the embodiment of the present invention is a JSON-formatted data, which contains the detailed information of the optimal customer group. After obtaining the result of the trial calculation space algorithm, the customer screening model converts the field information into a natural language description through the Data-to-Text ability, and at the same time combines the information of the knowledge graph and the pain point labels to clearly convey the characteristics and advantages of the optimal customer group to the marketers.

[0161] Refer to Figure 3 , which shows the flowchart of the training steps of the customer screening model of a customer group determination method provided by the embodiment of the present invention;

[0162] In some embodiments, the customer screening model is trained in the following manner:

[0163] Step 201, obtain the customer screening requirements and extract key parameters from the customer screening requirements; the key parameters include at least one of business scenarios, target scale, screening conditions, and label combinations;

[0164] Obtaining the customer screening requirements is that the user inputs text data or voice data, and the customer screening requirements are parsed through natural language processing technology to extract structured key parameters, including business scenarios, target scale, screening conditions, and label combinations.

[0165] Step 202, obtain the marketing voice data set;

[0166] Select a data source according to the business scenario and collect multiple groups of marketing voice data from channels such as customer service calls and sales recordings to construct a marketing voice data set.

[0167] Step 203, extract the customer data set related to the business scenario in the preset database;

[0168] The customer data is stored in the database of the server, and the customer data can be directly obtained from the database. Based on the business scenario requirements, multiple groups of customer data associated with the business scenario are extracted from the preset database, and a customer data set is constructed.

[0169] Step 204, obtain a quantization conversion probability value set for the marketing voice data set and the customer data set through an integrated algorithm;

[0170] Based on the marketing voice data set, it is converted into a semantic feature data set after text through an automatic speech recognition algorithm, and combined with the customer data set, and multi-modal feature joint training is performed through an integrated algorithm model to quantitatively output the conversion probability value of each marketing voice data and customer data, and each conversion probability value is combined into a conversion probability value set.

[0171] Step 205, construct a basic data set according to the conversion probability value set;

[0172] Based on the customer conversion probability value set, associate the original customer data and business tags, construct a structured basic data set through data fusion, and finally output a standardized table-based data set containing probability values, feature variables, and business tags, providing high-quality input for subsequent analysis and model iteration.

[0173] In some embodiments, the step 205 further includes the following sub-steps:

[0174] Sub-step S51, convert the marketing voice data into text data and preprocess the text data;

[0175] Collect marketing voice data from channels such as customer service calls and sales product recordings. Use the automatic speech recognition algorithm to convert the voice data into text data, and perform data cleaning on the text data to ensure the clarity and accuracy of the voice data, remove noise, silence, repetitions, and meaningless segments, unify the format of the cleaned data, and perform grammar correction and typo correction to convert voice data in different formats and encodings into a unified format.

[0176] Sub-step S52, configure a set of user pain point tags for the preprocessed text data;

[0177] User pain point tags include insufficient traffic, network lag, and too expensive voice call charges, etc. Based on the preprocessed text data, a preset large model automatically labels user pain point tags through rules, and combines manual sampling verification to ensure the accuracy of the tags. Finally, an enhanced data set containing structured pain point tags, confidence levels, and sources is output to support product optimization and customer experience analysis.

[0178] Sub-step S53, analyze the relationship between the conversion probability value, the user pain point tags, and the product by constructing a Bayesian network;

[0179] Based on the conversion probability value, user pain point tags, and product feature data, construct a Bayesian network model. By quantifying the conditional probability dependence relationship between variables, identify key influencing factors, and finally output a visual probability inference result to analyze the relationship between the conversion probability value, user pain point tags, and the product, providing causal insights for precision marketing and product iteration.

[0180] Sub-step S54, construct a basic data set according to the Bayesian network analysis result.

[0181] Combined with the conditional probability table, node dependency relationship, and sensitivity index output by the Bayesian network analysis, associate and map them with the original user conversion data, pain point labels, and product features to construct a structured dataset containing probability inference results and business characteristics to support subsequent causal analysis and strategy optimization. The process of constructing the Bayesian network: By collecting the dataset, collect user characteristics, user pain point labels, and products with conversion probability values as three labels; construct the Bayesian network, add nodes, and learn the network structure from the dataset based on BIC; parameter learning, estimate the conditional probability between each node based on Bayesian estimation; Bayesian inference can analyze the relationship between user characteristics, pain points, and recommended products through the inference of the Bayesian network structure.

[0182] Step 206, input the basic dataset into a preset large model for model fine-tuning;

[0183] Based on the constructed basic dataset, use domain adaptation technology to perform supervised fine-tuning on the preset large model. Through the fine-tuned weights, the model learns the business-specific causal inference mode, and finally outputs a fine-tuned model with scenario capabilities, which can accurately predict user conversion behavior or generate targeted product improvement suggestions, and at the same time reduce the hallucination risk through adversarial training.

[0184] Refer to Figure 4 , which shows a schematic flowchart of model fine-tuning for a customer group determination method provided by an embodiment of the present invention;

[0185] In some embodiments, step 206 further includes the following sub-steps:

[0186] Sub-step S61, construct a query matrix, a key matrix, a value matrix, and a weight matrix;

[0187] Sub-step S62, decompose the query matrix, the key matrix, the value matrix, and the weight matrix into at least one low-rank matrix;

[0188] In a preset large model, construct a query (Q) matrix, a key (K) matrix, and a value (V) matrix in the self-attention mechanism, and decompose these weight matrices in the feed-forward neural network into two low-rank matrices A and B.

[0189] Sub-step S63, perform conversion on the elements in the low-rank matrix to generate a new weight matrix;

[0190] Each element of these two low-rank matrices A and B will be converted through a function to calculate the product of A and B to generate a new weight matrix, which is a low-rank approximation of the original weight matrix.

[0191] Sub-step S64, superimpose the new weight matrix on the low-rank matrix to determine the fine-tuned weight;

[0192] The newly generated low-rank matrix is superimposed on the original weight matrix to form the fine-tuned weights. By superimposing the new weight matrix on the original low-rank matrix, a fine-tuned weight matrix is generated, achieving efficient domain adaptation while keeping the main parameters of the model unchanged.

[0193] Sub-step S65, fine-tuning the preset large model based on the fine-tuned weights.

[0194] Based on the fine-tuned weights, the preset large model is lightly fine-tuned using the gradient backpropagation algorithm. By freezing the backbone parameters and only updating the low-rank matrix or the adaptation layer, while retaining the general capabilities of the original model, the model is adapted to specific business scenarios, and the performance of the fine-tuned model on specific tasks is evaluated and compared with the original model to determine the improvements brought by the fine-tuning.

[0195] Step 207, determining the user intention through the fine-tuned preset large model based on the customer screening requirements, and determining the corresponding trial calculation space algorithm for the customer screening requirements according to the user intention;

[0196] Based on the customer screening requirements, the fine-tuned large model is used to analyze the user intention, automatically match the trial calculation space algorithm according to the intention type, and drive the algorithm iteration through the objective function, and finally output the optimal customer group and the corresponding screening rules.

[0197] Step 208, inputting the key parameters into the fine-tuned preset large model, calling the trial calculation space algorithm to train the preset large model, and obtaining the customer screening model.

[0198] The key parameters in the customer screening requirements are input into the fine-tuned large model, which intelligently calls the adapted trial calculation space algorithm to dynamically construct a parameter search strategy, generates a high-precision customer screening model through an automated training process, and finally outputs a deployment-ready model with both business interpretability and prediction performance, supporting real-time determination of target customer segmentation and priority ranking.

[0199] An embodiment of the present invention provides a customer group determination method, which includes: obtaining customer screening requirements; classifying the intent of the customer screening requirements according to a customer screening model to determine the user intent; determining a corresponding trial space algorithm for the customer screening requirements according to the user intent; extracting key parameters from the customer screening requirements; the key parameters include at least one of a business scenario, a target scale, screening conditions, and a label combination; inputting the key parameters into the customer screening model, and calling the trial space algorithm through the customer screening model to determine a target customer group according to the key parameters. By automatically classifying the user requirements through the customer screening model, quickly and accurately identifying the user screening target, automatically matching the trial space algorithm according to the user intent, realizing the adaptation of the algorithm to the scenario, being able to adapt to the business screening of various business scenarios, and at the same time saving labor consumption, and quickly screening out the target customer group to improve efficiency.

[0200] It should be noted that for the method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the embodiments of the present invention are not limited by the described action sequences, because according to the embodiments of the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential for the embodiments of the present invention.

[0201] Refer to Figure 5 , which shows a structural block diagram of a customer group determination device provided by an embodiment of the present invention, and specifically may include the following modules:

[0202] A requirement acquisition module 301, configured to obtain customer screening requirements;

[0203] An intent determination module 302, configured to classify the intent of the customer screening requirements according to the customer screening model to determine the user intent;

[0204] An algorithm determination module 303, configured to determine a corresponding trial space algorithm for the customer screening requirements according to the user intent;

[0205] A parameter extraction module 304, configured to extract key parameters from the customer screening requirements; the key parameters include at least one of a business scenario, a target scale, screening conditions, and a label combination;

[0206] A customer determination module 305, configured to input the key parameters into the customer screening model, and call the trial space algorithm through the customer screening model to determine a target customer group according to the key parameters.

[0207] The trial space algorithm includes a specified scale optimal trial algorithm; the customer determination module 305 includes:

[0208] A data acquisition sub-module for acquiring customer data and the corresponding conversion probability values of the customer data; a first sorting result determination sub-module for sorting the customer data in descending order according to the conversion probability values to obtain a sorting result; a first customer group determination sub-module for determining a target customer group according to the sorting result and the target scale.

[0209] The trial space algorithm includes a specified condition optimal trial algorithm; the customer determination module 305 includes:

[0210] A condition conversion sub-module for converting the screening conditions into structured conditional statements; a data acquisition sub-module for acquiring customer data and the corresponding conversion probability values of the customer data; a condition screening sub-module for screening the customer data based on the structured conditional statements to obtain a candidate customer group; a second sorting result determination sub-module for sorting the candidate customer group in descending order according to the conversion probability values to obtain a sorting result; a second customer group determination sub-module for determining a target customer group based on the sorting result and the target scale.

[0211] The trial space algorithm includes a label optimal threshold trial algorithm; the customer determination module 305 includes:

[0212] A data acquisition sub-module for acquiring customer data and the corresponding conversion probability values of the customer data; a model construction sub-module for constructing a decision tree model based on the label combination and the conversion probability values; a data allocation sub-module for performing feature threshold division based on the gain rate of the Gini coefficient and allocating the customer data obtained through the feature threshold division to the smallest unit clusters in the decision tree model; a third sorting result determination sub-module for sorting the customer data in the smallest unit clusters in descending order to obtain a sorting result; a third customer group determination sub-module for screening the customer data based on the sorting result and the target scale to determine a target customer group and calculating the estimated conversion probability value corresponding to the target customer group; the estimated conversion probability value is the expected probability of the user purchasing the recommended product.

[0213] The trial space algorithm includes a label optimal combination trial algorithm; the customer determination module 305 includes:

[0214] A data extraction sub-module for extracting customer data related to the business scenario from a preset database; a rule determination sub-module for determining a label combination rule and a screening scale according to the business scenario; a fourth customer group determination sub-module for screening the customer data according to the label combination rule and the screening scale to determine a target customer group and calculating the estimated conversion probability value corresponding to the target customer group.

[0215] The data acquisition sub-module includes: a voice data acquisition unit for acquiring marketing voice data; a customer data extraction unit for extracting customer data related to the business scenario from a preset database; a probability value calculation unit for obtaining a conversion probability value corresponding to the customer data by quantifying the marketing voice data and the customer data through an integrated algorithm; the conversion probability value being the probability of a user's purchase intention for a product.

[0216] The customer screening model is trained in the following manner:

[0217] A screening requirement acquisition module for acquiring customer screening requirements and extracting key parameters from the customer screening requirements; the key parameters including at least one of a business scenario, a target scale, screening conditions, and a label combination.

[0218] A marketing voice data set acquisition module for acquiring a marketing voice data set.

[0219] A customer data set acquisition module for extracting a customer data set related to the business scenario from a preset database.

[0220] A probability value set determination module for obtaining a conversion probability value set by quantifying the marketing voice data set and the customer data set through an integrated algorithm.

[0221] A basic data set construction module for constructing a basic data set according to the conversion probability value set.

[0222] A model fine-tuning module for inputting the basic data set into a preset large model for model fine-tuning.

[0223] A user intention determination module for determining a user intention based on the customer screening requirements through the fine-tuned preset large model, and determining a trial calculation space algorithm corresponding to the customer screening requirements according to the user intention.

[0224] A model training module for inputting the key parameters into the fine-tuned preset large model, calling the trial calculation space algorithm to train the preset large model, and obtaining the customer screening model.

[0225] The basic data set construction module includes: a data conversion sub-module for converting the marketing voice data into text data and preprocessing the text data; a label configuration sub-module for configuring a user pain point label set for the preprocessed text data; a network construction sub-module for analyzing the relationship between the conversion probability value, the user pain point label, and the product by constructing a Bayesian network; and a result analysis sub-module for constructing a basic data set according to the Bayesian network analysis result.

[0226] The model fine-tuning module includes: a matrix construction sub-module for constructing a query matrix, a key matrix, a value matrix, and a weight matrix; a matrix factorization sub-module for factorizing the query matrix, the key matrix, the value matrix, and the weight matrix into at least one low-rank matrix; an element conversion sub-module for converting the elements in the low-rank matrix to generate a new weight matrix; a weight determination sub-module for superimposing the new weight matrix on the low-rank matrix to determine the fine-tuned weight; and a preset model fine-tuning sub-module for fine-tuning the preset large model based on the fine-tuned weight.

[0227] An embodiment of the present invention provides a customer group determination device. The requirement acquisition module acquires customer screening requirements; the intention determination module classifies the customer screening requirements according to a customer screening model to determine the user intention; the algorithm determination module determines the corresponding trial space algorithm for the customer screening requirements according to the user intention; the parameter extraction module extracts key parameters from the customer screening requirements; and the customer determination module inputs the key parameters into the customer screening model, and the customer screening model calls the trial space algorithm to determine the target customer group according to the key parameters. The customer screening model automatically classifies the user requirements by intention, quickly and accurately identifies the user screening target, automatically matches the trial space algorithm according to the user intention, realizes the adaptation of the algorithm to the scenario, can adapt to the business screening of various business scenarios, and can also save labor consumption, and screen out the target customer group faster to improve efficiency.

[0228] For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple. For the relevant parts, please refer to the partial description of the method embodiment.

[0229] An embodiment of the present invention further provides an electronic device, including:

[0230] It includes a processor, a memory, and a computer program stored on the memory and capable of running on the processor. When the computer program is executed by the processor, it realizes each process of the above-mentioned method embodiment for determining a customer group, and can achieve the same technical effect. To avoid repetition, it will not be described here again.

[0231] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it realizes each process of the above-mentioned method embodiment for determining a customer group, and can achieve the same technical effect. To avoid repetition, it will not be described here again.

[0232] Each embodiment in this specification is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other.

[0233] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, devices, or computer program products. Therefore, the embodiments of the present invention can take the form of all-hardware embodiments, all-software embodiments, or embodiments combining software and hardware aspects. Moreover, the embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0234] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal devices generate a device for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0235] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0236] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable terminal device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0237] Although the preferred embodiments of the embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they know the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

[0238] Finally, it should also be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or terminal device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or terminal device comprising said element.

[0239] The above provides a detailed introduction to a method and a device for determining a customer group provided by the present invention. Specific examples are used in this text to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for determining a customer group, characterized in that, The method includes: Obtaining customer screening requirements; Classifying the intent of the customer screening requirements according to a customer screening model to determine the user intent; Determining the corresponding trial calculation space algorithm for the customer screening requirements according to the user intent; Extracting key parameters from the customer screening requirements; the key parameters include at least one of a business scenario, a target scale, screening conditions, and a label combination; Inputting the key parameters into the customer screening model, and calling the trial calculation space algorithm through the customer screening model to determine a target customer group according to the key parameters.

2. The customer group determination method according to claim 1, characterized in that The trial calculation space algorithm includes a specified scale optimal trial calculation algorithm; the step of calling the trial calculation space algorithm through the customer screening model to determine a target customer group according to the key parameters includes: Obtaining customer data and the corresponding conversion probability value of the customer data; Performing a descending order sorting on the customer data according to the conversion probability value to obtain a sorting result; Determining a target customer group according to the sorting result and the target scale.

3. The customer group determination method according to claim 1, characterized in that The trial calculation space algorithm includes a specified condition optimal trial calculation algorithm; the step of calling the trial calculation space algorithm through the customer screening model to determine a target customer group according to the key parameters includes: Converting the screening conditions into structured conditional statements; Obtaining customer data and the corresponding conversion probability value of the customer data; Filtering the customer data based on the structured conditional statements to obtain a candidate customer group; Performing a descending order sorting on the candidate customer group according to the conversion probability value to obtain a sorting result; Determining a target customer group based on the sorting result and the target scale.

4. The customer group determination method according to claim 1, wherein The trial calculation space algorithm includes a label optimal threshold trial calculation algorithm; the step of calling the trial calculation space algorithm through the customer screening model to determine a target customer group according to the key parameters includes: Obtaining customer data and the corresponding conversion probability value of the customer data; Constructing a decision tree model based on the label combination and the conversion probability value; Performing feature threshold partitioning based on the gain rate of the Gini coefficient, and allocating the customer data partitioned by the feature threshold to the smallest cell cluster in the decision tree model; Performing a descending order sorting on the customer data in the smallest cell cluster to obtain a sorting result; Filtering the customer data based on the sorting result and the target scale to determine a target customer group, and calculating the corresponding estimated conversion probability value of the target customer group; the estimated conversion probability value is the expected probability of a user purchasing a recommended product.

5. The customer group determination method according to claim 1, characterized in that The trial calculation space algorithm includes a label optimal combination trial calculation algorithm; the step of calling the trial calculation space algorithm through the customer screening model to determine a target customer group according to the key parameters includes: Extracting customer data related to the business scenario from a preset database; Determining a label combination rule and a screening scale according to the business scenario; Filtering the customer data according to the label combination rule and the screening scale to determine a target customer group, and calculating the corresponding estimated conversion probability value of the target customer group.

6. The customer group determination method according to any one of claims 2-4, characterized in that, The step of obtaining the corresponding conversion probability value of the customer data includes: Obtaining marketing voice data; Extracting customer data related to the business scenario from a preset database; Obtain the conversion probability value corresponding to the customer data by quantifying the marketing voice data and the customer data through an integration algorithm; the conversion probability value is the probability of the user's purchase intention for the product.

7. The customer group determination method according to claim 1, characterized in that The customer screening model is trained in the following way: Obtain the customer screening requirements and extract key parameters from the customer screening requirements; the key parameters include at least one of business scenario, target scale, screening conditions, and label combination; Obtain a marketing voice data set; Extract the customer data set related to the business scenario from the preset database; Obtain a conversion probability value set by quantifying the marketing voice data set and the customer data set through an integration algorithm; Construct a basic data set according to the conversion probability value set; Input the basic data set into a preset large model for model fine-tuning; Determine the user intention through the fine-tuned preset large model based on the customer screening requirements, and determine the trial calculation space algorithm corresponding to the customer screening requirements according to the user intention; Input the key parameters into the fine-tuned preset large model, call the trial calculation space algorithm to train the preset large model, and obtain the customer screening model.

8. The method according to claim 7, wherein The constructing the basic data set according to the conversion probability value set includes: Convert the marketing voice data into text data and preprocess the text data; Configure a user pain point label set for the preprocessed text data; Analyze the relationship among the conversion probability value, the user pain point label, and the product by constructing a Bayesian network; Construct a basic data set according to the Bayesian network analysis result.

9. The method according to claim 7, characterized in that The inputting the basic data set into a preset large model for model fine-tuning includes: Construct a query matrix, a key matrix, a value matrix, and a weight matrix; Decompose the query matrix, the key matrix, the value matrix, and the weight matrix into at least one low-rank matrix; Perform conversion on the elements in the low-rank matrix to generate a new weight matrix; Overlay the new weight matrix on the low-rank matrix to determine the fine-tuned weight; Fine-tune the preset large model based on the fine-tuned weight.

10. A customer group determination device, characterized in that The device includes: A requirement acquisition module for obtaining customer screening requirements; An intention determination module for classifying the intention of the customer screening requirements according to the customer screening model to determine the user intention; An algorithm determination module for determining the trial calculation space algorithm corresponding to the customer screening requirements according to the user intention; A parameter extraction module for extracting key parameters from the customer screening requirements; the key parameters include at least one of business scenario, target scale, screening conditions, and label combination; A customer determination module for inputting the key parameters into the customer screening model, and calling the trial calculation space algorithm by the customer screening model to determine the target customer group according to the key parameters.

11. An electronic device, characterized in that, It includes: A processor, a memory, and a computer program stored on the memory and capable of running on the processor. When the computer program is executed by the processor, it implements the steps of a customer group determination method as described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of a customer group determination method according to any one of claims 1-9 are implemented.

Images