A method and apparatus for identifying potential users based on artificial intelligence

Abstract

This invention relates to a method and apparatus for identifying potential users based on artificial intelligence. It pertains to the field of artificial intelligence technology. The method includes: acquiring business input information and multi-dimensional data of target objects; wherein the business input information includes at least target product description information and customer group description information; modeling and processing the multi-dimensional data to construct feature engineering; wherein the feature engineering includes a general feature set for response rate models and a feature set for specific products; based on the target product description information and customer group description information, using a large language model agent, obtaining preliminary potential users; and based on the feature engineering, multi-dimensional data, and preliminary potential users, determining target potential users. This invention can improve the accuracy of identifying potential users.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically to a method and apparatus for identifying potential users based on artificial intelligence. Background Technology

[0002] With the acceleration of digital transformation in government and enterprises, identifying potential users for specific products has become a core aspect of government and enterprise marketing.

[0003] In related technologies, massive amounts of customer data are processed based on multiple manually designed key business factors to construct a user feature dataset associated with the key factors, and then potential users are screened based on the user feature dataset.

[0004] However, manually designed key business factors require business and technical personnel to sort out rules, design features, and write screening conditions for different products and customer groups, which consumes a lot of manpower and time, resulting in low efficiency in identifying potential users.

[0005] Therefore, there is an urgent need for a method that can improve the efficiency of identifying potential users. Summary of the Invention

[0006] The technical problem that this invention aims to solve is that multiple key business factors are designed manually, requiring business and technical personnel to sort out rules, design features, and write screening conditions for different products and customer groups one by one, which consumes a lot of manpower and time, resulting in low efficiency in identifying potential users.

[0007] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for identifying potential users based on artificial intelligence, the method comprising: acquiring business input information and multi-dimensional data of target objects; wherein, the business input information includes at least target product description information and customer group description information; modeling and processing the multi-dimensional data to construct feature engineering; wherein, the feature engineering includes a general feature set of response rate model and a specific product expert feature set; obtaining preliminary potential users based on the target product description information and customer group description information using a large language model intelligent agent; and determining target potential users based on feature engineering, multi-dimensional data, and preliminary potential users.

[0008] The beneficial effects of this invention are as follows: By acquiring business input information containing target product description information and customer group description information, the large language model intelligent agent directly processes the product and customer group descriptions in natural language form, automatically obtaining preliminary potential users without relying on manual design of key business factors. At the same time, feature engineering composed of a general feature set of response rate model and a specific product expert feature set is used for modeling, which greatly reduces manpower and time costs, improves recognition efficiency, avoids bias and omissions caused by human experience, improves screening consistency and stability, and the general feature set can be reused across products. Only special expert features need to be added for new products to complete the adaptation, without the need for overall remodeling, which significantly improves the model adaptation speed and scenario expansion capability, and reduces the construction cost and cycle of new product marketing models.

[0009] Based on the above technical solution, the present invention can be further improved as follows.

[0010] Furthermore, based on feature engineering, multi-dimensional data, and preliminary potential users, target potential users are identified, including: training a machine learning model based on feature engineering and multi-dimensional data to obtain a response rate classifier; wherein, the response rate classifier is used to predict the marketing response probability of the target product; based on the preliminary potential users, the marketing response probability of the target product is obtained using the response rate classifier; and based on the marketing response probability, target potential users are identified.

[0011] The beneficial effect of adopting the above-mentioned further solution is that by using feature engineering and multi-dimensional data training to obtain a response rate classifier, and using this response rate classifier to predict the target product marketing response probability of preliminary potential users, and then determining potential users based on the marketing response probability, the potential user identification results can be more accurate and reliable, ensuring that the prediction output is stable and controllable, has high interpretability, and effectively avoids the problem of unstable output of a single language model.

[0012] Furthermore, based on the marketing response probability, target potential users are identified, including: based on a preset probability threshold, the marketing response probability is screened to identify target potential users from the initial potential users whose marketing response probability is not lower than the preset probability threshold.

[0013] The beneficial effect of adopting the above-mentioned further solution is that by filtering the marketing response probability based on a preset probability threshold, it is possible to accurately identify target potential users whose marketing response probability is not lower than the preset probability threshold from the initial potential users, thereby achieving quantitative screening and precise positioning of target potential users and improving the objectivity and consistency of target potential user screening.

[0014] Furthermore, multi-dimensional data is modeled and processed to construct feature engineering, including: selecting modeling samples based on multi-dimensional data; setting observation points for the modeling samples and extracting the original multi-dimensional data before the observation points in the modeling samples; determining the positive and negative labels of the modeling samples based on whether the customer purchased the target product during the performance period; dividing the target modeling samples containing positive and negative labels into training and testing sets according to a preset ratio or cross-validation method; constructing a general feature set for the response rate model based on the first original multi-dimensional data of the training set and the second original multi-dimensional data of the testing set; extracting the target product business attributes based on the target product description information in the business input information; constructing a product-specific expert feature set based on the target product business attributes; and determining the feature engineering based on the general feature set for the response rate model and the product-specific expert feature set.

[0015] The beneficial effect of adopting the above-mentioned further solution is that by selecting modeling samples based on multi-dimensional data, setting observation points and extracting the original multi-dimensional data before the observation points, determining positive and negative labels based on whether the customer has purchased the target product during the performance period, dividing the target modeling samples into training and testing sets according to a preset ratio or cross-validation method, constructing a general feature set for the response rate model based on the original multi-dimensional data of the training and testing sets, extracting business attributes based on the target product description information and constructing a specific product expert feature set, and finally determining the feature engineering that includes the general feature set and the expert feature set, the effectiveness and reliability of the feature data can be improved.

[0016] Furthermore, modeling samples are selected based on multi-dimensional data, including: cleaning the multi-dimensional data to obtain effective multi-dimensional data; where effective multi-dimensional data is multi-dimensional data after removing invalid data, abnormal data, and data lacking key information; based on the effective multi-dimensional data, target objects that have not yet processed the target product at the observation point are selected as candidate modeling samples; based on the customer group description information in the business input information, the candidate modeling samples are initially screened, and modeling samples that do not conform to the customer group description information are removed to obtain the modeling samples.

[0017] The beneficial effect of adopting the above-mentioned further solution is that by cleaning and processing multi-dimensional data to remove invalid data, abnormal data, and data with missing key information, effective multi-dimensional data is obtained. Based on the effective multi-dimensional data, target objects that have not yet processed the target product at the observation point are selected as candidate modeling samples. Based on the customer group description information in the business input information, the candidate modeling samples are initially screened and samples that do not conform to the customer group description information are removed to obtain modeling samples. This can ensure the purity and effectiveness of the modeling samples, improve the quality of sample data, and make the samples match the target customer group and business scenario.

[0018] Furthermore, the large language model intelligent agent includes a product query intelligent agent, a customer query intelligent agent, a feature analysis intelligent agent, a potential customer query intelligent agent, and a self-calibration intelligent agent. Based on the target product description information and customer group description information, the large language model intelligent agent obtains preliminary potential users, including: using the product query intelligent agent to standardize and parse the target product description information, generate product definition query statements, and execute the product definition query statements to determine the data range and business scope of the target product; using the customer query intelligent agent to query customer data that has ordered or used the target product based on the target product definition to obtain the current customer set; using the feature analysis intelligent agent to perform feature analysis on the current customer set based on the customer group description information to extract key feature conditions for screening potential users; using the potential customer query intelligent agent to generate potential customer query statements based on key feature conditions and retrieve a set of candidate potential users from the customer attribute database; and using the self-calibration intelligent agent to determine preliminary potential users based on the set of candidate potential users.

[0019] The beneficial effect of adopting the above-mentioned further scheme is that by dividing the large language model agent into a product query agent, a customer query agent, a feature analysis agent, a potential customer query agent, and a self-calibration agent, and then using the product query agent to perform standardized parsing and product definition determination, the customer query agent to obtain the current customer set, the feature analysis agent to extract key feature conditions, the potential customer query agent to generate query statements and retrieve the candidate potential user set, and the self-calibration agent to determine the initial potential users, the screening process for the initial potential users can be clearly divided, improving the accuracy of natural language processing and data query, and ensuring the reliability of the output results of each stage.

[0020] Furthermore, the present invention provides a potential user identification device, comprising: an acquisition module for acquiring business input information and multi-dimensional data of target objects; wherein the business input information includes at least target product description information and customer group description information; a construction module for modeling and processing the multi-dimensional data to construct feature engineering; wherein the feature engineering includes a general feature set for response rate models and a specific product expert feature set; a first determination module for obtaining preliminary potential users based on the target product description information and customer group description information using a large language model agent; and a second determination module for determining target potential users based on feature engineering, multi-dimensional data, and preliminary potential users.

[0021] Furthermore, this application provides a computer device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the artificial intelligence-based potential user identification method described in the first aspect or any corresponding embodiment.

[0022] Furthermore, this application provides a computer-readable storage medium storing computer instructions for causing a computer to execute the artificial intelligence-based potential user identification method described in the first aspect or any corresponding embodiment above.

[0023] Furthermore, this application provides a computer program product, including computer instructions for causing a computer to execute the artificial intelligence-based method for identifying potential users as described in the first aspect or any corresponding embodiment. Attached Figure Description

[0024] Figure 1 A flowchart illustrating the potential user identification method based on artificial intelligence provided by this invention; Figure 2 A flowchart illustrating another potential user identification method based on artificial intelligence provided by the present invention; Figure 3 A flowchart illustrating another potential user identification method based on artificial intelligence provided by the present invention; Figure 4 A flowchart illustrating another potential user identification method based on artificial intelligence provided by the present invention; Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0025] The principles and features of the present invention are described below. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0026] With the acceleration of digital transformation in government and enterprises, identifying potential users for specific products has become a core aspect of government and enterprise marketing.

[0027] In related technologies, massive amounts of customer data are processed based on multiple manually designed key business factors to construct a user feature dataset associated with the key factors, and then potential users are screened based on the user feature dataset.

[0028] However, manually designed key business factors require business and technical personnel to sort out rules, design features, and write screening conditions for different products and customer groups, which consumes a lot of manpower and time, resulting in low efficiency in identifying potential users.

[0029] Based on this, the present invention provides an artificial intelligence-based method for identifying potential users. By acquiring business input information containing target product descriptions and customer group descriptions, a large language model agent is used to directly process the product and customer group descriptions in natural language form, automatically obtaining preliminary potential users without relying on manual design of key business factors. At the same time, feature engineering, consisting of a general feature set of response rate model and a feature set of specific product experts, is used for modeling, which significantly reduces manpower and time costs, improves identification efficiency, avoids biases and omissions caused by human experience, and improves the consistency and stability of screening. Furthermore, the general feature set can be reused across products, and only special expert features need to be added for new products to complete the adaptation, without the need for overall remodeling. This significantly improves the model adaptation speed and scenario expansion capability, and reduces the construction cost and cycle of new product marketing models.

[0030] According to an embodiment of the present invention, an embodiment of a method for identifying potential users based on artificial intelligence is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0031] This embodiment provides a method for identifying potential users based on artificial intelligence, which can be used in computer devices such as computers and servers. Figure 1 This is a flowchart of a potential user identification method based on artificial intelligence according to an embodiment of this application, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain business input information and multi-dimensional data of the target object; wherein, the business input information includes at least target product description information and customer group description information.

[0032] Business input information can be understood as textual information of business requirements submitted by the business side.

[0033] The target product description information can be understood as the textual description of the functions, benefits, application rules, scope of application, etc. of the product to be promoted and marketed.

[0034] Customer profile information can be understood as the limiting descriptions of the intended user group, such as attributes, identity characteristics, and behavioral preferences, defined by the business side.

[0035] The target audience can be understood as all users within the region who are eligible to participate in product marketing and have the qualifications to conduct business.

[0036] Multidimensional data can be understood as comprehensive raw user data, including basic user attributes, business transaction data, behavioral trajectory data, consumption data, and social connection data.

[0037] Specifically, the system collects text-based business requirement information entered into the front-end business platform, and simultaneously connects to the back-end data platform and user information database to batch retrieve various raw data of all target users. For example, if business personnel enter information about low-interest online consumer loan products (target product description information) and urban working people aged 25 to 45 with stable income (customer group description information), the system simultaneously retrieves multi-dimensional data such as age, occupation, transaction history, past loan records, and online browsing behavior of all users in that region.

[0038] As an example, it connects to the enterprise office collaboration form interface, automatically captures the product descriptions and customer requirements filled in the form, and simultaneously imports multi-dimensional user data in batches through offline data file import.

[0039] As an example, a speech-to-text module is used to convert verbal product introductions and customer requirements from business personnel into text input information. Then, big data crawlers are used to legally capture publicly available and compliant user tag data to complete multi-dimensional data acquisition.

[0040] Step S102: Model and process the multi-dimensional data to construct feature engineering; wherein, feature engineering includes a general feature set for the response rate model and a product-specific expert feature set.

[0041] Feature engineering can be understood as extracting, combining, and deriving a quantitative feature system from raw user data that can be used for model training and user judgment.

[0042] The general feature set of the response rate model can be understood as a set of basic user features that are applicable to all categories of marketing products, have strong universality, and can be reused across businesses.

[0043] The specific product expert feature set can be understood as a collection of differentiated business features that combine the business attributes and processing logic of a single product and are customized specifically for it.

[0044] Specifically, modeling operations such as sample screening, time point extraction, positive and negative sample labeling, and dataset splitting are carried out on the massive amounts of multi-dimensional user data collected. General user profile features are extracted from the standardized data, and then customized business features are created by combining these features with the current marketing product's business attributes. The two types of features are then integrated to complete the entire feature engineering project. For example, for credit products, the general feature set includes user age, gender, region, monthly income, and basic credit data, while the specific product expert feature set includes exclusive features such as user online borrowing frequency, small loan delinquency status, and frequency of inquiries about similar products.

[0045] Please refer to Table 1. Feature engineering can include a general feature set for the response rate model. This general feature set can include basic customer behavior, customer business structure, customer product ordering behavior and structure, and business characteristics. It is universal for various marketing scenarios of operator-oriented enterprise products and is applicable to the construction of marketing models for various products. The list of features engineering corresponding to this part is shown in Table 1: Please refer to Table 2. The expert feature set for a specific product varies from product to product. The core idea is to design features unique to the potential customers of that product, tailored to its specific characteristics. Features customized by experts or business stakeholders can be included in this section. The following examples use AI computing products.

[0046] AI computing products refer to hardware / computing infrastructure such as GPU cloud servers, AI-native storage, and AI computing clusters, as well as supporting platform systems such as AI service platforms and computing power platforms. Their customers typically have a strong willingness to pay for technology, and investment decisions in these projects often have strategic attributes for companies or governments. The business attributes of AI computing products themselves also place high demands on company size, accumulated data resources, and computing power investment, and are commonly found in high-tech enterprises and industrial parks. Table 2 below is a list of expert characteristic sets for a specific product, using this product as an example. Step S103: Based on the target product description information and customer group description information, the initial potential users are obtained using a large language model intelligent agent.

[0047] Large language model agents can be understood as a large model function execution cluster composed of multiple functionally independent and cooperative agents, possessing the capabilities of semantic parsing, data querying, feature analysis, and result calibration.

[0048] Preliminary potential users can be understood as a large group of intended users who have been initially screened by the large language model agent, roughly meet the basic requirements of the product and customer group, and have not been precisely scored by the model.

[0049] Specifically, product descriptions and customer group descriptions in natural language are input into five collaborative intelligent agents. These agents sequentially complete the following steps: determining product scope, retrieving existing users, extracting user characteristics, searching for candidate users, and self-calibrating results. This process batch-filters potential users matching basic criteria to form a preliminary list. For example, after inputting short video membership promotion product information and information about young online entertainment customers, the multiple agents automatically analyze the product promotion scope, retrieve users who have already activated similar memberships, extract user entertainment consumption characteristics, search for users with similar attributes, calibrate the filtering results, and output a large batch of preliminary potential users.

[0050] Step S104: Based on feature engineering, multi-dimensional data, and preliminary potential users, identify target potential users.

[0051] Target potential users can be understood as precise users with the highest willingness to respond to marketing and the best conversion probability within the initial potential user range, after probability prediction and threshold screening.

[0052] Specifically, after identifying feature engineering, multi-dimensional data, and initial potential users, target potential users can be further identified from the initial potential users.

[0053] As an example, a marketing response rate prediction model is trained based on the completed feature engineering and standardized multi-dimensional data. Initial potential users are entered into the model to complete the quantitative prediction of response probability. A second precise screening is performed according to the preset probability threshold, and finally, the precise users that can be reached by the marketing are determined. For example, for users who are initially interested in short video membership, the model is trained based on dual feature sets to predict the user's activation probability. Users whose activation probability is higher than the set threshold are selected as target potential users.

[0054] In one possible implementation, please refer to Table 3. The list of target potential users can be linked to corresponding users and sent to those users via email or other means. Taking natural customers in the target region as an example, the AI-based potential user identification method of this invention is applied to identify and score potential customers for "Yunxi". See Table 3 for details: The present invention provides an AI-based method for identifying potential users. By acquiring business input information containing target product descriptions and customer group descriptions, and utilizing a large language model agent to directly process the product and customer group descriptions in natural language form, it automatically obtains preliminary potential users without relying on manual design of key business factors. Furthermore, it employs feature engineering, consisting of a general feature set of the response rate model and a feature set of specific product experts, for modeling. This significantly reduces manpower and time costs, improves identification efficiency, avoids biases and omissions caused by human experience, and enhances screening consistency and stability. Moreover, the general feature set can be reused across products; adaptation can be completed simply by supplementing specific expert features for new products, without requiring overall model remodeling. This significantly improves model adaptation speed and scenario expansion capabilities, and reduces the construction cost and cycle of new product marketing models.

[0055] Based on step S105 of this embodiment, which uses feature engineering, multi-dimensional data, and preliminary potential users to determine target potential users, such as... Figure 2 The implementation of steps S2051 to S2053 is as follows: Step S2051: Train the machine learning model based on feature engineering and multi-dimensional data to obtain a response rate classifier; wherein, the response rate classifier is used to predict the marketing response probability of the target product.

[0056] Machine learning models can be understood as intelligent algorithm models that autonomously learn data patterns based on labeled sample data and have the ability to classify and predict probabilities.

[0057] A response rate classifier can be understood as a classification and prediction model that, after training, is specifically designed to determine a user's willingness to accept a specified marketing product and output a quantitative response probability.

[0058] Marketing response probability can be understood as a numerical result that quantifies the likelihood of a target audience actively inquiring about, processing, or accepting a corresponding target marketing product.

[0059] Specifically, after obtaining feature engineering and multi-dimensional data, a machine learning model can be trained to obtain a response rate classifier.

[0060] As an example, feature engineering integrates a pre-built general feature set with product-specific expert feature sets. This, combined with cleaned and labeled multi-dimensional user data, is then input into a basic machine learning model for iterative training. The model's internal parameters are continuously optimized, and after the training error converges, a finalized and usable response rate classifier is generated. The core function of this classifier is to output the probability of a user's marketing response to a given product. For instance, using broadband speed-up services as a marketing product, the model integrates general features from basic user profiles, expert features specific to home network usage, and labeled data from users' past broadband upgrade subscription behaviors. This results in a broadband service response rate classifier that can predict a user's willingness to subscribe to broadband speed-up services.

[0061] Step S2052: Based on the initial potential users, use a response rate classifier to obtain the marketing response probability of the target product.

[0062] Specifically, the feature data of all preliminary potential users selected in the early stage and the multi-dimensional raw data are uniformly organized into a model-recognizable input format, and imported in batches into the trained response rate classifier. The classifier then performs batch inference calculations and outputs the marketing response probability of the target product for each preliminary potential user.

[0063] Step S2053: Identify target potential users based on marketing response probability.

[0064] After determining the probability of marketing response, potential users can be further identified. The specific methods for identification will be described in detail below.

[0065] Specifically, step S2052 includes: Step a1: Based on a preset probability threshold, the marketing response probability is screened to identify target potential users whose marketing response probability is not lower than the preset probability threshold from the initial potential users.

[0066] The preset probability threshold can be understood as a fixed probability threshold value set in advance by the business side based on marketing needs and conversion standards, which serves as a standard for user screening and judgment.

[0067] Specifically, according to the pre-defined probability criteria, the marketing response probability of each initial potential user is compared one by one. Users with a probability value greater than or equal to a set threshold are selected and designated as target potential users for final business implementation and promotion. For example, for mobile data package promotion, if the business personnel preset a probability threshold of 0.6, the system will iterate through the data package application response probabilities of all initially interested users and only retain users with a response probability ≥ 0.6, identifying this group as target potential users for targeted data package promotion.

[0068] The artificial intelligence-based potential user identification method provided by this invention obtains a response rate classifier through feature engineering and multi-dimensional data training, and uses the response rate classifier to predict the marketing response probability of target products for preliminary potential users. Then, based on the marketing response probability, the potential users are determined. This makes the potential user identification results more accurate and reliable, ensures that the prediction output is stable and controllable, has high interpretability, and effectively avoids the problem of unstable output of a single language model.

[0069] Furthermore, by filtering marketing response probabilities based on preset probability thresholds, it is possible to accurately identify target potential users whose marketing response probability is not lower than the preset probability threshold from the initial potential users, thereby achieving quantitative screening and precise positioning of target potential users and improving the objectivity and consistency of target potential user screening.

[0070] Based on the multi-dimensional data modeling process in step S102 of this embodiment, feature engineering is constructed as follows: Figure 3 The implementation of steps S3021 to S3028 in the above steps: Step S3021: Select modeling samples based on multi-dimensional data.

[0071] Modeling samples can be understood as compliant user sample data selected from all target objects that can be used for model training to learn patterns.

[0072] Specifically, from the massive amount of multi-dimensional user data that has been collected, user groups that meet the basic business access criteria are selected according to business screening rules and used as modeling samples for subsequent modeling. For example, for car insurance promotion business, people without vehicle registration information are removed from the multi-dimensional data of all car owners, and car owners with valid vehicle information are selected as models for car insurance business.

[0073] Specifically, step S3021 includes: Step b1 involves cleaning the multi-dimensional data to obtain valid multi-dimensional data; where valid multi-dimensional data is multi-dimensional data that has been freed from invalid data, abnormal data, and missing key information.

[0074] Effective multidimensional data can be understood as multidimensional data after removing invalid and redundant data, abnormal data that deviates from the normal range, and data with missing key fields.

[0075] Specifically, a comprehensive cleaning and governance process is carried out on all the aggregated user multi-dimensional raw data. Each data item is examined, and invalid and redundant data without business value, as well as abnormal data with values ​​deviating from normal ranges, are centrally removed. Simultaneously, data with large gaps in core key fields or incomplete information is removed. Ultimately, data with complete fields, uniform format, and authentic and usable data is retained, forming standardized and effective multi-dimensional data. For example, in the modeling scenario for low-interest loan products, user credit data, income data, and borrowing behavior data are cleaned, removing duplicate and invalid transaction data, abnormal data with income values ​​far exceeding reasonable ranges, and user data with incomplete key information such as age, occupation, and residential address. This process yields clean and compliant effective multi-dimensional data for loan users.

[0076] Step b2: Based on valid multi-dimensional data, select target objects that have not yet processed the target product at the time of observation as candidate modeling samples.

[0077] Candidate modeling samples can be understood as alternative user samples that have undergone data cleaning and have not yet processed the corresponding target product at the observation point node, and which initially meet the basic conditions for modeling.

[0078] Specifically, by using effective multi-dimensional data and pre-set observation points as time criteria, the business processing records of the target objects are screened to identify user groups who have never applied for, activated, or processed the current target product before that time point, and these users are uniformly defined as candidate modeling samples.

[0079] Step b3: Based on the customer group description information in the business input information, perform preliminary screening of candidate modeling samples, remove modeling samples that do not conform to the customer group description information, and obtain modeling samples.

[0080] Specifically, the customer description information entered and retained in the early stage of business is retrieved and used as the criteria for judging the target audience. All candidate modeling samples obtained in the previous step are then screened again to obtain modeling samples.

[0081] As an example, the customer description information entered and retained in the early stage of business is retrieved and used as the criteria for judging the target audience. All candidate modeling samples obtained in the previous step are then subjected to a second precise verification and screening. Samples that do not meet the requirements of customer positioning in terms of age, occupation, spending power, geographical range, usage habits, etc. are directly eliminated, and high-quality samples that fit the business marketing target audience positioning are retained to form officially usable modeling samples.

[0082] Step S3022: Set observation points for the modeling sample and extract the original multi-dimensional data before the observation points in the modeling sample.

[0083] An observation point can be understood as an artificially defined time boundary node, serving as the endpoint for capturing historical behavioral data.

[0084] Raw multidimensional data can be understood as the original data such as user basic attributes, behaviors, and business transactions that have not undergone feature derivation or computational processing.

[0085] Specifically, a fixed time point is uniformly set as the observation point for all selected modeling samples. All original user behaviors, attributes, and business data retained up to and including that time point are uniformly extracted as the historical basis for model learning. For example, the 1st of each month is set as the unified observation point, and all original multi-dimensional data such as consumption records, business browsing records, and basic information of all modeling samples within the six months prior to the 1st of that month are extracted.

[0086] Step S3023: Determine the positive and negative labels for the modeling sample based on whether the customer has processed the target product during the performance period.

[0087] The performance period can be understood as a fixed observation period set after the observation point, used to determine whether the user has engaged in the target product purchase behavior.

[0088] Positive labels can be understood as sample markers for successful transactions of the target product during the performance period, representing high-intent samples.

[0089] Negative labels can be understood as marking samples that did not process the target product during the performance period, representing low-intent samples.

[0090] Specifically, a fixed period of time is defined after the observation point as the performance period. The business transactions of all modeled samples during this period are statistically analyzed. Successful transactions for the target product are marked with a positive label, while those not completed are marked with a negative label. For example, if the performance period is set to three months after the observation point, samples that successfully completed credit card installment transactions within these three months are marked with a positive label, while samples that did not complete the transaction are marked with a negative label.

[0091] Step S3024: Divide the target modeling samples containing positive and negative labels into training and test sets according to a preset ratio or cross-validation method.

[0092] Specifically, after all the modeling samples with completed positive and negative labeling are integrated, they are divided into a training set for model training and a test set for performance verification, according to a fixed ratio preset by the business or by adopting a cross-validation method. For example, the labeled credit business modeling samples are divided in a fixed ratio of 7:3, with 70% of the samples assigned to the training set and 30% assigned to the test set.

[0093] Step S3025: Based on the first original multi-dimensional data of the training set and the second original multi-dimensional data of the test set, construct a general feature set for the response rate model.

[0094] The first set of original multidimensional data can be understood as the original data before the observation point corresponding to the training set sample.

[0095] The second set of original multidimensional data can be understood as the original data before the observation point corresponding to the test set sample.

[0096] Specifically, the first set of original multi-dimensional data corresponding to the training set and the second set of original multi-dimensional data corresponding to the test set are extracted respectively. From these two types of data, user basic attributes, behavioral habits, and basic qualification features applicable to all categories of marketing products are extracted and summarized to form a general feature set. For example, based on the original user data in the training and test sets, basic features such as user age, gender, residence location, income level, and online activity time are extracted to build a general feature set for a universal response rate model.

[0097] Step S3026: Extract the target product business attributes based on the target product description information in the business input information.

[0098] The target product's business attributes can be understood as the product's unique business characteristics, which consist of product functions, application thresholds, target audience, usage scenarios, and rights and benefits rules.

[0099] Specifically, the system reads the target product description information in natural language format entered by business personnel, and extracts and sorts out the target product's business attributes such as product application conditions, target audience, business purpose, and rights and benefits rules. For example, from the description information of a whole-house smart home package, it extracts multiple product business attributes such as targeting people decorating new homes, including home appliance linkage functions, and supporting installment payment.

[0100] Step S3027: Based on the business attributes of the target product, construct a feature set of specific product experts.

[0101] Specifically, after determining the business attributes of the target product, a set of features for specific product experts can be constructed.

[0102] As an example, based on the extracted target product business attributes, combined with front-line business operation experience, industry business processing logic, and data patterns accumulated from marketing of similar products, we deeply mine differentiated data dimensions that are strongly correlated with the target product's business scenarios, access conditions, usage needs, consumption habits, and business processing procedures. We then design specific quantitative and behavioral indicators that fit the product's audience needs, business processing thresholds, rights usage rules, and conversion paths. We eliminate general basic features and focus on the product's specific business scenarios for feature derivation, statistical aggregation, and dimension decomposition. We compile all specific business features that are only applicable to the currently promoted product and can accurately reflect users' demand for the product and their processing inclinations. We classify, organize, and standardize these features according to feature type, data source, and business function, and finally integrate and summarize them to form a specific product expert feature set that is clearly structured, comprehensive in dimensions, and fits the actual business logic of the product.

[0103] Step S3028: Based on the general feature set of the response rate model and the expert feature set of specific products, determine the feature engineering.

[0104] Specifically, the general feature set of the response rate model, which has the ability to be reused across businesses, is integrated and spliced ​​with the feature set of specific product experts that fits the business logic of individual products. The format and field standards are unified and standardized, and finally combined to form a complete feature engineering that can be directly used in the model.

[0105] The potential user identification method provided by this invention selects modeling samples based on multi-dimensional data, sets observation points and extracts the original multi-dimensional data before the observation points, determines positive and negative labels based on whether the customer has purchased the target product during the performance period, divides the target modeling samples into training and testing sets according to a preset ratio or cross-validation method, constructs a general feature set for the response rate model based on the original multi-dimensional data of the training and testing sets, extracts business attributes based on the target product description information and constructs a specific product expert feature set, and finally determines the feature engineering that includes the general feature set and the expert feature set, which can improve the effectiveness and reliability of feature data.

[0106] Furthermore, by cleaning and processing multi-dimensional data to remove invalid, abnormal, and missing key information, effective multi-dimensional data is obtained. Based on the effective multi-dimensional data, target objects who have not yet processed the target product at the observation point are selected as candidate modeling samples. The candidate modeling samples are then preliminarily screened based on the customer group description information in the business input information, and samples that do not conform to the customer group description information are removed to obtain modeling samples. This ensures the purity and effectiveness of the modeling samples, improves the quality of sample data, and makes the samples match the target customer group and business scenario.

[0107] The large language model intelligent agent includes a product query intelligent agent, a customer query intelligent agent, a feature analysis intelligent agent, a potential customer query intelligent agent, and a self-calibration intelligent agent. Based on step S103 of this embodiment, which uses the target product description information and customer group description information, the large language model intelligent agent obtains preliminary potential users such as... Figure 4 The implementation of steps S4031 to S4035 is as follows: Step S4031: The product query intelligent agent performs standardized parsing of the target product description information, generates a product definition query statement, and executes the product definition query statement to determine the data range and business definition of the target product.

[0108] Product definition query statements can be understood as standardized data retrieval statements that are derived from natural language product descriptions, conform to database query specifications, and can be executed directly.

[0109] Business scope can be understood as a unified business judgment standard, including the scope of business statistics, the definition of processing rules, the user statistics dimensions, and the scope of business affiliation for the target product.

[0110] The data scope can be understood as the data source, data field, data time interval, and other data access boundaries defined when conducting user data statistics, retrieval, and filtering.

[0111] Specifically, the target product description information can be input into the product query agent to obtain a product definition query statement. This query statement is then executed to complete business data boundary verification, determining the data scope and business definition of the target product. For example, by inputting the description information of a home broadband speed-up package, the product query agent completes standardized text parsing, automatically generates and executes a query statement for existing broadband users, ultimately determining that the product only counts residential users, excluding commercial office users. Simultaneously, it defines the business data from the past year as the data retrieval range, thus obtaining the business definition and data retrieval scope.

[0112] Step S4032: Through the customer query agent, based on the target product criteria, query customer data that has ordered or used the target product to obtain the current customer set.

[0113] Specifically, the customer query intelligent agent is activated, and based on the business processing database and user behavior database, it retrieves in batches all customer data that have successfully subscribed, activated, and are using the target product normally on a daily basis. All retrieved existing users are then integrated and aggregated to form a complete set of current customers.

[0114] Step S4033: The feature analysis agent performs feature analysis on the current customer set based on the customer group description information, and extracts key feature conditions for screening potential users.

[0115] Key characteristics can be understood as conditions extracted from existing high-quality users that represent the core attributes of the target customer group.

[0116] Specifically, a feature analysis intelligent agent is activated, using pre-entered customer group descriptions as the core reference standard. This agent conducts comprehensive, in-depth data analysis, behavioral pattern assessment, and attribute summarization on the integrated existing customer base. It deeply mines commonalities among high-quality existing users, such as shared identity attributes, consumption habits, service preferences, and behavioral patterns, extracting several key characteristic conditions with strong discriminative power that can be directly used for rapid screening of similar groups. For example, combining descriptions of middle-aged and young home-based consumers, the feature analysis intelligent agent analyzes data on the existing broadband speed-up user base, extracting several key characteristic screening conditions, such as residing in residential communities, having a large number of home network devices, high online audio-visual data consumption, and regularly subscribing to home network value-added services.

[0117] Step S4034: The potential customer query agent generates a potential customer query statement based on key feature conditions and retrieves a set of candidate potential users from the customer attribute database.

[0118] A potential customer query statement can be understood as a statement that queries for potential users.

[0119] Specifically, the intelligent agent for potential customer query is run. Based on all the key feature conditions extracted in the feature analysis stage, the feature conditions in natural language form are automatically converted into standardized potential customer query statements that are adapted to the retrieval rules of the customer attribute database. Based on these query statements, the system accesses the full-domain customer attribute database to carry out large-scale batch matching and retrieval of people. All users who meet the feature conditions are collected and summarized to form a large set of candidate potential users.

[0120] Step S4035: Based on the candidate potential user set, the self-calibrating agent determines the preliminary potential users.

[0121] After determining the set of candidate potential users, a self-calibrating agent can be used to obtain preliminary potential users.

[0122] As an example, the self-calibrating agent is invoked to perform comprehensive compliance verification, data review, and scope correction on the candidate potential user set obtained from the previous batch retrieval. It automatically removes user data with abnormal information, serious discrepancies with the core customer group positioning, business conflicts, and duplicate redundancy within the set. At the same time, it optimizes the scope of audience screening, completes batch audience simplification and compliance correction, and obtains preliminary potential users.

[0123] The potential user identification method provided by this invention divides the large language model agent into a product query agent, a customer query agent, a feature analysis agent, a potential customer query agent, and a self-calibration agent. The product query agent performs standardized parsing and determines product specifications; the customer query agent obtains the current customer set; the feature analysis agent extracts key feature conditions; the potential customer query agent generates query statements and retrieves a set of candidate potential users; and the self-calibration agent determines preliminary potential users. This approach ensures a clear division of labor in the preliminary potential user screening process, improves the accuracy of natural language processing and data querying, and guarantees reliable output results at each stage.

[0124] In some embodiments, the potential user identification device of the present invention can be implemented in a combination of hardware and software. As an example, the potential user identification device of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the artificial intelligence-based potential user identification method of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0125] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0126] An electronic device according to an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned artificial intelligence-based potential user identification methods. That is, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store the computer program; the processor is used to execute the artificial intelligence-based potential user identification method shown in any embodiment of the present invention by calling the computer program.

[0127] In one alternative embodiment, an electronic device is provided, such as Figure 5 As shown, Figure 5The illustrated electronic device 5000 includes a processor 5001 and a memory 5003. The processor 5001 and the memory 5003 are connected, for example, via a bus 5002. Optionally, the electronic device 5000 may further include a transceiver 5004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 5004 is not limited to one type, and the structure of the electronic device 5000 does not constitute a limitation on the embodiments of the present invention.

[0128] Processor 5001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 5001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0129] Bus 5002 may include a path for transmitting information between the aforementioned components. Bus 5002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 5002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus 5002 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.

[0130] The memory 5003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0131] The memory 5003 stores application code (computer program) for executing the present invention, and its execution is controlled by the processor 5001. The processor 5001 executes the application code stored in the memory 5003 to implement the content shown in the foregoing method embodiments.

[0132] Among them, electronic devices can also be terminal devices, which can be any device that can install applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0133] It should be noted that, Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0134] An embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-described artificial intelligence-based methods for identifying potential users.

[0135] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0136] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned artificial intelligence-based potential user identification method.

[0137] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0138] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0139] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EEPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0140] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.

[0141] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0142] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and do not imply a specific order or sequence. Where appropriate, the order of use of similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0143] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a circuit, module, or system. Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0144] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for identifying potential users based on artificial intelligence, characterized in that, The method includes: Acquire multi-dimensional data on business input information and target objects; wherein, the business input information includes at least target product description information and customer group description information; The multi-dimensional data is modeled and processed to construct feature engineering; wherein, the feature engineering includes a general feature set for the response rate model and a product-specific expert feature set; Based on the target product description information and customer group description information, a preliminary potential user is obtained using a large language model intelligent agent; Based on the feature engineering, the multi-dimensional data, and the preliminary potential users, target potential users are identified.

2. The method for identifying potential users based on artificial intelligence according to claim 1, characterized in that, The process of identifying target potential users based on the feature engineering, the multi-dimensional data, and the initial potential users includes: The machine learning model is trained based on the feature engineering and the multi-dimensional data to obtain a response rate classifier; wherein, the response rate classifier is used to predict the marketing response probability of the target product. Based on the initial potential users, the marketing response probability of the target product is obtained using the response rate classifier. Based on the marketing response probability, target potential users are identified.

3. The method for identifying potential users based on artificial intelligence according to claim 2, characterized in that, Based on the aforementioned marketing response probability, target potential users are identified, including: Based on a preset probability threshold, the marketing response probability is screened to identify target potential users whose marketing response probability is not lower than the preset probability threshold from the initial potential users.

4. The method for identifying potential users based on artificial intelligence according to claim 1, characterized in that, The process of modeling and processing the multi-dimensional data to construct feature engineering includes: Modeling samples are selected based on the aforementioned multi-dimensional data; Set observation points for the modeling sample and extract the original multi-dimensional data in the modeling sample before the observation points; Determine the positive and negative labels for the modeling sample based on whether the customer purchases the target product during the performance period; The target modeling samples containing positive and negative labels are divided into training and test sets according to a preset ratio or cross-validation method. Based on the first original multi-dimensional data of the training set and the second original multi-dimensional data of the test set, a general feature set for the response rate model is constructed. Extract the target product's business attributes based on the target product description information in the business input information; Based on the business attributes of the target product, construct a feature set of specific product experts; The feature engineering is determined based on the general feature set of the response rate model and the expert feature set of the specific product.

5. The method for identifying potential users based on artificial intelligence according to claim 4, characterized in that, The selection of modeling samples based on the multi-dimensional data includes: The multi-dimensional data is cleaned to obtain effective multi-dimensional data; wherein, the effective multi-dimensional data is multi-dimensional data after removing invalid data, abnormal data and missing key information. Based on the effective multi-dimensional data, target objects that have not yet processed the target product at the time of the observation point are selected as candidate modeling samples. Based on the customer group description information in the business input information, the candidate modeling samples are initially screened, and modeling samples that do not conform to the customer group description information are eliminated to obtain the modeling samples.

6. The method for identifying potential users based on artificial intelligence according to any one of claims 1-5, characterized in that, The large language model intelligent agent includes a product query intelligent agent, a customer query intelligent agent, a feature analysis intelligent agent, a potential customer query intelligent agent, and a self-calibration intelligent agent. Based on the target product description information and customer group description information, the large language model intelligent agent obtains preliminary potential users, including: The product query intelligence agent performs standardized parsing of the target product description information, generates a product definition query statement, and executes the product definition query statement to determine the data range and business definition of the target product. The customer query agent queries customer data that has ordered or used the target product based on the target product criteria, thus obtaining the current customer set. The feature analysis agent performs feature analysis on the current customer set based on the customer group description information, and extracts key feature conditions for screening potential users. The potential customer query agent generates a potential customer query statement based on the key feature conditions and retrieves a set of candidate potential users from the customer attribute database. The self-calibrating agent determines preliminary potential users based on the set of candidate potential users.

7. A device for identifying potential users, characterized in that, The device includes: The acquisition module is used to acquire business input information and multi-dimensional data of the target object; wherein, the business input information includes at least target product description information and customer group description information; A construction module is used to model and process the multi-dimensional data and construct feature engineering; wherein, the feature engineering includes a general feature set for the response rate model and a product-specific expert feature set; The first determining module is used to obtain preliminary potential users based on the target product description information and customer group description information, using a large language model intelligent agent; The second determining module is used to determine target potential users based on the feature engineering, the multi-dimensional data, and the preliminary potential users.

8. An electronic device, characterized in that, include: A processor and a memory connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to cause the processor to perform the artificial intelligence-based potential user identification method as described in any one of claims 1-6.

9. A non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform any one of claims 1-6, the method for identifying potential users based on artificial intelligence.

10. A computer program product comprising a computer program that, when executed by a processor, implements the steps of the artificial intelligence-based potential user identification method according to any one of claims 1-6.

Images