User portrait construction method and device, equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence, can be applied to business systems of intelligent medical treatment, financial science and technology and the like, and discloses a user portrait construction method, device, equipment and medium, and the method comprises the steps: obtaining multi-source data of a user, classifying the multi-source data according to a preset data classification system, and obtaining a classification result, the classification result at least comprises a data type and a service attribute label of each piece of data; based on the classification result, one or more prompt templates containing data field placeholders and feature extraction task description are matched and called from a pre-constructed prompt template library; filling the data field placeholders of the called prompt template with the multi-source data corresponding to the user, and generating a feature extraction prompt set of the user; analyzing and executing each feature extraction prompt in the feature extraction prompt set, and fusing all extracted features to obtain fused features; and constructing a multi-dimensional portrait of the user based on the fusion features.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and natural language processing, and in particular to a method, apparatus, device and medium for constructing user profiles. Background Technology

[0002] In recent years, with the development of artificial intelligence (AI) technology and its deep penetration into the fintech and healthcare sectors, the technological implementation of risk control and personalized services has become a crucial support for the business operations of financial and medical institutions. Building accurate user profiles is fundamental for risk control, personalized product recommendations, and efficient customer service. For example, in smart healthcare, personalized diagnosis and treatment and disease risk prediction can be achieved through user profiles; in fintech, user profiles can be used to assess the probability of default during loan approval, thereby improving risk control levels.

[0003] However, existing technologies typically rely on structured data to build user profiles based on traditional machine learning or basic deep learning models. For example, in rehabilitation interventions, models usually assess treatment based on users' diagnostic records, test results, and medication records; in credit approvals, they typically assess risk based on users' transaction records and credit scores. This neglects the value of textual data, resulting in incomplete profile information and inaccurate feature extraction. Furthermore, the application of traditional user profile models has many limitations, making it difficult to cope with rapidly changing scenario requirements and unable to provide targeted support based on new business scenarios and needs. This affects the effectiveness of user profiles in business decisions such as personalized medicine and financial risk control decisions. Summary of the Invention

[0004] This invention provides a user profile construction method, apparatus, device, and medium to solve the technical problems of existing user profile construction models having inaccurate feature extraction and being unable to cope with rapidly changing scenario requirements, and failing to provide targeted support according to new business scenarios and needs.

[0005] Firstly, a user profile construction method is provided, including: The system acquires multi-source data from users and classifies the multi-source data according to a preset data classification system to obtain classification results. The classification results include at least the data type and business attribute tags of each data. The multi-source data includes structured data and unstructured text data. Based on the classification results, one or more prompt templates are matched and called from a pre-built prompt template library; each prompt template includes data field placeholders and a description of the feature extraction task, and is associated with data type and business attribute tags; The user's corresponding multi-source data is filled into the data field placeholders of the invoked prompt template to generate the user's feature extraction prompt set; Each feature extraction prompt in the set of feature extraction prompts is parsed and executed to extract features based on the feature extraction task description and fill data in the feature extraction prompts, and all extracted features are fused to obtain fused features; Based on the fusion features, a multi-dimensional profile of the user is constructed.

[0006] Secondly, a user profile building device is provided, including: The classification module is used to acquire multi-source data from users and classify the multi-source data according to a preset data classification system to obtain classification results. The classification results include at least the data type and business attribute tags of each data. The multi-source data includes structured data and unstructured text data. The template matching module is used to match and call one or more prompt templates from a pre-built prompt template library based on the classification results; wherein each prompt template includes data field placeholders and a feature extraction task description, and is associated with data type and business attribute tags; The prompt generation module is used to fill the data field placeholders of the prompt template called with the corresponding multi-source data of the user, and generate the feature extraction prompt set of the user; The feature extraction module is used to parse and execute each feature extraction prompt in the feature extraction prompt set, to extract features according to the feature extraction task description and fill data in the feature extraction prompt, and to fuse all extracted features to obtain fused features; The profile building module is used to build a multi-dimensional profile of the user based on the fused features.

[0007] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described user profile construction method.

[0008] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the user profile construction method described above.

[0009] In the aforementioned user profile construction method, apparatus, device, and medium, the acquired multi-source user data can be classified to determine the data type and business attribute tags of each data. Then, based on the data type and business attribute tags, one or more prompt templates including data field placeholders and feature extraction task descriptions are matched and called from a pre-built prompt template library. Corresponding data is then filled into the data field placeholders of the called prompt templates to generate a feature extraction prompt set for the user. Each feature extraction prompt in the set is then parsed and executed to extract features based on the feature extraction task description and filled data in the prompt. All extracted features are then fused to obtain fused features. Finally, based on… The fusion features construct a multi-dimensional profile of the user. In this invention, for business scenarios such as personalized diagnosis and treatment in smart healthcare and loan risk assessment in fintech, a pre-built prompt template can be intelligently matched based on data type and business attributes. The feature extraction task description in the prompt template guides the targeted feature extraction of relevant data (including structured and unstructured data) from the current user's multi-source data. This fully utilizes unstructured text data, making the profile closer to the user's real state and achieving guided feature extraction. It can quickly provide targeted support according to different business scenarios, improving the depth and credibility of the profile while adapting to rapidly changing business scenarios. It also allows for clear tracing of the prompt template for profile feature extraction, enhancing interpretability. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of an application environment for a user profile construction method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the user profile construction method provided in the first embodiment of the present invention; Figure 3 yes Figure 2 A schematic diagram of a specific implementation of step S120; Figure 4 yes Figure 2 A schematic diagram of a specific implementation of step S140; Figure 5 This is a flowchart illustrating the user profile construction method provided in the second embodiment of the present invention; Figure 6This is a schematic diagram of a user profile building device according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 8 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] The user profile construction method provided in this embodiment of the invention can be applied to, for example... Figure 1In this application environment, the client communicates with the server via a network. The server can periodically obtain multi-source data from the user, including structured and unstructured text data, from the network, or it can obtain multi-source data from the client. The server categorizes the obtained multi-source data, determines the data type and business attribute tags of each data point, and then, based on the data type and business attribute tags, matches and calls one or more prompt templates from a pre-built prompt template library. These prompt templates include data field placeholders and feature extraction task descriptions. The corresponding data is then filled into the data field placeholders of the called prompt templates, generating a feature extraction prompt set for the user. Each feature extraction prompt in the set is then parsed and executed to extract features based on the feature extraction task description and filled data in the prompt. Finally, all extracted features are fused. The fused features are obtained, and finally, a multi-dimensional profile of the user is constructed based on these features. In this invention, for business scenarios such as personalized diagnosis and treatment in smart healthcare and loan risk assessment in fintech, a pre-built prompt template can be intelligently matched according to data type and business attributes. The feature extraction task description in the prompt template guides the targeted feature extraction of relevant data (including structured and unstructured data) from the current user's multi-source data, fully utilizing unstructured text data to make the profile more closely reflect the user's true state. It also achieves guided feature extraction, providing rapid and targeted support for different business scenarios. This enhances the depth and credibility of the profile while adapting to rapidly changing business scenarios. Furthermore, it allows for clear traceability of the prompt templates used for profile feature extraction, improving interpretability. In this invention, the client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a separate server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0014] Please see Figure 2 As shown, Figure 2 A flowchart illustrating the user profile construction method provided in the first embodiment of the present invention includes the following steps S110-S150: S110: Obtain multi-source data from users and classify the multi-source data according to a preset data classification system to obtain classification results. The classification results include at least the data type and business attribute tags of each data.

[0015] In this embodiment, the multi-source data includes structured data and unstructured text data. Specifically, structured data refers to data with a fixed format that can be represented by tables or database fields, such as age, gender, transaction records, blood pressure, BMI, number of hospitalizations, etc. Unstructured text data refers to text data without a fixed format that is difficult to represent directly in tables and requires feature extraction through natural language processing (NLP) and other methods, such as customer reviews, chat logs, doctor's records, medical records, and symptom descriptions.

[0016] In this step, the data in the multi-source data are classified according to a preset data classification system, which includes pre-defined data type determination rules and business attribute label mapping rules.

[0017] In this invention, data types can include structured data and unstructured text data. Specifically, based on data type determination rules, each piece of data in the multi-source data is identified as either structured or unstructured text data. Then, using pre-defined business attribute tag mapping rules, one or more business attribute tags are assigned to each piece of data based on its source, content keywords, or field meaning. For example, in a financial scenario, business attribute tags can include tags such as credit attributes, asset attributes, transaction behavior attributes, and risk sentiment attributes. Data from credit reports can be labeled as credit attributes, text containing keywords such as "borrowing" and "turnover" can be labeled as "funding need attributes," and text containing keywords such as "buy," "sell," and "transaction amount" can be labeled as "transaction behavior attributes." In a smart healthcare scenario, business attribute tags can include attribute tags such as blood glucose monitoring attributes, blood glucose control attributes, and symptom descriptions. Text data from medical records can be labeled as "symptom descriptions," and data containing keywords such as "blood glucose sequence" can be labeled as "blood glucose monitoring attribute tags."

[0018] S120: Based on the classification results, match and call one or more prompt templates from the pre-built prompt template library; wherein each prompt template includes data field placeholders and a feature extraction task description, and is associated with data type and business attribute tags.

[0019] In this step, based on the data types and business attribute tags of each data point in the multi-source data, a matching prompt template is quickly retrieved from a pre-built prompt template library. This prompt template includes data field placeholders and a description of the feature extraction task, which can be used to guide the targeted extraction of user features. This ensures that the final extracted features are strongly correlated with the actual business scenario task, providing more reliable data support for critical decisions such as high-risk financial approvals and personalized medical recommendations.

[0020] Specifically, such as Figure 3As shown, step S120 includes the following steps S121-S122: S121: Based on the data type and business attribute tags of each data in the multi-source data, retrieve all associated prompt templates from the pre-built prompt template library; wherein, the prompt templates are associated with dynamic weights.

[0021] In this invention, each prompt template is associated with a corresponding data type and business attribute tag. The prompt template library can include prompt templates for different business scenarios, as well as prompt templates for different data types and business attribute tags for the same business scenario. Different prompt templates are used to guide feature extraction for different business scenarios or data types and business attributes. For example, for structured data and transaction behavior attributes in a financial scenario, the prompt template could be "Based on user [transaction amount], [transaction frequency], and [transaction type] data, analyze the user's average monthly consumption, main consumption areas, and consumption stability." For unstructured text data and symptom description attributes in a smart healthcare online diagnosis and treatment scenario, the prompt template could be "From the user's [description text], identify keywords related to diabetic complications and assess the severity of the symptom description (mild, moderate, severe)."

[0022] In this step, a search and matching process is performed in a pre-built prompt template library based on the data type and business attribute tags of each data in the multi-source data to obtain all prompt templates that match the data type and business attribute tags of the data.

[0023] S122: Sort the prompt templates according to their dynamic weights and select one or more prompt templates with dynamic weights higher than a preset threshold for invocation.

[0024] In this step, the matched prompt templates are sorted according to their associated dynamic weights, and one or more prompt templates with dynamic weights higher than a preset threshold are selected for invocation. If multiple prompt templates need to be invoked, they are invoked in order of sorting priority.

[0025] S130: Fill the data field placeholders of the called prompt template with the corresponding multi-source data of the user to generate the feature extraction prompt set of the user.

[0026] In this invention, the data field placeholders in the prompt template are configured to accept the filling of structured data and / or unstructured text data.

[0027] In this step, structured data and / or unstructured text data from the multi-source data that correspond to the data type and business attribute tags of the invoked prompt template are filled into the data field placeholders to generate feature extraction prompts. All feature extraction prompts generated after filling the invoked prompt template constitute a feature extraction prompt set. For example, a user's transaction data (average monthly transaction amount of 5,000 yuan, transaction frequency of 10 times / month, and main transaction type of daily consumption) is filled into a prompt template associated with structured data and transaction behavior attributes to generate a feature extraction prompt: "Based on the user's average monthly transaction amount of 5,000 yuan, transaction frequency of 10 times / month, and main transaction type of daily consumption, analyze the user's average monthly consumption, main consumption areas, and consumption stability."

[0028] S140: Parse and execute each feature extraction prompt in the feature extraction prompt set to extract features according to the feature extraction task description and fill data in the feature extraction prompt, and fuse all extracted features to obtain fused features.

[0029] In this step, each feature extraction prompt in the feature extraction prompt set is executed. Based on the feature extraction task description and the filled data contained therein, the corresponding analysis model is called for processing to obtain the corresponding feature value. Preferably, in this embodiment, all output features are concatenated to achieve feature fusion and obtain fused features. In some other embodiments, a weighted summation method can also be used to fuse all extracted features.

[0030] Specifically, such as Figure 4 As shown, step S140 includes the following steps S141-S142: S141: Parse each feature extraction prompt in the feature extraction prompt set, and identify the data type of the data filled in the data field placeholder of the feature extraction prompt and the corresponding feature extraction task.

[0031] In this step, the data type of the feature extraction prompt generated by filling the data placeholders in the prompt template is identified, and the corresponding feature extraction task is obtained so that different models can be called for feature extraction analysis for different data types.

[0032] S142: Based on the feature extraction task, the corresponding analysis model is called to perform feature extraction according to the data type of the filled data.

[0033] In this invention, a prompt-guided feature extraction method is adopted, that is, a task-driven feature extraction method. Under the guidance of the feature extraction task description in the feature extraction prompt, a quantified feature value is generated by calling an appropriate analysis model.

[0034] Specifically, when the data being filled is structured data, a statistical analysis model or machine learning model is invoked to extract numerical features. That is, the statistical analysis model or machine learning model is invoked to perform calculations and obtain corresponding numerical feature values, such as a user's credit score or repayment ability index. When the data being filled is unstructured text data, a pre-trained large language model is used to extract text features. That is, the powerful semantic understanding and reasoning capabilities of the large language model are utilized to perform semantic understanding and quantitative analysis on the unstructured text data according to the feature extraction task, and output text-derived feature values, such as outputting the severity of diabetic complications as mild.

[0035] S150: Based on the fusion features, construct a multi-dimensional profile of the user.

[0036] In this step, the multi-dimensional profile includes one or more profile labels obtained based on the feature values ​​in the fusion features. For example, for smart healthcare scenarios, the user's fusion features include blood glucose volatility (e.g., large short-term volatility) and complication risk level (peripheral neuropathy: moderate risk, retinopathy: low risk). The overall risk assessment label can be calculated by a rule engine and is labeled as medium to high risk. Each feature value in the fusion features is associated with a prompt template and a data source for filling. The above features, the prompt templates and data sources associated with the above features, and the overall risk assessment together constitute the user's multi-dimensional profile, that is, a multi-dimensional profile including a blood glucose control profile and a complication risk profile and their associated information is obtained. For financial scenarios, the fusion features may include credit attribute features (credit rating, default probability, etc.) and transaction behavior attribute features (consumption preference, consumption amount, etc.), etc., to obtain a multi-dimensional profile including a credit profile and a consumption demand profile.

[0037] Furthermore, in some embodiments, the user profile construction method may further include: obtaining actual feedback information on the business decision results made based on the constructed multi-dimensional profile; and adjusting the dynamic weights and / or feature extraction task descriptions of the prompt templates in the prompt template library according to the actual feedback information. That is, the multi-dimensional profile is applied to business decisions, and feedback on the decision effect is collected. Based on the decision effect feedback, the dynamic weights associated with the prompt templates used to guide the generation of relevant features are adjusted, and / or the specific content of the feature extraction task description in the prompt template is optimized. For example, if the profile generated by a certain prompt template has a low accuracy rate in actual business applications, the dynamic weight of that prompt template is reduced, causing its ranking to drop. Through continuous feedback optimization, the accuracy and adaptability of user profile analysis and generation are improved.

[0038] As can be seen, in the above solution, for business scenarios such as personalized diagnosis and treatment in smart healthcare and loan risk assessment in fintech, a pre-built prompt template can be intelligently matched based on data type and business attributes. The prompt template can guide the targeted feature extraction of relevant data (including structured and unstructured data) from multiple sources of current user data according to the feature extraction task description. This fully utilizes unstructured text data, making the profile closer to the user's real state. It also achieves guided feature extraction, which can quickly provide targeted support according to different business scenarios. While improving the depth and credibility of the profile, it can also adapt to rapidly changing business scenarios. Furthermore, each feature used for profile generation can be clearly traced back to the specific prompt template and source on which it was based, enhancing the interpretability and credibility of the profile generation process.

[0039] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0040] Please see Figure 5 As shown, Figure 5 A flowchart illustrating the user profile construction method provided in the second embodiment of the present invention includes the following steps S210-S280: S210: Obtain historical multi-source data and its feature annotations for historical users, and classify the historical multi-source data according to the data classification system to determine the data type and business attribute tags of each data.

[0041] In this step, feature annotation can be either real feature values ​​obtained from historical feature extraction tasks on historical multi-source data, or feature values ​​directly annotated by domain experts on historical multi-source data. This annotation can guide the construction of the prompt template and serve as a benchmark for evaluating the accuracy of feature extraction in the prompt template. For example, in a financial scenario, feature annotation could be a credit score determined based on the user's final repayment performance; in a medical scenario, it could be a disease risk level determined based on the final diagnosis.

[0042] S220: Based on data of different data types, business attribute tags and their feature annotations, generate a prompt template containing corresponding data field placeholders and feature extraction task descriptions, and associate the prompt template, data type and business attribute tags.

[0043] In this step, for data with similar data types and business attribute tags, a natural language task description (i.e., feature extraction task description) that can guide the extraction of the feature is generated in reverse based on its corresponding feature annotation, and a prompt template containing data field placeholders is formed, and the association between the prompt template and the data type and business attribute tags is established.

[0044] S230: Gather all prompt templates to form the prompt template library.

[0045] Furthermore, after constructing the prompt template library, the historical accuracy rate of each prompt template in performing its feature extraction task can be obtained, and the current weight value of the prompt template can be calculated based on the historical accuracy rate to obtain the dynamic weight associated with the prompt template, i.e., according to the formula... The dynamic weight of the prompt template is calculated. ,in, This represents the historical accuracy rate of performing feature extraction tasks. This step involves adjusting parameters to control the weights' sensitivity to changes in accuracy. Based on this step, each prompt template can also be assigned a corresponding dynamic weight based on historical accuracy, facilitating subsequent sorting and retrieval of prompt templates from the prompt template library.

[0046] Based on the above steps S210-S230, it can be seen that the prompt templates are designed with different feature extraction tasks for different business scenarios. That is, the business logic of different business scenarios is encapsulated in a flexibly configurable prompt template library. When the business scenario changes, business experts only need to design or adjust the corresponding prompt templates to guide the extraction of new feature dimensions, which greatly improves the efficiency of profile analysis and generation.

[0047] S240: Obtain multi-source data from users and classify the multi-source data according to a preset data classification system to obtain classification results. The classification results include at least the data type and business attribute tags of each data.

[0048] This step is similar to step S110, and will not be described again here.

[0049] S250: Based on the classification results, match and call one or more prompt templates from the pre-built prompt template library.

[0050] This step is similar to step S120, and will not be described again here.

[0051] S260: Fill the data field placeholders of the called prompt template with the corresponding multi-source data of the user to generate the feature extraction prompt set of the user.

[0052] This step is similar to step S130, and will not be described again here.

[0053] S270: Parse and execute each feature extraction prompt in the feature extraction prompt set to extract features according to the feature extraction task description and fill data in the feature extraction prompt, and fuse all extracted features to obtain fused features.

[0054] This step is similar to step S140, and will not be described again here.

[0055] S280: Based on the fusion features, construct a multi-dimensional profile of the user.

[0056] This step is similar to step S150, and will not be described again here.

[0057] As can be seen, the above solution, by constructing a dynamically adjustable prompt template library, enables the rapid and accurate extraction of user features from structured and unstructured text data. Furthermore, the prompt template library can be dynamically updated according to business needs, supporting rapid scenario adaptation. For business scenarios such as personalized diagnosis and treatment in smart healthcare and loan risk assessment in fintech, it can intelligently match pre-built prompt templates based on data type and business attributes. The prompt templates guide the targeted feature extraction of relevant data (including structured and unstructured data) from multiple sources of user data, fully utilizing unstructured text data to make the profile more closely reflect the user's true state. It also achieves guided feature extraction, providing rapid and targeted support for different business scenarios. This enhances the depth and credibility of the profile while adapting to rapidly changing business scenarios. Moreover, each feature ultimately used for profile generation can be clearly traced back to the specific prompt template and source upon which its generation was based, enhancing the interpretability and credibility of the profile generation process.

[0058] In one embodiment, a user profile building apparatus is provided, such as... Figure 6 As shown, the user profile building device includes a classification module 301, a template matching module 302, a prompt generation module 303, a feature extraction module 304, and a profile building module 305. Detailed descriptions of each functional module are as follows: The classification module 301 is used to acquire multi-source data from users and classify the multi-source data according to a preset data classification system to obtain classification results. The classification results include at least the data type and business attribute tags of each data. The multi-source data includes structured data and unstructured text data. The template matching module 302 is used to match and call one or more prompt templates from a pre-built prompt template library based on the classification results; wherein each prompt template includes data field placeholders and a feature extraction task description, and is associated with data type and business attribute tags; The prompt generation module 303 is used to fill the data field placeholders of the prompt template called with the corresponding multi-source data of the user, and generate the feature extraction prompt set of the user; The feature extraction module 304 is used to parse and execute each feature extraction prompt in the feature extraction prompt set, to extract features according to the feature extraction task description and fill data in the feature extraction prompt, and to fuse all the extracted features to obtain fused features; The profile building module 305 is used to build a multi-dimensional profile of the user based on the fused features.

[0059] In one embodiment, the template matching module 302 is specifically used for: Based on the data type and business attribute tags of each data in the multi-source data, all associated prompt templates are retrieved from the pre-built prompt template library; wherein, the prompt templates are associated with dynamic weights. The prompt templates are sorted according to their dynamic weights, and one or more prompt templates with dynamic weights higher than a preset threshold are selected for invocation.

[0060] In one embodiment, the user profile building apparatus further includes a feedback optimization module 306, which is specifically used for: Obtain actual feedback information on the business decision results based on the constructed multi-dimensional profile; and adjust the dynamic weights and / or feature extraction task descriptions of the prompt templates in the prompt template library according to the actual feedback information.

[0061] In one embodiment, the user profile building apparatus further includes a template building module 307, wherein, The classification module 301 is also used to: acquire historical multi-source data of historical users and their feature annotations, and classify the historical multi-source data according to the data classification system to determine the data type and business attribute tags of each data. The template construction module 307 is specifically used to: generate prompt templates containing corresponding data field placeholders and feature extraction task descriptions based on data of different data types, business attribute tags and their feature annotations, and associate the prompt templates, data types and business attribute tags; and collect all prompt templates to form the prompt template library.

[0062] In one embodiment, the feedback optimization module 306 is further specifically used for: Obtain the historical accuracy of each prompt template in performing its feature extraction task; The current weight value of the prompt template is calculated based on the historical accuracy, thus obtaining the dynamic weight of the prompt template.

[0063] In one embodiment, the feature extraction module 304 is specifically used for: Analyze each feature extraction prompt in the feature extraction prompt set, and identify the data type of the data filled in the data field placeholder of the feature extraction prompt and the corresponding feature extraction task; Based on the feature extraction task, the corresponding analysis model is invoked to extract features according to the data type of the filled data.

[0064] In one embodiment, the feature extraction module 304 is further specifically used for: When the data type of the data being filled is structured data, a statistical analysis model or a machine learning model is invoked to extract numerical features. When the data type of the filling data is unstructured text data, a pre-trained large language model is used for text feature extraction.

[0065] This invention provides a user profile construction device that can classify acquired multi-source user data, determine the data type and business attribute tags of each data point, and then, based on the data type and business attribute tags, match and call one or more prompt templates from a prompt template library, each including data field placeholders and a feature extraction task description. The device then fills the corresponding data into the data field placeholders of the called prompt templates, generating a feature extraction prompt set for the user. The device parses and executes each feature extraction prompt in the set to extract features based on the feature extraction task description and the filled data. All extracted features are then fused to obtain a fused feature. Finally, the fused feature is used to... Features are used to construct a multi-dimensional profile of the user. During operation, the system can intelligently match pre-built prompt templates based on data type and business attributes. The prompt templates guide the extraction of structured and unstructured features from the current user's multi-source data, making full use of unstructured text data to make the profile more closely resemble the user's true state. It also achieves guided feature extraction, which can quickly provide targeted support according to different business scenarios. While improving the depth and credibility of the profile, it can also adapt to rapidly changing business scenarios. Furthermore, each feature used to generate the profile can be clearly traced back to the specific prompt template and source on which it was generated, enhancing the interpretability and credibility of the profile generation process.

[0066] Specific limitations regarding the user profile building device can be found in the limitations of the user profile building method described above, and will not be repeated here. Each module in the aforementioned user profile building device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0067] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a user profile construction method on the server side.

[0068] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the client-side functions or steps of a user profile building method.

[0069] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: The system acquires multi-source data from users and classifies the multi-source data according to a preset data classification system to obtain classification results. The classification results include at least the data type and business attribute tags of each data. The multi-source data includes structured data and unstructured text data. Based on the classification results, one or more prompt templates are matched and called from a pre-built prompt template library; each prompt template includes data field placeholders and a description of the feature extraction task, and is associated with data type and business attribute tags; The user's corresponding multi-source data is filled into the data field placeholders of the invoked prompt template to generate the user's feature extraction prompt set; Each feature extraction prompt in the feature extraction prompt set is parsed and executed to extract features based on the feature extraction task description and fill data in the feature extraction prompt, and all extracted features are fused to obtain fused features; Based on the fusion features, a multi-dimensional profile of the user is constructed.

[0070] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: The system acquires multi-source data from users and classifies the multi-source data according to a preset data classification system to obtain classification results. The classification results include at least the data type and business attribute tags of each data. The multi-source data includes structured data and unstructured text data. Based on the classification results, one or more prompt templates are matched and called from a pre-built prompt template library; each prompt template includes data field placeholders and a description of the feature extraction task, and is associated with data type and business attribute tags; The user's corresponding multi-source data is filled into the data field placeholders of the invoked prompt template to generate the user's feature extraction prompt set; Each feature extraction prompt in the feature extraction prompt set is parsed and executed to extract features based on the feature extraction task description and fill data in the feature extraction prompt, and all extracted features are fused to obtain fused features; Based on the fusion features, a multi-dimensional profile of the user is constructed.

[0071] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0072] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0073] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0074] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for constructing user profiles, characterized in that, include: The system acquires multi-source data from users and classifies the multi-source data according to a preset data classification system to obtain classification results. The classification results include at least the data type and business attribute tags of each data. The multi-source data includes structured data and unstructured text data. Based on the classification results, one or more prompt templates are matched and called from the pre-built prompt template library; wherein each prompt template contains data field placeholders and a description of the feature extraction task, and is associated with data type and business attribute tags; The user's corresponding multi-source data is filled into the data field placeholders of the invoked prompt template to generate the user's feature extraction prompt set; Each feature extraction prompt in the set of feature extraction prompts is parsed and executed to extract features based on the feature extraction task description and fill data in the feature extraction prompts, and all extracted features are fused to obtain fused features; Based on the fusion features, a multi-dimensional profile of the user is constructed.

2. The user profile construction method as described in claim 1, characterized in that, The step of matching and calling one or more prompt templates from a pre-built prompt template library based on the classification results includes: Based on the data type and business attribute tags of each data in the multi-source data, all associated prompt templates are retrieved from the pre-built prompt template library; wherein, the prompt templates are associated with dynamic weights. The prompt templates are sorted according to their dynamic weights, and one or more prompt templates with dynamic weights higher than a preset threshold are selected for invocation.

3. The user profile construction method as described in claim 2, characterized in that, The user profile construction method also includes: Obtain actual feedback information on the business decision-making results based on the constructed multi-dimensional profile; Based on the actual feedback information, adjust the dynamic weights and / or feature extraction task descriptions of the prompt templates in the prompt template library.

4. The user profile construction method as described in claim 1, characterized in that, Before matching and calling one or more prompt templates from a pre-built prompt template library based on the classification results, the process further includes: Acquire historical multi-source data and its feature annotations of historical users, and classify the historical multi-source data according to the data classification system to determine the data type and business attribute tags of each data. Based on data of different data types, business attribute tags and their feature annotations, generate a prompt template containing corresponding data field placeholders and feature extraction task descriptions, and associate the prompt template, data type and business attribute tags; All prompt templates are compiled to form the prompt template library.

5. The user profile construction method as described in claim 4, characterized in that, After collecting all the prompt templates to form the prompt template library, the process further includes: Obtain the historical accuracy of each prompt template in performing its feature extraction task; The current weight value of the prompt template is calculated based on the historical accuracy, thus obtaining the dynamic weight of the prompt template.

6. The user profile construction method as described in claim 1, characterized in that, The step of parsing and executing each feature extraction prompt in the feature extraction prompt set to perform feature extraction based on the feature extraction task description and filler data in the feature extraction prompt includes: Analyze each feature extraction prompt in the feature extraction prompt set, and identify the data type of the data filled in the data field placeholder of the feature extraction prompt and the corresponding feature extraction task; Based on the feature extraction task, the corresponding analysis model is invoked to extract features according to the data type of the filled data.

7. The user profile construction method as described in claim 6, characterized in that, The feature extraction based on the data type of the filled data, by calling the corresponding analysis model, includes: When the data type of the data being filled is structured data, a statistical analysis model or a machine learning model is invoked to extract numerical features. When the data type of the filling data is unstructured text data, a pre-trained large language model is used for text feature extraction.

8. A user profile building device, characterized in that, include: The classification module is used to acquire multi-source data from users and classify the multi-source data according to a preset data classification system to obtain classification results. The classification results include at least the data type and business attribute tags of each data. The multi-source data includes structured data and unstructured text data. The template matching module is used to match and call one or more prompt templates from a pre-built prompt template library based on the classification results; wherein each prompt template contains data field placeholders and a feature extraction task description, and is associated with data type and business attribute tags; The prompt generation module is used to fill the data field placeholders of the prompt template called with the corresponding multi-source data of the user, and generate the feature extraction prompt set of the user; The feature extraction module is used to parse and execute each feature extraction prompt in the feature extraction prompt set, to extract features according to the feature extraction task description and fill data in the feature extraction prompt, and to fuse all extracted features to obtain fused features; The profile building module is used to build a multi-dimensional profile of the user based on the fused features.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the user profile construction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the user profile construction method as described in any one of claims 1 to 7.

Images