Financial service decision method and device, electronic device and storage medium

Abstract

The application discloses a kind of financial service decision-making method and its device, electronic equipment and storage medium, it is related to artificial intelligence technical field or other related fields, the method comprises: using multiple collaborative intelligent agents, service request is parsed, and service entity is obtained;Business entity is mapped into target business node in financial knowledge graph using financial knowledge graph, and structured query intent is generated accordingly;Data filtering conditions are generated according to data authority information, and prompt words are constructed based on structured query intent and data filtering conditions;The prompt word is input to script generation intelligent agent, and data query script is generated by script generation intelligent agent, and data query script is executed, and target structured data is acquired;The financial business decision-making strategy is generated in combination with the causal effect side of financial knowledge graph and target structured data.The application solves the technical problems that the data analysis means used by financial institutions in the related art cannot meet the real-time and precision requirements of financial services.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology or other related technical fields. Specifically, it relates to a financial business decision-making method and apparatus, electronic device and storage medium. Background Technology

[0002] In the daily operations and management of financial institutions, data analytics is a key element supporting business decisions, optimizing customer experience, and improving operational efficiency. In the context of increasingly fierce market competition and digital transformation, data analytics has transcended its role as a mere technical support. With the exponential growth of financial data, the ability to extract necessary business data from massive, multi-source, and unstructured data determines whether financial institutions can seize opportunities, mitigate risks, and achieve refined operations in a rapidly changing market environment.

[0003] In related technologies, financial institutions generally rely on traditional data processing tools and reporting systems for data analysis, such as spreadsheet software, traditional business intelligence (BI) reporting tools, and dedicated data query platforms. These business data processing methods have significant technical limitations in data acquisition, analysis, and strategy execution, making it difficult to meet the growing demands of financial businesses for real-time performance, accuracy, and automation.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This invention provides a financial business decision-making method, apparatus, electronic device, and storage medium to at least address the technical problem that data analysis methods used by financial institutions are insufficient to meet the real-time and accuracy requirements of financial business.

[0006] According to one aspect of the present invention, a financial business decision-making method is provided, comprising: receiving a business request from a financial business terminal and parsing the business request to obtain a set of business entities; employing an intent parsing agent to map the business entities in the set of business entities to target business nodes in the financial knowledge graph using a pre-constructed financial knowledge graph, and generating a structured query intent based on the target business nodes; obtaining data permission information from the financial business terminal through a security and compliance agent, generating data filtering conditions based on the data permission information, and constructing prompt words based on the structured query intent and the data filtering conditions; inputting the prompt words into a script generation agent, generating a data query script through the script generation agent, and executing the data query script to obtain target structured data; and generating a financial business decision-making strategy by combining the causal effect edges of the financial knowledge graph and the target structured data.

[0007] Further, the steps of receiving business requests from the financial business terminal and parsing the business requests to obtain a set of business entities include: extracting natural language text from the business requests; performing word segmentation on the natural language text to extract a sequence of business keywords; inputting the sequence of business keywords into a pre-trained named entity recognition model to identify the types of financial entities in the sequence of business keywords and generate an initial entity list; performing disambiguation processing on the initial entity list based on predefined business context rules to eliminate polysemous word confusion and obtain an initial set of business entities; and performing matching verification between each business entity in the initial set of business entities and a pre-set financial business terminology dictionary, and outputting a set of business entities that have passed the verification.

[0008] Furthermore, the step of mapping business entities in the business entity set to target business nodes in the financial knowledge graph using a pre-constructed financial knowledge graph, and generating structured query intents based on the target business nodes, includes: traversing the financial knowledge graph, finding target business nodes with the same identifier as each business entity in the business entity set, and mapping the business entities to target business nodes in the financial knowledge graph based on the same identifier; obtaining attribute information of the business nodes, wherein the attribute information includes indicator type, dimension label, and associated calculation logic formula; decomposing each target business node into at least one sub-node according to the calculation logic formula, and using the sub-node as a sub-unit of the query task; and performing topological sorting of the sub-units according to the dependency order in the calculation logic formula to generate a structured query intent link containing the execution order.

[0009] Furthermore, the step of obtaining data permission information of the financial business terminal through a security and compliance intelligent agent and generating data filtering conditions based on the data permission information includes: parsing the authentication token of the current session and extracting the user identifier and organizational structure level of the financial business terminal; querying the user permission database based on the user identifier and organizational structure level to obtain the data access permission policy corresponding to the financial business terminal; determining the data filtering conditions based on the data access permission policy, and encapsulating the data filtering conditions into data constraint instructions, wherein the data filtering conditions include at least: field values ​​of the filter row.

[0010] Further, the step of inputting the prompt word into the script generation agent and generating a data query script through the script generation agent includes: inputting the prompt word into the script generation agent and generating a data query script for the target database through the script generation agent; pre-executing the data query script in a sandbox environment and capturing execution information generated during the execution process, wherein the execution information includes at least one of the following: error logs or exception stack information; generating a correction prompt word based on the execution information and sending the correction prompt word back to the script generation agent to regenerate the data query script, until the data query script is successfully pre-executed to obtain the final data query script.

[0011] Furthermore, the step of generating a financial business decision-making strategy by combining the causal effect edges of the financial knowledge graph and the target structured data includes: extracting the business indicator values ​​corresponding to the target business nodes from the target structured data, comparing the business indicator values ​​with historical benchmark values, and calculating the magnitude of change of the business indicator values; traversing the causal effect edges associated with the target business nodes in the financial knowledge graph, and obtaining the dynamic weight coefficient corresponding to each causal effect edge; calculating the contribution of the upstream nodes corresponding to the target business nodes to the indicator changes based on the magnitude of change and the dynamic weight coefficient, and filtering out root cause nodes whose contribution exceeds a preset contribution threshold; determining abnormal root causes based on the root cause nodes, querying a preset mapping matrix based on the abnormal root causes, matching the target execution action corresponding to the abnormal root causes, and generating a financial business decision-making strategy based on the target execution action.

[0012] Furthermore, after acquiring the target structured data, the method further includes: converting the target structured data into visualized data; and rendering the visualized data onto the interactive interface of the financial business terminal.

[0013] According to another aspect of the present invention, a financial business decision-making apparatus is also provided, comprising: a receiving unit, configured to receive a business request from a financial business terminal and parse the business request to obtain a set of business entities; a mapping unit, configured to use an intent parsing agent to map the business entities in the set of business entities to target business nodes in the financial knowledge graph using a pre-built financial knowledge graph, and generate a structured query intent based on the target business nodes; a construction unit, configured to obtain data permission information from the financial business terminal through a security and compliance agent, generate data filtering conditions according to the data permission information, and construct prompt words based on the structured query intent and the data filtering conditions; an execution unit, configured to input the prompt words to a script generation agent, generate a data query script through the script generation agent, and execute the data query script to obtain target structured data; and a generation unit, configured to combine the causal effect edges of the financial knowledge graph and the target structured data to generate a financial business decision-making strategy.

[0014] Further, the receiving unit includes: a first extraction module, used to extract natural language text from the business request, perform word segmentation on the natural language text, and extract a sequence of business keywords; a first recognition module, used to input the sequence of business keywords into a pre-trained named entity recognition model, recognize the financial entity types in the sequence of business keywords, and generate an initial entity list; a first processing module, used to perform disambiguation processing on the initial entity list based on predefined business context rules, eliminate polysemous word confusion, and obtain an initial set of business entities; and a first verification module, used to match and verify each business entity in the initial set of business entities with a pre-set financial business terminology dictionary, and output a set of business entities that have passed the verification.

[0015] Further, the mapping unit includes: a first traversal module, used to traverse the financial knowledge graph, find target business nodes with the same identifier as each business entity in the business entity set, and map the business entities to target business nodes in the financial knowledge graph based on the same identifier; a first acquisition module, used to acquire attribute information of the business nodes, wherein the attribute information includes indicator type, dimension label and associated calculation logic formula; a first decomposition module, used to decompose each target business node into at least one child node according to the calculation logic formula, and use the child node as a sub-unit of the query task; and a first generation module, used to perform topological sorting of the sub-units according to the dependency order in the calculation logic formula, and generate a structured query intent link containing the execution order.

[0016] Furthermore, the construction unit includes: a first parsing module, used to parse the authentication token of the current session and extract the user identifier and organizational structure level of the financial business terminal; a first query module, used to query the user permission database based on the user identifier and organizational structure level to obtain the data access permission policy corresponding to the financial business terminal; and a first determining module, used to determine the data filtering conditions based on the data access permission policy and encapsulate the data filtering conditions into data constraint instructions, wherein the data filtering conditions include at least: field values ​​of the filter row.

[0017] Further, the execution unit includes: a second generation module, used to input the prompt words to the script generation agent, and generate a data query script for the target database through the script generation agent; a first execution module, used to pre-execute the data query script in a sandbox environment, and capture execution information generated during the execution process, wherein the execution information includes at least one of the following: error logs or exception stack information; and a first correction module, used to generate correction prompt words based on the execution information, and send the correction prompt words back to the script generation agent to regenerate the data query script, until the data query script is successfully pre-executed to obtain the final data query script.

[0018] Further, the generation unit includes: a first comparison module, used to extract the business indicator value corresponding to the target business node from the target structured data, compare the business indicator value with the historical benchmark value, and calculate the change amplitude of the business indicator value; a second acquisition module, used to traverse the causal effect edges associated with the target business node in the financial knowledge graph, and acquire the dynamic weight coefficient corresponding to each causal effect edge; a first calculation module, used to calculate the contribution of the upstream node corresponding to the target business node to the indicator change based on the change amplitude and the dynamic weight coefficient, and filter out the root cause nodes whose contribution exceeds a preset contribution threshold; and a third generation module, used to determine the abnormal root cause based on the root cause node, query a preset mapping matrix based on the abnormal root cause, match the target execution action corresponding to the abnormal root cause, and generate a financial business decision strategy based on the target execution action.

[0019] Furthermore, the financial business decision-making device also includes: a first conversion module for converting the target structured data into visualized data; and a first rendering module for rendering the visualized data onto the interactive interface of the financial business terminal.

[0020] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform any of the above-described financial business decision-making methods.

[0021] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement any of the above-described financial business decision-making methods.

[0022] According to another aspect of the present invention, a computer program product is also provided, the computer program product comprising a computer program, wherein the computer program, when executed by a processor, implements any of the above-described financial business decision-making methods.

[0023] In this application, the following steps are performed: receiving business requests from the financial business end and parsing the business requests to obtain a set of business entities; using a pre-built financial knowledge graph to map the business entities in the set of business entities to target business nodes in the financial knowledge graph; generating structured query intents based on the target business nodes; obtaining data permission information from the financial business end; generating data filtering conditions based on the data permission information; constructing prompt words based on the structured query intents and data filtering conditions; inputting the prompt words into a script generation agent; generating a data query script through the script generation agent; executing the data query script; obtaining target structured data; and generating a financial business decision-making strategy by combining the causal effect edges of the financial knowledge graph and the target structured data.

[0024] This application employs multiple collaborative intelligent agents, utilizing a pre-built financial knowledge graph to accurately map these entities to target business nodes, thereby generating structured query intents. It also acquires data permission information in real time and generates data filtering conditions accordingly. These two elements are combined to construct prompts containing security constraints, which are input into a script generation agent to generate and execute data query scripts, obtaining the target structured data required by the business. Furthermore, it combines causal effect edges in the financial knowledge graph with the target data for deep correlation analysis, automatically generating financial business decision-making strategies. This achieves end-to-end intelligent processing from natural language requests to structured data insights and automated decision-making, lowering the technical threshold for data analysis, improving operational efficiency, and leveraging the causal reasoning capabilities of the knowledge graph to achieve deep attribution analysis and automatic strategy recommendation. This shortens the operational decision-making cycle and improves the real-time and accuracy of financial business operations, thus solving the technical problem that data analysis methods used by financial institutions often fail to meet the real-time and accuracy requirements of financial business. Attached Figure Description

[0025] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0026] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing financial business decision-making methods is shown.

[0027] Figure 2 This is a flowchart of an optional financial business decision-making method according to an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of an optional financial business decision-making process according to an embodiment of the present invention;

[0029] Figure 4 This is a schematic diagram of an optional financial business decision-making device according to an embodiment of the present invention;

[0030] Figure 5 This is a hardware structure block diagram of an electronic device (or mobile device) for executing a financial business decision-making method according to an embodiment of the present invention. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0033] It should be noted that the financial business decision-making method and apparatus in this application can be used in the field of artificial intelligence technology to make financial business decisions based on knowledge graphs and artificial intelligence, and can also be used in any field other than artificial intelligence technology to make financial business decisions based on knowledge graphs and artificial intelligence. This application does not limit the application field of the financial business decision-making method and apparatus.

[0034] It should be noted that the information collected in this application (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data all comply with relevant laws, regulations, and standards, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding access points are provided for users to choose to authorize or refuse. For example, interfaces are set up between this system and relevant users or organizations, providing users with corresponding access points to choose to agree to or refuse automated decision-making results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0035] The following embodiments of the present invention can be applied to various financial business decision-making systems / applications / devices. The present invention is based on knowledge graphs and intelligent agents to automatically parse business intent, generate structured query statements, and thereby obtain the structured data required by the business. Simultaneously, it leverages the causal reasoning capabilities of knowledge graphs to achieve in-depth analysis and automated decision-making for business decisions, significantly shortening the operational decision-making cycle and improving the real-time performance, accuracy, and automation level of financial business operations.

[0036] The present invention will now be described in detail with reference to various embodiments.

[0037] Example 1

[0038] According to an embodiment of the present invention, an embodiment of a financial business decision-making method 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.

[0039] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing financial business decision-making methods is shown. Figure 1As shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0040] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0041] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the financial business decision-making method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned financial business decision-making method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0042] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0043] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0044] Under the aforementioned operating environment, this application provides the following: Figure 2 The financial business decision-making method shown is implemented by a financial business decision-making system. This invention can be applied to business scenarios such as intelligent analysis of financial businesses, customer segmentation, and product recommendation, and financial business decisions are made through the deployment of multiple intelligent agents.

[0045] Figure 2 This is a flowchart of an optional financial business decision-making method according to an embodiment of the present invention, such as... Figure 2 As shown, the method includes the following steps:

[0046] Step S201: Receive business requests from the financial business side and parse the business requests to obtain a set of business entities.

[0047] In step S201 above, the system receives an application programming interface (API) request from the front-end interface or mobile terminal of the financial business. This request is typically in the form of natural language text, such as an operator's input: "Query the top ten financial outlets in region A with credit card delinquency rates higher than the average yesterday." Subsequently, the system calls the built-in natural language parsing module to perform semantic analysis on the received text, thereby identifying the key business elements contained in the text, i.e., business entities. Business entities include, but are not limited to, time dimensions, spatial dimensions, business indicators, and object attributes. By extracting and deduplicating key elements, a set of business entities is constructed. This set not only retains the core semantic information in the original request but also eliminates redundant noise in colloquial expressions, laying the foundation for mapping these entities to standardized knowledge graph nodes in subsequent steps.

[0048] Based on the above steps, the contradiction between vague and unstructured questions from business personnel and rigid database fields is resolved, achieving an initial conversion from natural language to machine-readable semantics, and ensuring that subsequent data processing can accurately understand the user's true intentions.

[0049] In one optional embodiment, the steps of receiving a business request from a financial business terminal and parsing the business request to obtain a set of business entities include: extracting natural language text from the business request; performing word segmentation on the natural language text to extract a sequence of business keywords; inputting the sequence of business keywords into a pre-trained named entity recognition model to identify the types of financial entities in the sequence of business keywords and generate an initial entity list; performing disambiguation processing on the initial entity list based on predefined business context rules to eliminate polysemous word confusion and obtain an initial set of business entities; and performing matching verification between each business entity in the initial set of business entities and a pre-set financial business terminology dictionary, and outputting a set of business entities that have passed the verification.

[0050] Furthermore, the process of receiving business requests from the financial business side and parsing them to obtain a set of business entities aims to transform fuzzy natural language queries into precise structured semantic units through a multi-layered semantic processing mechanism, thereby eliminating ambiguity in natural language and matching the standard data model within the system.

[0051] Specifically, firstly, the system extracts the core natural language text from the received business requests, removes non-semantic characters generated by interface interactions, and then performs fine-grained word segmentation on the text. Based on the language habits of the financial field, the continuous text is divided into the smallest units with independent semantics, thereby extracting a sequence of business keywords that can reflect the user's intent. This transforms unstructured text into a computable sequence, providing standardized input data for subsequent entity recognition.

[0052] Subsequently, the system inputs the generated sequence of business keywords into a pre-trained named entity recognition model. This model, trained on a large corpus of financial business data, possesses the ability to identify entities in specific domains. The model analyzes each keyword in the sequence to determine its corresponding financial entity type. For example, it distinguishes between user-level entities (e.g., customers, delinquency rates, indicators, regions A, etc.) and generates an initial entity list containing entity type tags, thereby automatically capturing text using machine learning algorithms.

[0053] However, polysemy is common in natural language. Therefore, the system introduces predefined business context rules to disambiguate the initial entity list. This rule base stores vocabulary usage preferences and contextual dependencies under different business scenarios. Based on the current query scenario and the surrounding context, the system corrects polysemous words in the initial entity list, eliminating recognition errors caused by semantic confusion, thus obtaining a more semantically accurate initial set of business entities. This ensures that the extracted entities have a unique and correct meaning in a specific business scenario, preventing subsequent data query deviations caused by ambiguity.

[0054] Finally, to ensure that the extracted entities can be recognized and executed by the downstream data query engine, the system performs a one-by-one matching and verification of each entity in the initial business entity set against a pre-built financial business terminology dictionary. This dictionary stores the standard definitions of all legal indicator names, dimension labels, and attribute fields in the system's underlying data model. Through verification, the system filters out invalid entities that, although syntactically recognized as entities, do not have corresponding fields in the current data model. Simultaneously, it standardizes the naming format of entities to ensure complete consistency with database fields. The system outputs the set of business entities that have passed verification, establishing a precise mapping relationship between natural language terms and underlying data fields, ensuring that subsequently generated query statements can be executed successfully and avoiding query failures due to field name mismatches.

[0055] Furthermore, business context rules include static and dynamic rules. Static rules maintain a vector database. For a specific entity, the cosine similarity between it and standard entities in the vector database is calculated, and the specific entity is mapped to the standard entity based on the cosine similarity. For example, the cosine similarity between Customer ID and CUST_ID is calculated to map Customer ID to CUST_ID. Dynamic rules analyze the distribution of field values ​​in real time and use KL divergence to detect distribution differences. If the distribution difference exceeds the KL divergence threshold, dynamic mapping adjustment is triggered; for other invalid entities, manual correction is used, and the corrected results are recorded in the vector database.

[0056] In addition, embodiments of the present invention also introduce a dialect mapping mechanism to integrate common dialect fields, generate a mapping dictionary, or train a named entity recognition model to automatically recognize dialect fields.

[0057] Through a series of rigorous processing steps, including word segmentation, named entity recognition, context disambiguation, and terminology dictionary matching and verification, the accuracy and robustness of extracting business entities from natural language requests can be significantly improved.

[0058] Step S202: An intent parsing agent is used to map the business entities in the business entity set to target business nodes in the financial knowledge graph using a pre-built financial knowledge graph, and a structured query intent is generated based on the target business nodes.

[0059] In step S202 above, the intent parsing agent performs intent parsing, effectively solving the semantic understanding bias problem caused by the lack of domain knowledge in traditional text analysis by utilizing the structured knowledge reserves of the financial knowledge graph. By mapping business entities to target business nodes, the system not only retains the precise meaning of the entities but also inherits the business logic relationships contained in the graph. The structured query intent generated based on the target business nodes provides a standardized and unambiguous operational blueprint for subsequent access control, script generation, and data execution, significantly improving the accuracy and efficiency of data queries and laying a key foundation for building an intelligent financial data analysis closed loop.

[0060] In one optional embodiment, the steps of mapping business entities in a business entity set to target business nodes in the financial knowledge graph using a pre-built financial knowledge graph, and generating structured query intents based on the target business nodes, include: traversing the financial knowledge graph, finding target business nodes with the same identifier as each business entity in the business entity set, and mapping business entities to target business nodes in the financial knowledge graph based on the same identifier; obtaining attribute information of the business nodes, wherein the attribute information includes indicator type, dimension label, and associated calculation logic formula; decomposing each target business node into at least one sub-node according to the calculation logic formula, and using the sub-node as a sub-unit of the query task; and performing topological sorting of the sub-units according to the dependency order in the calculation logic formula to generate a structured query intent link containing the execution order.

[0061] Furthermore, the process of mapping business entities in the business entity set to target business nodes in the financial knowledge graph using a pre-built financial knowledge graph, and generating structured query intents based on these target business nodes, aims to transform abstract business requirements into a sequence of executable computational tasks through precise semantic matching and logical decomposition. Specifically, this includes: First, the system traverses the financial knowledge graph stored in the graph database, which contains rich financial business metadata and relation definitions. It retrieves each business entity in the business entity set one by one, searching for target business nodes in the graph that share the same identifier as these entities. Through precise matching, it ensures that words in natural language accurately correspond to standard data model nodes defined internally by the system, thereby completing the mapping from the user's perspective to the system's perspective, eliminating semantic ambiguity, and establishing the data objects to be queried.

[0062] Subsequently, the system acquires the attribute information associated with these target business nodes. This attribute information specifically includes the metric type, dimension labels, and associated calculation logic formulas. The metric type defines whether the node is an atomic or composite metric, the dimension labels describe the perspective from which the data is observed, and the calculation logic formulas detail how the metric is composed of other basic metrics or dimensions through specific operators. This provides a basis for subsequent structured decomposition, ensuring the system can understand the business meaning and calculation path behind the metrics.

[0063] Based on the acquired calculation logic formula, the system breaks down each target business node into at least one child node, and uses these child nodes as sub-units of the query task. For example, if the target node is sales revenue, its calculation logic formula may involve sales volume multiplied by unit price. The system will then break down the sales revenue into two child nodes: sales volume and unit price. This breakdown process reduces complex compound queries to the most basic atomic query units, decreasing the complexity of subsequent script generation and ensuring that each sub-unit corresponds to a specific field or dimension in the underlying data table.

[0064] Finally, the system performs topological sorting of the decomposed sub-units according to the dependency order in the computational logic formula, generating a structured query intent link that includes the execution order. Topological sorting is a linear sorting algorithm that ensures that all its predecessor dependent nodes have been computed before a node is computed. Through this sorting, the system generates a clear execution link, specifying the order of each sub-unit in the query process, and generating a structured query intent link that includes the execution order. This provides clear logical guidance for the subsequent script generation agent, ensuring that the generated query script conforms to data dependencies, avoiding execution errors, and thus improving the accuracy and efficiency of data querying.

[0065] By employing identifier matching, attribute acquisition, logical decomposition, and topological sorting, a deep transformation from business entities to structured query intent is achieved. This transforms complex business logic into an executable sequence of computational steps, ensuring the structural rationality and execution feasibility of subsequent data query tasks, and providing solid technical support for building an efficient and accurate closed loop for financial data analysis.

[0066] Step S203: Obtain data permission information from the financial business terminal through the security and compliance intelligent agent, generate data filtering conditions based on the data permission information, and construct prompt words based on the structured query intent and data filtering conditions.

[0067] In step S203 above, dynamic security constraints are executed by a dynamic security compliance intelligent agent. Data permission constraints are dynamically injected to ensure the security and compliance of data queries, while providing complete contextual input for subsequent large-scale model script generation. The static, hard-coded permission verification mechanism of traditional databases is dynamically and intelligently embedded into the natural language processing flow, ensuring data security isolation. By constructing prompts containing security constraints, not only is the accuracy of large-scale model generation scripts improved, avoiding query errors or data leakage risks due to permission omissions, but it also meets the stringent data compliance requirements of financial institutions while maintaining a simple user experience, providing crucial assurance for building a trustworthy and secure intelligent operating system.

[0068] In one optional embodiment, the steps of obtaining data permission information from the financial business terminal and generating data filtering conditions based on the data permission information include: parsing the authentication token of the current session, extracting the user identifier and organizational structure level of the financial business terminal; querying the user permission database based on the user identifier and organizational structure level to obtain the data access permission policy corresponding to the financial business terminal; determining the data filtering conditions based on the data access permission policy, and encapsulating the data filtering conditions into data constraint instructions, wherein the data filtering conditions include at least: field values ​​of the filter row.

[0069] Furthermore, the process of acquiring data access permission information from the financial business side and generating data filtering conditions aims to ensure that data query operations strictly comply with the security and compliance requirements of financial institutions through a refined identity recognition and permission mapping mechanism. Specifically, the system first parses the authentication token of the current session. This token is a security credential issued by the identity authentication center when a user logs into the system, used to prove the user's identity and the legitimacy of the session. By parsing this token, the system can accurately extract the user's identifier and organizational structure level from the financial business side. The purpose of parsing the authentication token and extracting key identity information is to establish a connection between the current operator and the system's permission system, providing an accurate index for subsequent permission queries and avoiding permission matching based on incorrect identities.

[0070] Subsequently, the system queries the user permission database based on the extracted user identifier and organizational hierarchy to obtain the data access permission policies corresponding to the financial business end. The user permission database stores detailed permission configurations for all users and their organizations, including the scope of accessible data, the types of operable metrics, and specific row-level data isolation rules. By querying this database, the system can dynamically obtain the specific data scope that the current user is authorized to access. Based on this, abstract organizational structure information is transformed into specific data access rules, enabling the granularity of permission control to be refined to the specific business data row.

[0071] The system determines data filtering conditions based on the acquired data access permission policies and encapsulates these conditions into data constraint instructions. Data filtering conditions are a set of rules used at the database query level to block unauthorized access to data; their core requirement is that they include at least the field values ​​of the filtered rows. Encapsulating data filtering conditions into data constraint instructions transforms the permission policies into machine-executable technical instructions, enabling them to be recognized by subsequent script-generating agents and embedded into the query logic. This process ensures that regardless of the user's query intent, the underlying query statement will automatically attach these filtering conditions, thereby achieving implicit and mandatory row-level data isolation.

[0072] By parsing authentication tokens, querying permission databases, and encapsulating data constraint instructions, dynamic acquisition and standardized expression of data permissions are achieved.

[0073] Step S204: Input the prompt words into the script generation agent, generate a data query script through the script generation agent, and execute the data query script to obtain the target structured data.

[0074] In step S204 above, the aim is to utilize script-generated intelligent agents to transform prompts containing business logic and security constraints into executable data query commands and obtain the final analysis results. This can improve the efficiency, accuracy, and automation level of data retrieval in financial business scenarios, providing crucial data support for building an intelligent operational decision-making closed loop.

[0075] In one optional embodiment, the step of inputting a prompt word into a script generation agent and generating a data query script through the script generation agent includes: inputting the prompt word into the script generation agent and generating a data query script for a target database through the script generation agent; pre-executing the data query script in a sandbox environment and capturing execution information generated during execution, wherein the execution information includes at least one of the following: error logs or exception stack information; generating a corrected prompt word based on the execution information and sending the corrected prompt word back to the script generation agent to regenerate the data query script, until the data query script pre-execution is successful, thereby obtaining the final data query script.

[0076] Furthermore, the process of inputting prompts into the script generation agent and iteratively optimizing the final data query script aims to address potential syntax errors or logical biases in the large model-generated code through sandbox execution and reflective agent-based pre-execution and reflection processing, ensuring the availability and stability of the query script in the production environment. Specifically, the system first inputs prompts containing business intent and security constraints into the script generation agent. This agent, based on a pre-trained large language model and combined with database metadata information, generates an initial data query script for the target database. This transforms the business requirements described in natural language into structured query instructions that can be executed by a computer, thus initially constructing the logical path for data retrieval.

[0077] Subsequently, the system pre-executes the generated initial data query script in an isolated sandbox environment. The sandbox environment is a test space logically isolated from the production database but with the same data structure, simulating the actual execution effect of the script without affecting real business data. During pre-execution, the system captures execution information generated in real time. This information includes at least error logs or exception stack traces. Error logs can specifically contain error messages such as syntax errors, missing table or field names, and insufficient permissions; exception stack traces provide the call chain and context details where the error occurred. Capturing execution information allows for quick identification of defects in the script generation process, providing accurate error feedback for subsequent corrections and avoiding resource waste caused by blind retries.

[0078] The system generates correction prompts based on the captured execution information. These prompts, supplemented with specific error messages and corrective guidelines, are sent back to the script generation agent. The agent then incorporates the new error information to reflect on its actions and refactor the code, regenerating the corrected data query script. This execution feedback loop guides the large model to automatically correct errors, achieving script self-healing.

[0079] The system repeats the above pre-execution and correction cycle until the data query script is successfully pre-executed in the sandbox environment, meaning no error logs or exception stack traces are generated, resulting in the final data query script. Based on this, only scripts that have undergone rigorous verification, are logically correct, and syntactically valid will be submitted to the production database for execution, thus ensuring the accuracy of subsequent data query results and the security of system operation.

[0080] By introducing sandbox pre-execution and an iterative correction mechanism based on execution information, the success rate and robustness of script generation can be improved.

[0081] Step S205: Combine the causal effect edges of the financial knowledge graph with the target structured data to generate financial business decision-making strategies.

[0082] In step S205 above, the process of generating financial business decision-making strategies by combining the causal effect edges of the financial knowledge graph with the target structured data is described. Causal effect edges refer to directed edges with weighted coefficients in the financial knowledge graph. This step aims to perform in-depth attribution analysis and intelligent decision-making on structured data, thereby achieving intelligent generation and automated execution of business decision-making strategies.

[0083] In one optional embodiment, the steps of generating a financial business decision-making strategy by combining causal effect edges of a financial knowledge graph and target structured data include: extracting business indicator values ​​corresponding to target business nodes from the target structured data, comparing the business indicator values ​​with historical benchmark values, and calculating the magnitude of change in the business indicator values; traversing the causal effect edges associated with the target business nodes in the financial knowledge graph and obtaining the dynamic weight coefficients corresponding to each causal effect edge; calculating the contribution of upstream nodes corresponding to the target business nodes to the indicator changes based on the magnitude of change and the dynamic weight coefficients, and filtering out root cause nodes whose contribution exceeds a preset contribution threshold; determining abnormal root causes based on the root cause nodes, querying a preset mapping matrix based on the abnormal root causes, matching the target execution actions corresponding to the abnormal root causes, and generating a financial business decision-making strategy based on the target execution actions.

[0084] Furthermore, by combining the causal effect edges of financial knowledge graphs with the process of generating financial business decision-making strategies from target structured data, the aim is to accurately pinpoint the root causes of business anomalies through quantitative attribution analysis and automatically generate executable operational strategies. Specifically, the system first extracts the business indicator values ​​corresponding to the target business nodes from the target structured data. These business indicator values ​​are the specific values ​​obtained by the system at the current moment, such as the current daily sales or conversion rate. Subsequently, the system compares the extracted business indicator values ​​with historical benchmark values, calculating the magnitude of change in the business indicator values. The historical benchmark values ​​can be the values ​​from the previous day, the same period last week, or a preset normal fluctuation range. Calculating the magnitude of change can quantify the degree of deviation of the current business status, clarify whether there are significant abnormal fluctuations, and provide quantitative triggering conditions for subsequent attribution analysis.

[0085] Subsequently, the system traverses the causal effect edges in the financial knowledge graph that are directly related to the target business node, obtaining the dynamic weight coefficient corresponding to each causal effect edge. The causal effect edge represents the influence relationship between upstream nodes and downstream nodes, while the dynamic weight coefficient is calculated based on historical time-series data using algorithms such as causal testing, reflecting the strength and stability of this influence relationship. This quantifies the potential influence weight of each upstream factor on changes in the target indicator, ensuring the scientific validity and timeliness of the attribution results.

[0086] The system calculates the contribution of upstream nodes corresponding to the target business node to the change in the indicator based on the magnitude of the change and dynamic weighting coefficients. This calculation process essentially distributes the actual change in the indicator to each upstream node according to causal weights, thereby estimating the specific contribution ratio of each upstream factor to the overall abnormal fluctuation. By calculating the contribution ratio, the system transforms ambiguous causes into specific responsibility ratios, giving the attribution results a mathematical basis. Based on this, the system filters out root cause nodes whose contribution exceeds a preset contribution threshold.

[0087] Once the root cause node is identified, the system recognizes it as the anomaly root cause, i.e., the fundamental reason for the abnormal business metrics. Based on this anomaly root cause, the system queries a pre-built mapping matrix, which stores the correspondence between various business anomaly scenarios and standard operational actions. The action mapping agent queries the mapping matrix to match the target execution action corresponding to the anomaly root cause. For example, if the root cause is a decrease in ad click-through rate, the matched action might be to optimize ad creatives or adjust delivery channels. Finally, the system generates a financial business decision-making strategy based on the matched target execution action. This strategy includes specific operational suggestions, parameter adjustment values, and expected effect evaluations. This forms a complete closed-loop solution, guiding operations personnel in subsequent business adjustments.

[0088] By quantifying changes in indicators, using dynamic causal weights for contribution analysis, and executing actions based on mapping matrix matching, the entire process of automated decision-making from data anomaly detection to strategy generation has been achieved.

[0089] In one optional embodiment, after acquiring the target structured data, the method further includes: converting the target structured data into visualized data; and rendering the visualized data onto the interactive interface of the financial business terminal.

[0090] Furthermore, after acquiring the target structured data, the system performs data visualization and interface rendering steps to transform abstract numerical results into intuitive business insights, improving users' information acquisition efficiency and interactive experience. Specifically, the system first transforms the target structured data into visualized data. This process includes cleaning, aggregating, and mapping the dimensions of the raw data, and selecting appropriate chart formats based on data type and business semantics. For example, time series data is converted into line charts to display trends, categorical data is converted into bar charts or pie charts to display proportions, and geographically distributed data is converted into map heatmaps to display regional differences.

[0091] Subsequently, the system renders the generated visualization data onto the interactive interface of the financial business side. The rendering process involves integrating visualization components with the front-end framework to ensure that charts can adapt to different terminal devices with a responsive layout and support dynamic interactive functions, such as hovering to display details and clicking to drill down for detailed information. This allows business personnel to directly view the analysis results on a unified platform without needing to switch to other tools or view raw data tables, achieving instant presentation of data insights and convenient consumption.

[0092] Through the above steps, business requests from the financial business side are received and parsed to obtain a set of business entities. A pre-built financial knowledge graph is used to map the business entities in the set to target business nodes in the graph. A structured query intent is generated based on these target business nodes to obtain data permission information from the financial business side. Data filtering conditions are generated based on the data permission information. Prompt words are constructed based on the structured query intent and data filtering conditions. These prompt words are input into a script generation agent, which generates and executes a data query script to obtain the target structured data. Finally, a financial business decision-making strategy is generated by combining the causal effect edges of the financial knowledge graph with the target structured data.

[0093] In this embodiment, a pre-constructed financial knowledge graph is used to accurately map these entities to target business nodes, thereby generating structured query intents. Data permission information is obtained in real time and data filtering conditions are generated accordingly. The two are combined to construct prompts containing security constraints, which are input into a script generation agent to generate and execute a data query script to obtain the target structured data required by the business. On this basis, the causal effect edges in the financial knowledge graph are further combined with the target data for deep correlation analysis, automatically generating financial business decision-making strategies. This realizes end-to-end intelligent processing from natural language requests to structured data insights to automated decision-making, reducing the technical threshold of data analysis, improving operational efficiency, and leveraging the causal reasoning capabilities of the knowledge graph to achieve deep attribution analysis of data and automatic strategy recommendation, shortening the operational decision-making cycle, and improving the real-time and accuracy of financial business operations. This solves the technical problem in related technologies where the data analysis methods used by financial institutions are difficult to meet the real-time and accuracy requirements of financial business.

[0094] The following describes in detail another optional implementation method.

[0095] Figure 3 This is a schematic diagram of an optional financial business decision-making process according to an embodiment of the present invention, such as... Figure 3 As shown, the financial business decision-making process includes:

[0096] Step 1: The user enters the original text;

[0097] Construct a dynamic causal indicator graph with a heterogeneous graph structure, which includes indicator nodes, dimension nodes, and action nodes, as well as computational dependency edges between nodes and dynamic weight coefficients. Causal effect edge, dynamic weight coefficient Causal testing algorithms based on historical time-series behavioral data are updated periodically, for example: .

[0098] It receives unstructured natural language query commands from users and maps the colloquial expressions in the unstructured natural language query commands to corresponding standard indicator nodes and dimension nodes through a dynamic causal indicator graph.

[0099] Step two: Perform intent parsing through the intent parsing agent;

[0100] The mapped standard indicator nodes are decomposed into a directed acyclic graph (DAG) for graph query tasks, including time-series trend query, multi-dimensional drill-down query, and causal chain tracing, using an intent parsing agent.

[0101] Step 3: Perform dynamic security compliance through a dynamic security compliance intelligent agent;

[0102] The security and compliance intelligent agent extracts the current user's session identifier to identify their data permission level, dynamically generates row-level filtering isolation conditions based on the data permission level, and injects the row-level filtering isolation conditions as mandatory constraints into the prompt words to generate prompt words with security constraints.

[0103] Step four: Generate the query statement by generating an intelligent agent through a script;

[0104] The prompt words with security constraints are input into the script generation agent to generate a structured query language script for the target distributed online analytical processing database.

[0105] Step 5: Collect execution results through sandbox execution and reflective agent, and provide self-healing feedback for problematic query statements;

[0106] The query script is pre-executed in an isolated sandbox environment using a sandbox execution and reflection agent. When execution fails, the error backtracking information of the execution stack is captured, and correction prompts are dynamically generated in conjunction with the metadata of the dynamic causal index graph. The correction prompts are dynamically sent back to the script generation agent for iterative self-healing correction until execution is successful and structured analysis data is output.

[0107] The prompt word template for iterative self-healing correction is represented as follows:

[0108] [System Context]: You are a database self-healing expert;

[0109] [Historical Error]: The previously generated SQL [SELECT channel, SUM(gmv) FROM t...] failed to execute;

[0110] [Database Feedback]: The error message is [Code:47.DB::Exception:Missing columns:'channel'];

[0111] [Schema Correction]: According to the retrieved graph, the standard column name is 'channel_id', and its associated dimension is 'channel_name';

[0112]

Task

[0113] To avoid endless iterations, a maximum number of iterations is set. Once the maximum number of iterations is reached, a manual processing flow is triggered, the application is submitted for manual review, and important business scenarios are manually annotated.

[0114] Step 6: Perform root cause analysis through the root cause analysis module;

[0115] Step 7: Execute the action mapping through the action mapping agent;

[0116] Step 8: Business Decisions.

[0117] Following the causal effect edges in the dynamic causal index graph, i.e., those with weighted coefficients The directed edges are used to perform drill-down attribution reasoning on structured analysis data to identify the core root cause nodes causing data anomalies. A preset indicator-action mapping matrix is ​​used to match target action nodes associated with the core root cause nodes, automatically assembling corresponding application programming interface (API) request messages. Simultaneously, a front-end interactive interface outputs visualized data charts, diagnostic reports, and a one-click action button. When a user triggers the one-click action button, and the parameters in the API call request message do not trigger the preset safety circuit breaker threshold, the API call gateway invokes an external digital system to execute the target action.

[0118] This invention, based on knowledge graphs and intelligent agents, automatically parses business intent, generates structured query statements, and thereby obtains the structured data required by the business. At the same time, it leverages the causal reasoning capabilities of knowledge graphs to achieve in-depth analysis and automated decision-making for business decisions, significantly shortening the operational decision-making cycle and improving the real-time performance, accuracy, and automation level of financial business operations.

[0119] The following is a detailed description with reference to another embodiment.

[0120] Example 2

[0121] The financial business decision-making device provided in this embodiment includes multiple implementation units, each of which corresponds to a specific implementation step in the above embodiment one. The specific implementation method and beneficial effects can be referred to the foregoing method embodiment, and will not be repeated here.

[0122] Figure 4 This is a schematic diagram of an optional financial business decision-making device according to an embodiment of the present invention, such as... Figure 4 As shown, the financial business decision-making device may include: a receiving unit 41, a mapping unit 42, a construction unit 43, an execution unit 44, and a generation unit 45, wherein,

[0123] The receiving unit 41 is used to receive business requests from the financial business terminal and parse the business requests to obtain a set of business entities.

[0124] The mapping unit 42 is used to employ an intent parsing agent to map business entities in the business entity set to target business nodes in the financial knowledge graph using a pre-built financial knowledge graph, and to generate structured query intents based on the target business nodes.

[0125] Construction unit 43 is used to obtain data permission information from the financial business terminal through a security and compliance intelligent agent, generate data filtering conditions based on the data permission information, and construct prompt words based on structured query intent and data filtering conditions;

[0126] The execution unit 44 is used to input prompt words into the script generation agent, generate a data query script through the script generation agent, and execute the data query script to obtain the target structured data;

[0127] Generation unit 45 is used to generate financial business decision-making strategies by combining the causal effect edges of the financial knowledge graph and the target structured data.

[0128] The aforementioned financial business decision-making device receives business requests from the financial business terminal through receiving unit 41, parses the business requests to obtain a set of business entities; through mapping unit 42, an intent parsing agent is used to map the business entities in the set of business entities to target business nodes in the financial knowledge graph using a pre-built financial knowledge graph, and generates a structured query intent based on the target business nodes; construction unit 43 obtains data permission information from the financial business terminal through a security and compliance agent, generates data filtering conditions based on the data permission information, and constructs prompt words based on the structured query intent and data filtering conditions; through execution unit 44, the prompt words are input to a script generation agent, which generates a data query script and executes the data query script to obtain target structured data; through generation unit 45, a financial business decision-making strategy is generated by combining the causal effect edges of the financial knowledge graph and the target structured data.

[0129] In this embodiment, multiple collaborative intelligent agents are employed. These entities are precisely mapped to target business nodes using a pre-constructed financial knowledge graph, thereby generating structured query intents. Data permission information is acquired in real time, and data filtering conditions are generated accordingly. These two are combined to construct prompts containing security constraints, which are then input into a script generation agent to generate and execute a data query script. This retrieves the target structured data required by the business. Furthermore, deep correlation analysis is performed between the causal effect edges in the financial knowledge graph and the target data, automatically generating financial business decision-making strategies. This achieves end-to-end intelligent processing from natural language requests to structured data insights and automated decision-making, lowering the technical threshold for data analysis, improving operational efficiency, and leveraging the causal reasoning capabilities of the knowledge graph to achieve deep attribution analysis and automatic strategy recommendation. This shortens the operational decision-making cycle and improves the real-time and accuracy of financial business operations, thus solving the technical problem that data analysis methods used by financial institutions often fail to meet the real-time and accuracy requirements of financial business.

[0130] Furthermore, the receiving unit includes: a first extraction module, used to extract natural language text from the business request, perform word segmentation on the natural language text, and extract a sequence of business keywords; a first recognition module, used to input the sequence of business keywords into a pre-trained named entity recognition model, identify the financial entity types in the sequence of business keywords, and generate an initial entity list; a first processing module, used to perform disambiguation processing on the initial entity list based on predefined business context rules, eliminate polysemous word confusion, and obtain an initial set of business entities; and a first verification module, used to match and verify each business entity in the initial set of business entities with a pre-set financial business terminology dictionary, and output a set of business entities that have passed the verification.

[0131] Furthermore, the mapping unit includes: a first traversal module, used to traverse the financial knowledge graph, find target business nodes with the same identifier as each business entity in the business entity set, and map the business entities to target business nodes in the financial knowledge graph based on the same identifier; a first acquisition module, used to acquire the attribute information of the business nodes, wherein the attribute information includes indicator type, dimension label and associated calculation logic formula; a first decomposition module, used to decompose each target business node into at least one child node according to the calculation logic formula, and use the child node as a sub-unit of the query task; and a first generation module, used to perform topological sorting of the sub-units according to the dependency order in the calculation logic formula, and generate a structured query intent link containing the execution order.

[0132] Furthermore, the construction unit includes: a first parsing module, used to parse the authentication token of the current session and extract the user identifier and organizational structure level of the financial business terminal; a first query module, used to query the user permission database based on the user identifier and organizational structure level to obtain the data access permission policy corresponding to the financial business terminal; and a first determining module, used to determine the data filtering conditions based on the data access permission policy and encapsulate the data filtering conditions into data constraint instructions, wherein the data filtering conditions include at least the field values ​​of the filter rows.

[0133] Furthermore, the execution unit includes: a second generation module, used to input prompts to a script generation agent, and generate a data query script for the target database through the script generation agent; a first execution module, used to pre-execute the data query script in a sandbox environment, and capture execution information generated during the execution process, wherein the execution information includes at least one of the following: error logs or exception stack information; and a first correction module, used to generate correction prompts based on the execution information, and send the correction prompts back to the script generation agent to regenerate the data query script, until the data query script is successfully pre-executed, and the final data query script is obtained.

[0134] Furthermore, the generation unit includes: a first comparison module, used to extract the business indicator values ​​corresponding to the target business nodes from the target structured data, compare the business indicator values ​​with historical benchmark values, and calculate the magnitude of change in the business indicator values; a second acquisition module, used to traverse the causal effect edges associated with the target business nodes in the financial knowledge graph, and acquire the dynamic weight coefficients corresponding to each causal effect edge; a first calculation module, used to calculate the contribution of the upstream nodes corresponding to the target business nodes to the indicator changes based on the magnitude of change and the dynamic weight coefficients, and filter out the root cause nodes whose contribution exceeds a preset contribution threshold; and a third generation module, used to determine the abnormal root causes based on the root cause nodes, query a preset mapping matrix based on the abnormal root causes, match the target execution actions corresponding to the abnormal root causes, and generate financial business decision-making strategies based on the target execution actions.

[0135] Furthermore, the financial business decision-making device also includes: a first transformation module for converting target structured data into visualized data; and a first rendering module for rendering the visualized data to the interactive interface of the financial business terminal.

[0136] It should be noted that the receiving unit 41, mapping unit 42, construction unit 43, execution unit 44, and generation unit 45 mentioned above correspond to steps S201 to S205 in Embodiment 1. The instances and application scenarios implemented by the above units and corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules or units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above modules or units can also run as part of the device in the computer terminal 10 provided in Embodiment 1.

[0137] The invention will now be described in conjunction with another alternative embodiment.

[0138] Example 3

[0139] The present invention can also provide an electronic device. Figure 5 This is a hardware structure block diagram of an electronic device (or mobile device) for executing a financial business decision-making method according to an embodiment of the present invention, such as... Figure 5 As shown, the electronic device may include: one or more ( Figure 5 (Only one is shown) processor 502, memory 504, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0140] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the above-described methods. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0141] The processor can access information and applications stored in memory via a transmission device to perform the following steps: receiving business requests from the financial business end and parsing the requests to obtain a set of business entities; using an intent parsing agent, mapping the business entities in the set to target business nodes in the financial knowledge graph using a pre-built financial knowledge graph, and generating a structured query intent based on the target business nodes; obtaining data permission information from the financial business end through a security and compliance agent, generating data filtering conditions based on the data permission information, and constructing prompt words based on the structured query intent and data filtering conditions; inputting the prompt words into a script generation agent, generating a data query script through the script generation agent, executing the data query script, and obtaining the target structured data; and generating a financial business decision-making strategy by combining the causal effect edges of the financial knowledge graph and the target structured data.

[0142] The processor can access information and applications stored in memory via a transmission device to perform the following steps: extract natural language text from the business request, perform word segmentation on the natural language text, and extract a sequence of business keywords; input the sequence of business keywords into a pre-trained named entity recognition model to identify the financial entity types in the sequence of business keywords and generate an initial entity list; perform disambiguation processing on the initial entity list based on predefined business context rules to eliminate polysemous word confusion and obtain an initial set of business entities; perform matching verification between each business entity in the initial set of business entities and a pre-set financial business terminology dictionary, and output the set of business entities that pass the verification.

[0143] The processor can invoke information and applications stored in memory via a transmission device to perform the following steps: traverse the financial knowledge graph, find target business nodes with the same identifier as each business entity in the business entity set, and map the business entities to target business nodes in the financial knowledge graph based on the same identifier; obtain the attribute information of the business nodes, including indicator type, dimension label, and associated calculation logic formula; according to the calculation logic formula, decompose each target business node into at least one child node, and use the child node as a sub-unit of the query task; sort the sub-units topologically according to the dependency order in the calculation logic formula to generate a structured query intent link containing the execution order.

[0144] The processor can access information and applications stored in the memory via the transmission device to perform the following steps: parse the authentication token of the current session, extract the user identifier and organizational structure level of the financial business terminal; query the user permission database based on the user identifier and organizational structure level to obtain the data access permission policy corresponding to the financial business terminal; determine the data filtering conditions based on the data access permission policy, and encapsulate the data filtering conditions into data constraint instructions, wherein the data filtering conditions include at least the field values ​​of the filter rows.

[0145] The processor can access information and applications stored in memory via a transmission device to perform the following steps: inputting prompts to a script generation agent, generating a data query script for the target database; pre-executing the data query script in a sandbox environment, capturing execution information generated during execution, wherein the execution information includes at least one of the following: error logs or exception stack information; generating corrective prompts based on the execution information, and sending the corrective prompts back to the script generation agent to regenerate the data query script, until the data query script pre-execution is successful, resulting in the final data query script.

[0146] The processor can access information and applications stored in the memory via a transmission device to execute the following steps: extracting business indicator values ​​corresponding to the target business node from the target structured data, comparing the business indicator values ​​with historical benchmark values, and calculating the magnitude of change in the business indicator values; traversing the causal effect edges associated with the target business node in the financial knowledge graph, and obtaining the dynamic weight coefficient corresponding to each causal effect edge; calculating the contribution of the upstream nodes corresponding to the target business node to the indicator change based on the magnitude of change and the dynamic weight coefficient, and filtering out root cause nodes whose contribution exceeds a preset contribution threshold; determining abnormal root causes based on the root cause nodes, querying a preset mapping matrix based on the abnormal root causes, matching the target execution action corresponding to the abnormal root cause, and generating a financial business decision-making strategy based on the target execution action.

[0147] The processor can access information and applications stored in memory via a transmission device to perform the following steps: converting the target structured data into visualized data; and rendering the visualized data onto the interactive interface of the financial business terminal.

[0148] This invention provides a financial business decision-making solution. It employs multiple collaborative intelligent agents, utilizing a pre-constructed financial knowledge graph to precisely map these entities to target business nodes, thereby generating structured query intents. Data permission information is acquired in real-time, and data filtering conditions are generated accordingly. These two are combined to construct prompts containing security constraints, which are input into a script generation agent to generate and execute a data query script, obtaining the target structured data required by the business. Based on this, a deep correlation analysis is performed between the causal effect edges in the financial knowledge graph and the target data, automatically generating financial business decision-making strategies. This achieves end-to-end intelligent processing from natural language requests to structured data insights and automated decision-making, lowering the technical threshold for data analysis, improving operational efficiency, and leveraging the causal reasoning capabilities of the knowledge graph to achieve deep attribution analysis and automatic strategy recommendation, shortening the operational decision-making cycle and improving the real-time and accuracy of financial business operations. This solves the technical problem in related technologies where data analysis methods used by financial institutions are insufficient to meet the real-time and accuracy requirements of financial business operations.

[0149] Those skilled in the art will understand that Figure 5 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 5 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 5 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 5 The different configurations shown.

[0150] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0151] The invention will now be described in conjunction with another alternative embodiment.

[0152] Example 4

[0153] This invention also provides a computer-readable storage medium. Optionally, in this invention, the computer-readable storage medium can be used to store the program code executed by the financial business decision-making method provided in Embodiment 1.

[0154] Optionally, in this embodiment of the invention, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0155] This invention also provides a computer program product, which, when executed on a data processing device, is suitable for executing the steps of a financial business decision-making method: a receiving unit, used to receive business requests from the financial business end and parse the business requests to obtain a set of business entities; a mapping unit, used to use an intent parsing agent to map the business entities in the set of business entities to target business nodes in the financial knowledge graph using a pre-built financial knowledge graph, and to generate a structured query intent based on the target business nodes; a construction unit, used to obtain data permission information from the financial business end through a security and compliance agent, generate data filtering conditions based on the data permission information, and construct prompt words based on the structured query intent and data filtering conditions; an execution unit, used to input the prompt words into a script generation agent, generate a data query script through the script generation agent, and execute the data query script to obtain target structured data; and a generation unit, used to combine the causal effect edges of the financial knowledge graph and the target structured data to generate a financial business decision-making strategy.

[0156] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0157] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0158] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0159] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0160] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0161] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0162] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A financial business decision-making method, characterized in that, include: Receive business requests from the financial business side and parse the business requests to obtain a set of business entities; An intent parsing agent is used to map business entities in the business entity set to target business nodes in the financial knowledge graph using a pre-built financial knowledge graph, and to generate structured query intents based on the target business nodes. The system obtains data permission information from the financial business terminal through a security and compliance intelligent agent, generates data filtering conditions based on the data permission information, and constructs prompt words based on the structured query intent and the data filtering conditions. The prompt words are input into the script generation agent, which generates a data query script and executes the data query script to obtain the target structured data. The financial business decision-making strategy is generated by combining the causal effect edges of the financial knowledge graph and the target structured data.

2. The method according to claim 1, characterized in that, The steps of receiving business requests from the financial service provider and parsing the requests to obtain a set of business entities include: Natural language text is extracted from the business request, and word segmentation is performed on the natural language text to extract the business keyword sequence; The business keyword sequence is input into a pre-trained named entity recognition model to identify the financial entity types in the business keyword sequence and generate an initial entity list; The initial entity list is disambiguated based on predefined business context rules to eliminate polysemous word confusion and obtain the initial business entity set. Each business entity in the initial set of business entities is matched and verified against a pre-set dictionary of financial business terms, and the set of business entities that pass the verification is output.

3. The method according to claim 1, characterized in that, The steps of mapping business entities in the set of business entities to target business nodes in the financial knowledge graph using a pre-constructed financial knowledge graph, and generating structured query intents based on the target business nodes, include: Traverse the financial knowledge graph, find the target business node that has the same identifier as each business entity in the business entity set, and map the business entity to the target business node in the financial knowledge graph based on the same identifier. Obtain the attribute information of the business node, wherein the attribute information includes the indicator type, dimension label and associated calculation logic formula; According to the calculation logic formula, each target business node is decomposed into at least one sub-node, and the sub-node is used as a sub-unit of the query task; The sub-units are topologically sorted according to the dependency order in the calculation logic formula to generate a structured query intent link containing the execution order.

4. The method according to claim 1, characterized in that, The steps of obtaining data permission information from the financial business terminal through a security and compliance intelligent agent, and generating data filtering conditions based on the data permission information, include: Parse the authentication token of the current session to extract the user identifier and organizational structure level of the financial business terminal; Based on the user identifier and the organizational structure level to which they belong, query the user permission database to obtain the data access permission policy corresponding to the financial business terminal; The data filtering conditions are determined based on the data access permission policy, and the data filtering conditions are encapsulated into data constraint instructions, wherein the data filtering conditions include at least the field values ​​of the filtered rows.

5. The method according to claim 4, characterized in that, The steps of inputting the prompt words into the script-generating agent and generating a data query script through the script-generating agent include: The prompt words are input into the script-generating agent, which then generates a data query script for the target database. The data query script is pre-executed in a sandbox environment, and the execution information generated during the execution process is captured, wherein the execution information includes at least one of the following: error log or exception stack information; Based on the execution information, a correction prompt is generated, and the correction prompt is sent back to the script generation agent to regenerate the data query script. This process continues until the data query script is successfully pre-executed, resulting in the final data query script.

6. The method according to claim 1, characterized in that, The steps for generating financial business decision-making strategies by combining the causal effect edges of the financial knowledge graph and the target structured data include: Extract the business indicator values ​​corresponding to the target business nodes from the target structured data, compare the business indicator values ​​with historical benchmark values, and calculate the magnitude of change of the business indicator values; Traverse the causal effect edges in the financial knowledge graph that are associated with the target business node, and obtain the dynamic weight coefficient corresponding to each causal effect edge; Based on the change magnitude and the dynamic weight coefficient, the contribution of the upstream node corresponding to the target business node to the indicator change is calculated, and the root cause node whose contribution exceeds the preset contribution threshold is selected. The abnormal root cause is determined based on the root cause node, and a preset mapping matrix is ​​queried based on the abnormal root cause to match the target execution action corresponding to the abnormal root cause. A financial business decision-making strategy is generated based on the target execution action.

7. The method according to claim 1, characterized in that, After obtaining the target structured data, the process also includes: Transform the target structured data into visualized data; The visualized data is rendered onto the interactive interface of the financial business terminal.

8. A financial business decision-making device, characterized in that, include: The receiving unit is used to receive business requests from the financial business terminal and parse the business requests to obtain a set of business entities. The mapping unit is used to employ an intent parsing agent to map business entities in the set of business entities to target business nodes in the financial knowledge graph using a pre-built financial knowledge graph, and to generate structured query intents based on the target business nodes. The construction unit is used to obtain data permission information of the financial business terminal through a security and compliance intelligent agent, generate data filtering conditions based on the data permission information, and construct prompt words based on the structured query intent and the data filtering conditions; An execution unit is configured to input the prompt words into a script generation agent, generate a data query script through the script generation agent, and execute the data query script to obtain target structured data. The generation unit is used to combine the causal effect edges of the financial knowledge graph and the target structured data to generate financial business decision-making strategies.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the financial business decision-making method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the financial business decision-making method according to any one of claims 1 to 7.

11. A computer program product, characterized in that, The computer program product includes a computer program, wherein the computer program, when executed by a processor, implements the financial business decision-making method according to any one of claims 1 to 7.

Images