Dag topology-based financial ai element processing method and system

Abstract

The application relates to a financial AI element processing method based on a DAG topological structure, which comprises the following steps: constructing a DAG topological model to cover all-service AI elements of a financial institution, a construction method being as follows: abstracting a business system, a function module and AI elements in the financial institution into nodes of the DAG topological model, and each node being configured with a standardized attribute set; corresponding a hierarchical membership relationship and a business association relationship of each AI element to a directed edge between nodes, and a same node being capable of being associated with multiple upstream parent nodes; when an AI element calling request is received, adopting a breadth-first traversal algorithm based on topological sorting to locate target elements and associated elements according to a partial order relation and an edge weight priority of the DAG model; and performing permission control and progressive intent recognition based on the DAG topological model. The application also discloses a financial AI element processing system based on the DAG topological structure for realizing the above method.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of natural language processing, computer access control, and financial information processing, and particularly to a method and system for processing financial AI elements based on DAG topology. Background Technology

[0002] Currently, enterprise-level AI applications have entered a critical transition period from "pilot exploration" to "large-scale implementation." Financial institutions have much higher mandatory requirements for the compliance, real-time performance, and reliability of AI applications than for general-purpose scenarios. In the engineering implementation of financial AI technology, existing technologies generally suffer from the following shortcomings:

[0003] The core functional modules are fragmented, with poor coordination and high redundancy: In existing technologies, the three core functions of AI element resource management, access control, and user intent parsing mostly adopt independent module architectures. The underlying data structures and processing logic of each module are not unified. Data communication between modules requires multiple structural conversions and format adaptations, which not only increases the complexity of system development and operation and maintenance costs, but also easily leads to problems such as data inconsistency and excessive response delays, failing to meet the real-time requirements of financial business.

[0004] The basic topology selection is unreasonable and has inherent logical flaws: some existing solutions use tree structures for AI element management, which can only support hierarchical mapping of a single parent node. This cannot adapt to scenarios in the financial industry where the same AI element is shared by multiple business domains with multiple parent nodes, resulting in redundant element storage and low retrieval efficiency. Other solutions use ordinary directed graph structures, which cannot avoid the problem of circular paths. This can easily lead to failures such as infinite loops in permission propagation and infinite divergence of intent resolution paths, posing compliance risks such as system downtime and failure of permission control, which is contrary to the compliance requirements of the financial industry.

[0005] Deep coupling of multiple modules presents technical bottlenecks and insufficient adaptability: Existing technologies are unable to achieve deep coupling of the three major functions of AI element management, access control, and intent parsing. If data from multiple modules is forcibly integrated, new problems such as permission propagation conflicts, disconnect between resource ownership and access control, and redundancy in intent parsing paths will occur, leading to the failure of the system's layered design, a significant decrease in reliability, and an inability to adapt to the complex business scenarios and stringent regulatory requirements of the financial industry.

[0006] Intent parsing is disconnected from the underlying topology, resulting in insufficient processing efficiency and recognition accuracy: Existing intent parsing solutions mostly adopt manually set hierarchical rules that are not bound to the underlying topology. They cannot use the correlation to achieve dynamic optimization of the parsing path, leading to redundancy in single-round inference context and serious interference from irrelevant information. This not only consumes a lot of computing resources but also reduces the accuracy and response speed of intent recognition, making it unsuitable for the complex multi-round dialogues and accurate intent matching requirements in financial scenarios. Summary of the Invention

[0007] Based on the above-mentioned problems in this technical field, the present invention proposes the following technical solution:

[0008] A financial AI element processing method based on DAG topology includes the following steps:

[0009] A DAG topology model is constructed, which covers all AI elements of a financial institution's business. The construction method includes: abstracting the business systems, functional modules, and AI elements within the financial institution into nodes of the DAG topology model, and configuring each node with a standardized attribute set; mapping the hierarchical membership and business association relationships of each AI element to directed edges between nodes, and allowing the same node to be associated with multiple upstream parent nodes; when receiving an AI element call request, using a breadth-first traversal algorithm based on topological sorting, the target element and its associated elements are located according to the partial order relationship and edge weight priority of the DAG model.

[0010] The access control based on the DAG topology model includes: defining the default direction of permission propagation as from upstream node to downstream node; and for downstream nodes with multiple upstream nodes, using a permission priority algorithm based on DAG topology to resolve permission conflicts.

[0011] Based on the DAG topology model, progressive intent recognition is performed, specifically including: matching user request or question information with all intent descriptions and subdomain descriptions under the current node; if a subdomain description is matched, the intent parsing cursor is moved to the corresponding subdomain node and the matching is repeated according to the preset node association rules; if an intent description is matched, the traversal is terminated and the intent is output; if none of the above are matched, no matching intent is output and the process is terminated.

[0012] Furthermore, in the financial AI element processing method based on DAG topology, the method for constructing the DAG topology model further includes: using a topology sorting algorithm to transform the attribution relationship of the AI ​​elements into a partial order relationship of nodes in the DAG topology model, and automatically performing hierarchical aggregation and association management of the elements.

[0013] Furthermore, the financial AI element processing method based on DAG topology also includes: setting a caching mechanism for frequently accessed AI element nodes and storing frequently accessed AI element nodes in a cache pool; and setting a DAG topology dynamic update mechanism, which automatically triggers topology sorting updates when any AI element is added, deleted, or has its attributes changed.

[0014] Furthermore, in the financial AI element processing method based on DAG topology, the permission priority algorithm specifically includes: for any downstream node, it is determined to have access permission only when all upstream nodes allow access;

[0015] Construct role-based and policy-based access control models: The role-based access control model binds the access permissions of roles to nodes, and the policy-based access control model associates access policies with nodes; when a user initiates an access request, the topology traversal algorithm verifies the range of accessible nodes corresponding to the user's role, determines whether the access policy meets the requirements, and filters out AI elements within the permission range.

[0016] Furthermore, in the financial AI element processing method based on DAG topology, the progressive intent recognition reuses the parsing path automatically generated by the node association relationship of the DAG topology model. The parsing process proceeds layer by layer according to the directed edge direction of the DAG topology and terminates when a specific business intent is hit or a node without downstream child nodes is reached.

[0017] The inventors also proposed a financial AI element processing system based on DAG topology, including:

[0018] The DAG topology model construction module is used to construct a DAG topology model covering all AI elements of a financial institution's business. The construction includes: abstracting the business systems, functional modules, and AI elements within the financial institution into nodes of the DAG topology model, with each node configured with a standardized attribute set; mapping the hierarchical membership and business association relationships of each AI element to directed edges between nodes, with the same node being associated with multiple upstream parent nodes; and when receiving an AI element call request, using a breadth-first traversal algorithm based on topology sorting to locate the target element and its associated elements according to the partial order relationship and edge weight priority of the DAG model.

[0019] The access control module is used to perform access control based on the DAG topology model, specifically including: defining the default direction of permission propagation as from upstream node to downstream node; for downstream nodes with multiple upstream nodes, using a permission priority algorithm based on DAG topology to resolve permission conflicts;

[0020] The intent recognition module is used to perform progressive intent recognition based on the DAG topology model. Specifically, it includes: matching user request or question information with all intent descriptions and subdomain descriptions under the current node; if a subdomain description is matched, the intent parsing cursor is moved to the corresponding subdomain node and the matching is repeated according to the preset node association rules; if an intent description is matched, the traversal is terminated and the intent is output; if none of the matches are matched, no matching intent is output and the process is terminated.

[0021] Furthermore, in the aforementioned financial AI element processing system based on DAG topology, the DAG topology model construction module is also used to: transform the attribution relationship of the AI ​​elements into the partial order relationship of nodes in the DAG topology model through a topology sorting algorithm, and automatically perform hierarchical aggregation and association management of elements.

[0022] Furthermore, in the aforementioned financial AI element processing system based on DAG topology, the DAG topology model construction module is also used to: set up a caching mechanism for frequently accessed AI element nodes, storing frequently accessed AI element nodes in a cache pool; and set up a DAG topology dynamic update mechanism, automatically triggering topology sorting updates when any AI element is added, deleted, or has its attributes changed.

[0023] Furthermore, in the financial AI element processing system based on DAG topology, the permission priority algorithm used in the permission control module specifically includes: for any downstream node, it is determined to have access permission only when all upstream nodes allow access;

[0024] The access control module is also used to construct role-based and policy-based access control models: the role-based access control model binds the access permissions of roles and nodes, and the policy-based access control model associates access policies with nodes; when a user initiates an access request, the topology traversal algorithm is used to verify the range of accessible nodes corresponding to the user's role, determine whether the access policy meets the requirements, and filter out AI elements within the permission range.

[0025] Furthermore, in the financial AI element processing system based on DAG topology, the progressive intent recognition module reuses the parsing path automatically generated by the node association relationship of the DAG topology model. The parsing process proceeds layer by layer according to the directed edge direction of the DAG topology and terminates when a specific business intent is hit or a node without downstream child nodes is reached.

[0026] Compared with existing technologies, the advantages and technical effects of the method proposed in this invention include: (1) It can utilize the acyclicity and partial order of DAG to automatically realize the hierarchical collection and association management of elements, avoid the circular dependency problem of element ownership, ensure the global stability and logical rationality of the resource ownership structure, and solve the inherent defects of the circular path in ordinary directed graph structures; (2) It shortens the response time, realizes the efficient retrieval and reuse of AI elements, and solves the technical problems of low retrieval efficiency and low reuse rate caused by element silos. At the same time, it ensures that the topology structure is consistent with the actual business and avoids the hidden danger of data inconsistency; (3) It accurately filters AI elements within the scope of permissions, realizes "safe and controllable, on-demand sharing", and takes into account both compliance and business convenience. Attached Figure Description

[0027] Figure 1 This is a flowchart of the financial AI element processing method based on DAG topology described in this invention;

[0028] Figure 2 This is a structural diagram of the financial AI element processing device based on DAG topology described in this invention.

[0029] Explanation of reference numerals in the attached figures:

[0030] 1-DAG Topology Model Construction Module

[0031] 2-Access Control Module

[0032] 3-Intent Recognition Module Detailed Implementation

[0033] To explain in detail the technical content, structural features, objectives, and effects of the technical solution, the following description is provided in conjunction with specific embodiments and accompanying drawings.

[0034] Please see Figure 1 The present invention provides a financial AI element processing method based on DAG (Directed Acyclic Graph) topology, comprising the following steps:

[0035] S1. Construct a DAG topology model that covers all AI elements of financial institutions' businesses.

[0036] The business systems (such as credit approval systems, anti-fraud systems, etc.), functional modules (such as user authentication modules, log auditing modules, etc.), and AI elements (such as knowledge bases, MCPs, business sensitive words, business new words) within financial institutions are abstracted into nodes of the DAG topology model. Each node is configured with a standardized attribute set, which includes, but is not limited to: globally unique identifier of the element, element type, data sensitivity level, business department, creation time, update time, and operation log audit items.

[0037] The hierarchical affiliation and business relationship of each AI element are mapped to directed edges between nodes. Upstream nodes correspond to general AI elements in the upper layer, while downstream nodes correspond to AI elements in specific vertical fields. Furthermore, the same node can be associated with multiple upstream parent nodes, thereby achieving the goal of "covering AI elements across the entire business domain of financial institutions".

[0038] By using a topological sorting algorithm, the attribution relationship of the AI ​​elements is transformed into a partial order relationship of nodes in the DAG topology model, automatically performing hierarchical aggregation and association management of elements, and avoiding circular dependency problems.

[0039] S2. When receiving an AI element call request, a breadth-first traversal algorithm based on topological sorting is used to locate the target element and its associated elements according to the partial order relationship and edge weight priority of the DAG model.

[0040] For scenarios in the financial industry characterized by high concurrency, frequently accessed AI element nodes are stored in a cache pool to further shorten response time. At the same time, a DAG topology dynamic update mechanism is set up so that when any AI element is added, deleted, or its attributes are changed, the topology sorting is automatically triggered to ensure that the topology structure is consistent with the actual business.

[0041] S3. Perform access control based on the DAG topology model.

[0042] In access control, the default propagation direction of permissions is defined as from upstream nodes to downstream nodes. That is, the access permissions configured on an upstream node are automatically inherited to all its downstream nodes when there are no explicit blocking rules.

[0043] For downstream nodes with multiple upstream nodes, a permission priority algorithm based on DAG topology is used to resolve permission conflicts. Specifically, for any downstream node, access is only granted if all upstream nodes allow access; if any upstream node sets an access denial rule for that node, it is determined that access is denied.

[0044] Construct role-based and policy-based access control models: Role-based access control (RBAC) binds access permissions to nodes; for example, a regular employee role is only bound to permissions for low-sensitivity nodes. Policy-based access control (PBAC) associates access policies (including IP address, operating terminal, user level, and operation time) with nodes. When a user initiates an access request, a topology traversal algorithm verifies the range of accessible nodes corresponding to the user's role, determines whether the access policy meets the requirements, and filters out AI elements within the authorized scope.

[0045] S4. Perform progressive intent recognition based on the DAG topology model.

[0046] The system matches user requests or questions with all intent descriptions and subdomain descriptions under the current node. Specifically, after receiving a user's question-and-answer request, the system first locks the initial node of the DAG topology that the user can access based on the user's identity and permission information, and then performs large-scale intent recognition by matching the user's question content with the intent descriptions and subdomain descriptions under that initial node.

[0047] If a subdomain description is matched, the intent parsing cursor is moved to the corresponding subdomain node and the matching is repeated according to the preset node association rules; if an intent description is matched, the traversal is terminated and the intent is output; if neither is matched, no matching intent is output and the process is terminated.

[0048] The progressive intent recognition reuses the parsing path automatically generated by the node association relationship of the DAG topology model. The parsing process advances layer by layer according to the directed edge direction of the DAG topology and terminates when a specific business intent is hit or a node without downstream child nodes is reached.

[0049] More specifically, suppose there is an intelligent customer service system in a financial compliance scenario, managed using the financial AI element processing method based on DAG topology described in this invention, as shown in the following steps and effects:

[0050] Step A: Construct a DAG topology model.

[0051] The credit business system, anti-fraud function module, and AI elements (including credit knowledge base, sensitive word library, and MCP service) within the financial institution are abstracted as nodes of the DAG topology model. Each node is configured with a standardized attribute set, and more important attributes are extracted, such as data sensitivity level divided into several levels (e.g., L1-L4), business department affiliation and corresponding business line, etc.

[0052] The hierarchical affiliation and business relationships of each AI element are mapped to directed edges between nodes: for example, the "General Financial Knowledge Base" node is an upstream node, the "Credit Card Sub-Knowledge Base" node is a midstream node, and the "Credit Card Overdue Processing Knowledge Base" is a downstream node. The "Credit Card Overdue Processing Knowledge Base" can be simultaneously associated with two upstream parent nodes: the "Credit Card Sub-Knowledge Base" and the "General Risk Control Rules Base." A single node can be associated with multiple upstream parent nodes to adapt to multi-affiliation scenarios in real-world business.

[0053] Step B: AI element retrieval and caching.

[0054] When the intelligent customer service system receives a user query for "how to negotiate overdue credit card payments," the system receives an AI element retrieval request. Using a breadth-first traversal algorithm based on topological sorting, and following the partial order relationship and edge weight priority of the DAG model, it quickly locates the "Credit Card Overdue Processing Knowledge Base" node and its associated sensitive word elements and script template elements, starting from the "Credit Card Sub-Knowledge Base" node. The algorithm's time complexity is O(N+E) (where N is the number of nodes and E is the number of directed edges). The frequently accessed "Overdue Script Template" node is stored in a cache pool, allowing subsequent identical queries to retrieve data directly from the cache.

[0055] Step C: Implement access control based on the DAG topology model.

[0056] The default propagation direction for permissions is defined as from upstream nodes to downstream nodes. For example, if the "General Financial Knowledge Base" node is configured with "read-only" permission, then this permission will automatically propagate to all its downstream nodes.

[0057] For the downstream node of the "Credit Card Overdue Processing Knowledge Base" which has multiple upstream nodes ("Credit Card Sub-Knowledge Base" and "Risk Control General Rule Base"), a permission priority algorithm based on DAG topology is adopted: if the "Risk Control General Rule Base" node sets an access prohibition rule for the downstream node due to compliance requirements, the system determines that the downstream node has no access permission; only when all upstream nodes allow access is it determined to have access permission.

[0058] Simultaneously, a role-based and policy-based access control model is constructed: the role-based access control model binds roles like "customer service" to L1-L3 sensitive nodes, and the "risk control audit" role to L4 sensitive nodes; the policy-based access control model associates the access policy of "only allowing internal office IP addresses, working hours 9:00-18:00" with the credit knowledge base nodes. When a customer service specialist initiates an access request from the external network outside of working hours, a topology traversal algorithm verifies the accessible node range (only L1-L3 nodes) corresponding to the user role, determines that the access policy does not meet the requirements, and ultimately denies access and outputs a permission denied message.

[0059] Step D: Perform progressive intent recognition based on the DAG topology model.

[0060] The system matches the user's request or question, "Will my credit card be reported to the credit bureau if I'm three days overdue?", against all intent descriptions and subdomain descriptions under the current node. The initial node is "Credit Card Customer Service," whose subdomain descriptions include "Bill Inquiry," "Credit Limit Management," and "Overdue Processing." If a semantic match matches the "Overdue Processing" subdomain description, the intent parsing cursor is moved to the corresponding "Overdue Processing" subdomain node according to the preset node association rules, and the matching is repeated.

[0061] Under the "Overdue Processing" node, intent descriptions include "Overdue Penalty Interest Calculation," "Credit Reporting Rules," and "Negotiation Repayment." If the intent description "Credit Reporting Rules" is matched, the traversal terminates and that intent is output as the final identification result. If no intent or subdomain is matched during the process, no matching intent is output and the process terminates.

[0062] The progressive intent recognition reuses the parsing path automatically generated by the node association relationship of the DAG topology model. The parsing process proceeds layer by layer from "credit card customer service" → "overdue processing" → "credit reporting rules" according to the directed edge direction of the DAG topology, and terminates after hitting the specific business intent, without the need for additional data structure conversion.

[0063] Please participate Figure 2 Here is a structural diagram of a financial AI element processing system based on DAG topology as described in this invention, comprising:

[0064] DAG topology model construction module 1 is used to construct a DAG topology model covering all AI elements of a financial institution's business. The construction includes: abstracting the business systems, functional modules, and AI elements within the financial institution into nodes of the DAG topology model, with each node configured with a standardized attribute set; mapping the hierarchical membership and business association relationships of each AI element to directed edges between nodes, with the same node being associated with multiple upstream parent nodes; and when receiving an AI element call request, using a breadth-first traversal algorithm based on topology sorting to locate the target element and its associated elements according to the partial order relationship and edge weight priority of the DAG model.

[0065] The permission control module 2 is used to perform permission control based on the DAG topology model, specifically including: defining the default permission propagation direction as from upstream node to downstream node; for downstream nodes with multiple upstream nodes, using a permission priority algorithm based on DAG topology to resolve permission conflicts;

[0066] The intent recognition module 3 is used to perform progressive intent recognition based on the DAG topology model. Specifically, it includes: matching user request or question information with all intent descriptions and subdomain descriptions under the current node; if a subdomain description is matched, the intent parsing cursor is moved to the corresponding subdomain node and the matching is repeated according to the preset node association rules; if an intent description is matched, the traversal is terminated and the intent is output; if none of the above are matched, no matching intent is output and the process is terminated.

[0067] Furthermore, in the aforementioned financial AI element processing system based on DAG topology, the DAG topology model construction module 1 is also used to: transform the attribution relationship of the AI ​​elements into the partial order relationship of nodes in the DAG topology model through a topology sorting algorithm, and automatically perform hierarchical aggregation and association management of elements.

[0068] Furthermore, in the financial AI element processing system based on DAG topology, the DAG topology model construction module 1 is also used to: set up a caching mechanism for frequently accessed AI element nodes and store frequently accessed AI element nodes in a cache pool; and set up a DAG topology dynamic update mechanism so that when any AI element is added, deleted, or has its attributes changed, the topology sorting update is automatically triggered.

[0069] Furthermore, in the financial AI element processing system based on DAG topology, the permission priority algorithm used in the permission control module 2 specifically includes: for any downstream node, it is determined to have access permission only when all upstream nodes allow access;

[0070] The access control module 2 is also used to construct role-based and policy-based access control models: the role-based access control model binds the access permissions of roles and nodes, and the policy-based access control model associates access policies with nodes; when a user initiates an access request, the topology traversal algorithm is used to verify the range of accessible nodes corresponding to the user's role, determine whether the access policy meets the requirements, and filter out AI elements within the permission range.

[0071] Furthermore, in the financial AI element processing system based on DAG topology, the progressive intent recognition module 3 reuses the parsing path automatically generated by the node association relationship of the DAG topology model. The parsing process proceeds layer by layer according to the directed edge direction of the DAG topology and terminates when a specific business intent is hit or a node without downstream child nodes is reached.

[0072] This method uses computer-executed modeling algorithms to abstract the internal business systems, functional modules, and various AI elements (including knowledge bases, MCPs, business-sensitive words, business-related new words, etc.) of financial institutions into nodes of a DAG topology. Each node is configured with a standardized attribute set, including a globally unique identifier, element type, data sensitivity level, business department affiliation, creation time, update time, and operation log audit items. Directed edges between nodes are explicitly defined as hierarchical relationships and business dependencies between elements. Upstream nodes correspond to upper-level general domain AI elements, while downstream nodes correspond to subdivided vertical domain AI elements. This naturally supports the association of the same node with multiple upstream parent nodes, thereby constructing a unified DAG topology model covering all business domain AI elements of financial institutions.

[0073] Through topological sorting algorithms, the ownership relationships of AI elements can be transformed into partial order relationships of DAG nodes. Leveraging the acyclicity and partial order of DAGs, hierarchical aggregation and association management of elements are automatically achieved, avoiding circular dependencies in element ownership and ensuring the global stability and logical rationality of the resource ownership structure. This solves the inherent defects of circular paths in ordinary directed graph structures. When the system needs to call AI elements, a breadth-first search (BFS) algorithm based on topological sorting is used to quickly locate the target element and its associated elements according to the partial order relationship and edge weight priority of the DAG. The algorithm's time complexity is O(N+E) (where N is the number of nodes and E is the number of directed edges). For high-concurrency, large-node scenarios in finance, a node caching mechanism is added to store frequently called AI element nodes in a cache pool, further shortening the response time and achieving efficient retrieval and reuse of AI elements. This solves the technical problems of low retrieval efficiency and low reuse rate caused by element silos. Simultaneously, a dynamic DAG topology update mechanism is designed. When AI elements are added, deleted, or their attributes change, a topology sort update is automatically triggered to ensure that the topology structure is consistent with the actual business, avoiding potential data inconsistencies.

[0074] The access control algorithm accurately filters out AI elements within the scope of permissions, achieving "safe and controllable, on-demand sharing" while balancing compliance and business convenience.

[0075] The progressive precision intent recognition algorithm ensures that the parsing process has a definite endpoint by leveraging the acyclicity and partial order of the DAG topology. At the same time, it significantly reduces the size of the candidate set in a single round of matching, reduces interference from irrelevant information, and effectively improves the accuracy and response speed of intent recognition. Furthermore, intent parsing, element management, and access control share the same DAG topology, eliminating the need for data structure conversion, further reducing system redundancy and improving processing efficiency.

[0076] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Unless otherwise specified, an element defined by the phrase "comprising..." or "including..." does not exclude the presence of additional elements in the process, method, article, or terminal device that includes said element. Additionally, in this document, "greater than," "less than," "exceeding," etc., are understood to exclude the stated number; "above," "below," "within," etc., are understood to include the stated number.

[0077] Those skilled in the art will understand that the above embodiments can be provided as methods, apparatus, or computer program products. These embodiments may take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. All or part of the steps in the methods involved in the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium for executing all or part of the steps described in the methods of the above embodiments. The computer device includes, but is not limited to: personal computers, servers, general-purpose computers, special-purpose computers, network devices, embedded devices, programmable devices, smart mobile terminals, smart home devices, wearable smart devices, in-vehicle smart devices, etc.; the storage medium includes, but is not limited to: RAM, ROM, magnetic disks, magnetic tapes, optical disks, flash memory, USB flash drives, portable hard drives, memory cards, memory sticks, network server storage, network cloud storage, etc.

[0078] The above embodiments are described with reference to flowchart illustrations and / or block diagrams of the methods, apparatus (systems), and computer program products according to the embodiments. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a computer device to produce a machine, such that the instructions, which execute via the processor of the computer device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0079] These computer program instructions may also be stored in a computer device-readable storage medium that can direct a computer device to operate in a particular manner, such that the instructions stored in the computer device-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0080] These computer program instructions can also be loaded onto a computer device, causing a series of operational steps to be performed on the computer device to produce a computer-implemented process, thereby providing instructions that execute on the computer device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0081] Although the above embodiments have been described, those skilled in the art, once they understand the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the above descriptions are merely embodiments of the present invention and do not limit the scope of patent protection of the present invention. Any equivalent structural or procedural transformations made using the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A financial AI element processing method based on DAG topology, characterized in that, Including the following steps: A DAG topology model is constructed, which covers all AI elements of a financial institution's business. The construction method includes: abstracting the business systems, functional modules, and AI elements within the financial institution into nodes of the DAG topology model, and configuring each node with a standardized attribute set; mapping the hierarchical membership and business association relationships of each AI element to directed edges between nodes, and allowing the same node to be associated with multiple upstream parent nodes; when receiving an AI element call request, using a breadth-first traversal algorithm based on topological sorting, the target element and its associated elements are located according to the partial order relationship and edge weight priority of the DAG model. The access control based on the DAG topology model includes: defining the default direction of permission propagation as from upstream node to downstream node; and for downstream nodes with multiple upstream nodes, using a permission priority algorithm based on DAG topology to resolve permission conflicts. Based on the DAG topology model, progressive intent recognition is performed, specifically including: matching user request or question information with all intent descriptions and subdomain descriptions under the current node; if a subdomain description is matched, the intent parsing cursor is moved to the corresponding subdomain node and the matching is repeated according to the preset node association rules; if an intent description is matched, the traversal is terminated and the intent is output; if none of the above are matched, no matching intent is output and the process is terminated.

2. The financial AI element processing method based on DAG topology as described in claim 1, characterized in that, The method for constructing the DAG topology model further includes: using a topology sorting algorithm to transform the attribution relationship of the AI ​​elements into a partial order relationship of nodes in the DAG topology model, and automatically performing hierarchical aggregation and association management of the elements.

3. The financial AI element processing method based on DAG topology as described in claim 2, characterized in that, Also includes: A caching mechanism is set up for frequently accessed AI element nodes, and these nodes are stored in a cache pool. A DAG topology dynamic update mechanism is set up so that when any AI feature is added, deleted, or its attributes are changed, the topology sorting is automatically triggered to update.

4. The financial AI element processing method based on DAG topology as described in claim 1, characterized in that, The permission priority algorithm specifically includes: for any downstream node, it is determined to have access rights only when all upstream nodes allow access; Construct role-based and policy-based access control models: The role-based access control model binds the access permissions of roles to nodes, and the policy-based access control model associates access policies with nodes; when a user initiates an access request, the topology traversal algorithm verifies the range of accessible nodes corresponding to the user's role, determines whether the access policy meets the requirements, and filters out AI elements within the permission range.

5. The financial AI element processing method based on DAG topology as described in claim 1, characterized in that, The progressive intent recognition reuses the parsing path automatically generated by the node association relationship of the DAG topology model. The parsing process advances layer by layer according to the directed edge direction of the DAG topology and terminates when a specific business intent is hit or a node without downstream child nodes is reached.

6. A financial AI element processing system based on DAG topology, characterized in that, include: The DAG topology model construction module is used to construct a DAG topology model covering all AI elements of a financial institution's business. The construction includes: abstracting the business systems, functional modules, and AI elements within the financial institution into nodes of the DAG topology model, with each node configured with a standardized attribute set; mapping the hierarchical membership and business association relationships of each AI element to directed edges between nodes, with the same node being associated with multiple upstream parent nodes; and when receiving an AI element call request, using a breadth-first traversal algorithm based on topology sorting to locate the target element and its associated elements according to the partial order relationship and edge weight priority of the DAG model. The access control module is used to perform access control based on the DAG topology model, specifically including: defining the default direction of permission propagation as from upstream node to downstream node; for downstream nodes with multiple upstream nodes, using a permission priority algorithm based on DAG topology to resolve permission conflicts; The intent recognition module is used to perform progressive intent recognition based on the DAG topology model. Specifically, it includes: matching user request or question information with all intent descriptions and subdomain descriptions under the current node; if a subdomain description is matched, the intent parsing cursor is moved to the corresponding subdomain node and the matching is repeated according to the preset node association rules; if an intent description is matched, the traversal is terminated and the intent is output; if none of the matches are matched, no matching intent is output and the process is terminated.

7. The financial AI element processing system based on DAG topology as described in claim 6, characterized in that, The DAG topology model construction module is also used to: transform the ownership relationship of the AI ​​elements into the partial order relationship of nodes in the DAG topology model through a topology sorting algorithm, and automatically perform hierarchical collection and association management of elements.

8. The financial AI element processing system based on DAG topology as described in claim 7, characterized in that, The DAG topology model construction module is also used to: set up a caching mechanism for frequently accessed AI feature nodes, storing frequently accessed AI feature nodes in a cache pool; and set up a DAG topology dynamic update mechanism, automatically triggering topology sorting updates when any AI feature is added, deleted, or has its attributes changed.

9. The financial AI element processing system based on DAG topology as described in claim 6, characterized in that, The permission priority algorithm used in the permission control module specifically includes: for any downstream node, it is determined to have permission only when all upstream nodes allow access; The access control module is also used to construct role-based and policy-based access control models: the role-based access control model binds the access permissions of roles and nodes, and the policy-based access control model associates access policies with nodes; when a user initiates an access request, the topology traversal algorithm is used to verify the range of accessible nodes corresponding to the user's role, determine whether the access policy meets the requirements, and filter out AI elements within the permission range.

10. The financial AI element processing system based on DAG topology as described in claim 6, characterized in that, In the intent recognition module, the progressive intent recognition reuses the parsing path automatically generated by the node association relationship of the DAG topology model. The parsing process advances layer by layer according to the directed edge direction of the DAG topology and terminates when a specific business intent is hit or a node without downstream child nodes is reached.

Images