Business processing method and device

Abstract

The invention discloses a service processing method and device. The method comprises the steps of obtaining a business demand description text; determining a business operation task corresponding to the business demand description text, wherein the business operation task comprises an atomic operation, a standard step stream or a to-be-executed subtask; metadata of the business operation task is obtained, execution information corresponding to the business operation task is determined according to the metadata, and the execution information comprises an execution end and an execution time sequence; and completing the business operation task according to the execution information. By using the method, a redundancy planning link of a high-frequency simple demand is reduced, rapid response to the high-frequency simple demand and optimization of resource consumption are realized, and the problem that when an undefined complex composite task is faced, the task efficiency is greatly improved is solved. The method solves the problems of task matching failure and automatic business process stagnation caused by a scheme without automatic understanding and disassembling complex requirements, and improves the reasonability of resource allocation of an execution end by accurately determining execution information, thereby improving the efficiency of business automatic execution and the resource utilization rate.

Description

Technical Field

[0001] This invention relates to the field of computer processing technology, and in particular to a business processing method and apparatus. Background Technology

[0002] In the process of enterprise digital transformation, business automation has become a core requirement for cost reduction and efficiency improvement. Robotic Process Automation (RPA) is a technology that can automatically execute repetitive tasks with high efficiency and precision. In practical applications, RPA significantly improves business processing efficiency by automating the execution of standardized processes. Conversely, standardized processes provide a standardized input framework for the automated execution of robotic processes, and the two have a natural synergistic relationship.

[0003] However, existing business automation methods involving standardized processes and robotic process automation have many limitations in practical applications. On the one hand, their autonomous dynamic planning capabilities are insufficient. Existing methods can only handle predefined standardized workflows. When no business process matching the needs of the interaction object is found, they rely entirely on manual filling of knowledge gaps, without any automated fallback solution. This is especially problematic in scenarios with delayed user responses, leading to a complete halt in the process and severely impacting the quality of the closed-loop completion of the business process.

[0004] On the other hand, the allocation of execution-side resources is unreasonable. Existing methods usually optimize the utilization of remote resources based on subtask priorities and dependencies, but they do not consider the compatibility between the characteristics of the business operation tasks themselves and the execution-side resources. This leads to problems such as rough collaboration between local and remote resources and unreasonable resource allocation, which in turn causes resource waste or task blockage and affects the efficiency of business process processing. Summary of the Invention

[0005] This invention provides a business processing method and apparatus to solve the problems of insufficient autonomous dynamic planning capability and unreasonable resource allocation at the execution end in existing business automation processing methods.

[0006] In a first aspect, embodiments of the present invention provide a business processing method, the method comprising: Obtain the business requirement description text; Determine the business operation task corresponding to the business requirement description text, wherein the business operation task includes atomic operations, standard step flow, or subtasks to be executed; Obtain the metadata of the business operation task, and determine the execution information corresponding to the business operation task based on the metadata. The execution information includes the execution terminal and the execution sequence. Based on the execution information, complete the business operation task.

[0007] Secondly, embodiments of the present invention provide a business processing apparatus, the apparatus comprising: The acquisition module is used to acquire the business requirement description text; The task determination module is used to determine the business operation task corresponding to the business requirement description text. The business operation task includes atomic operations, standard step flow, or subtasks to be executed. An execution information determination module is used to obtain the metadata of the business operation task and determine the execution information corresponding to the business operation task based on the metadata. The execution information includes the execution end and the execution sequence. The task execution module is used to complete the business operation task based on the execution information.

[0008] The technical solution of this invention involves obtaining a business requirement description text; determining the corresponding business operation task, which includes atomic operations, standard step flows, or subtasks to be executed; obtaining the metadata of the business operation task; and determining the execution information corresponding to the business operation task based on the metadata, including the execution terminal and execution sequence; and completing the business operation task based on the execution information. This method, by determining a business operation task precisely matched to the business requirement description text from three types of business operation tasks, reduces redundant planning steps for high-frequency, simple requirements, achieves rapid response and resource consumption optimization for high-frequency, simple requirements, and solves the problem of task matching failure and automated business process stagnation caused by the lack of automated fallback and decomposition solutions for undefined complex composite tasks, thus improving the automated closed-loop completion capability of the business process. Furthermore, by utilizing the metadata of the business operation task to accurately determine the execution information, including the execution terminal and execution sequence, the rationality of execution terminal resource allocation is improved, thereby enhancing the efficiency and resource utilization of automated business execution.

[0009] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

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

[0011] Figure 1 A flowchart of a business processing method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a business processing device provided in an embodiment of the present invention; Figure 3 A schematic diagram of an electronic device that can be used to implement embodiments of the present invention is shown. Detailed Implementation

[0012] 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.

[0013] 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.

[0014] It is understood that before using the technical methods disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0015] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose, based on the prompt message, whether to provide personal information to the software or hardware such as the electronic device, application, server, or storage medium performing the operations of this disclosed technology.

[0016] As an optional but non-limiting implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0017] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0018] Specifically, embodiments of the present invention provide a business processing method. Figure 1 This is a flowchart of a business processing method provided by an embodiment of the present invention. The embodiment of the present invention is applicable to scenarios of automated business processing. The method can be executed by a business processing device, which can be implemented in the form of software and / or hardware. Optionally, it can be implemented by an electronic device, preferably a mobile terminal, desktop computer, laptop computer, or server.

[0019] like Figure 1 As shown, the business processing method provided in this embodiment of the invention may specifically include: S101. Obtain the business requirement description text.

[0020] The business requirement description text can be understood as text submitted in natural language or structured form to describe business requirements, such as "Requesting 3 days of annual leave from November 1st to November 3rd, 2025".

[0021] In this embodiment, the business requirement description text can be obtained by detecting an editing operation on the editing area of ​​the current interface, and then determining the business requirement description text based on the editing operation. The editing operation can be the user inputting the business requirement description text, or the user selecting the business requirement description text from a drop-down box, etc.

[0022] S102. Determine the business operation tasks corresponding to the business requirement description text. Business operation tasks include atomic operations, standard step flows, or sub-tasks to be executed.

[0023] Atomic operations can be considered as standardized, directly invoked units encapsulated from frequently occurring, single-step operations with no cross-operation dependencies (such as "clicking a button" or "entering text"). Each atomic operation must include a semantic description and execution parameters. The semantic description can be considered a precise definition of the operation's function using natural language, such as "input: write preset text content to the specified input box," used to match the business requirement description text. Execution parameters can be understood as the key parameters required for the preset operation; for example, "click" must include "interface elements" to ensure the operation can be executed directly.

[0024] A standard workflow can be considered a pre-set chain of complete process steps for business needs with clearly defined standardized procedures. For example, for leave application needs, the corresponding standard workflow could be: [Open the enterprise OA system, enter the leave application module, select the annual leave type, fill in the leave start and end dates (November 1, 2025 to November 3, 2025), fill in the reason for leave, submit the application form, the background pushes the approval to the superior, and notifies the employee that the application has been approved].

[0025] A subtask to be executed can be understood as a subtask to be executed at the current time step after breaking down a complex, non-fixed or undefined composite requirement into multiple ordered discrete subtasks.

[0026] In this embodiment, the method for determining the business operation task corresponding to the business requirement description text can be as follows: First, the business requirement description text is analyzed, and a pre-built atomic operation library can be used to determine whether the business requirement description text matches an atomic operation. If it matches, the business operation task is determined based on the matched atomic operation; otherwise, a pre-built standardized process knowledge graph can be searched to determine whether a standard step flow matching the business requirement description text exists. If it exists, the standard step flow is determined as a standard step flow; if it does not exist, the business requirement description text can be flexibly decomposed into a coherent sequence of subtasks based on a preset strategy, and the currently suitable subtask to be executed is determined from the subtask sequence, and the subtask to be executed is determined as the business operation task. The three types of business operation tasks cover simple, standard, and complex business requirement scenarios.

[0027] S103. Obtain the metadata of the business operation task, and determine the execution information corresponding to the business operation task based on the metadata. The execution information includes the execution end and the execution sequence.

[0028] The metadata of a business operation task may include its identification information, initiation time, and the process chain it belongs to. The execution end may include local terminals and cloud terminals; the execution sequence may include synchronous execution and asynchronous execution.

[0029] In this embodiment, the method for determining the execution terminal corresponding to a service operation task based on metadata can be as follows: Determine the attribute tags of the service operation task based on a predefined mapping table, where the mapping table represents the correspondence between the operation task type and the attribute tags, and the attribute tags include execution terminal adaptation attributes and real-time attributes; determine the execution terminal based on the execution terminal adaptation attributes in the attribute tags. Alternatively, the method for determining the execution terminal corresponding to a service operation task based on metadata can be as follows: Determine the attribute tags of the service operation task based on a predefined mapping table, and determine the execution terminal based on the execution terminal adaptation attributes in the attribute tags and the current network environment awareness data.

[0030] One way to determine the execution sequence of business operation tasks based on metadata is to determine the execution sequence based on the real-time attributes in the attribute tags and / or the dependencies of the preceding process chains.

[0031] S104. Complete the business operation task based on the execution information.

[0032] In this embodiment, business operation tasks can be directly assigned to corresponding execution terminals based on execution information. The execution terminals then complete the corresponding business operation tasks based on the execution sequence, essentially converting operations, steps, or tasks into specific interactive operations with the user interface. For example, if a business operation task is an atomic operation, the execution terminal can directly execute the atomic operation based on the execution sequence. If a business operation task is a standard step flow, the corresponding execution information can be determined for each standard step in the standard step flow, and each standard step can be executed according to the execution information in the order of the standard step flow. If a business operation task is a subtask to be executed, the execution terminal can execute the target action corresponding to the subtask to be executed based on the execution sequence, and after completing this step, return to step S102 to re-execute the relevant steps of the subtask to be executed corresponding to the business requirement description text, until the business process termination condition is met.

[0033] The business processing method provided in this invention involves: acquiring a business requirement description text; determining the business operation task corresponding to the business requirement description text, whereby the business operation task includes atomic operations, standard step flows, or subtasks to be executed; acquiring the metadata of the business operation task and determining the execution information corresponding to the business operation task based on the metadata, whereby the execution information includes the execution end and execution sequence; and completing the business operation task based on the execution information. This method, by determining a business operation task precisely matched to the business requirement description text from three types of business operation tasks, reduces redundant planning steps for high-frequency, simple requirements, achieves rapid response and resource consumption optimization for high-frequency, simple requirements, and solves the problem of task matching failure and automated business process stagnation caused by the lack of automated fallback and solutions for breaking down complex requirements when facing undefined complex composite tasks, thereby improving the automated closed-loop completion capability of the business process. Furthermore, by utilizing the metadata of the business operation task to accurately determine the execution information, including the execution end and execution sequence, the rationality of execution end resource allocation is improved, thereby enhancing the efficiency and resource utilization of automated business execution.

[0034] As a first optional embodiment of the present invention, based on the above embodiments, the step of determining the business operation task corresponding to the business requirement description text can be specified as the following steps: a1) Match the business requirement description text based on a pre-defined regular expression to determine whether the matching result meets the preset first matching condition.

[0035] In this embodiment, after obtaining the business requirement description text, regular expressions can be used to match the business requirement description text. If the matching result matches two or more actions, it means that the business requirement description text corresponds to a compound operation, and the matching result is considered not to meet the preset first matching condition. If the matching result matches a single action, the matching result is considered to meet the first matching condition.

[0036] b1) If the matching result satisfies the first matching condition, the word vector of the business requirement description text is determined, and the operation hit result is determined based on the word vector and the pre-built atomic operation library; if the operation hit result is a hit atomic operation, the business operation task is determined based on the hit atomic operation.

[0037] It is known that existing intelligent agents generally suffer from response chain redundancy when processing users' natural language requests. That is, regardless of whether the operation corresponding to the business request is a high-frequency and simple operation, the entire chain must be executed, resulting in high response latency for high-frequency and simple requests. In contrast, humans develop subconscious and highly proficient execution habits for basic operations such as "clicking buttons" and "entering text" through long-term use of intelligent devices, and can complete them efficiently without additional thought.

[0038] Therefore, in this embodiment, an atomic operation library can be pre-constructed to encapsulate high-frequency, simple operations into standardized, directly callable units, thereby shortening the response chain. This library stores a collection of reusable, directly executable atomic operations with standardized semantic descriptions.

[0039] In this embodiment, if the matching result meets the first matching condition, it can be further determined whether the business operation task corresponding to the business requirement description text is an atomic operation.

[0040] For example, the business requirement description text can be converted into word vectors using a pre-trained semantic model (such as BERT, Word2Vec, etc.); similarly, the semantic description of each atomic operation in the atomic operation library is converted into a semantic vector. Next, the cosine similarity between the word vectors and each semantic vector is calculated; if there is a semantic vector with a cosine similarity greater than or equal to a preset similarity threshold, the operation hit result is determined to be a hit atomic operation, that is, the business requirement description text hits the atomic operation corresponding to that semantic vector.

[0041] In this embodiment, the method for determining the business operation task based on the hit atomic operation can be as follows: if the business requirement description text hits an atomic operation, then the atomic operation is determined as the business operation task; if the business requirement description text hits multiple atomic operations, then the atomic operations can be prioritized according to their execution success rate, and the atomic operation with the highest execution success rate is determined as the business operation task.

[0042] In an optional embodiment, after determining the business operation task based on the hit atomic operation, parameters can be extracted from the business requirement description text and filled into the execution parameter list of the business operation task, such as the "search button" in "click the search button" corresponding to the "interface element" of "click", to ensure that the business operation task can be executed accurately.

[0043] c1) If the matching result does not meet the first matching condition, or if the operation hit result is a no-atomic-operation hit, then by querying the pre-built standardized process knowledge graph, it is determined whether there is a standard step flow that matches the business requirement description text, and if there is, the standard step flow is identified as a business operation task.

[0044] In this embodiment, when the business requirement description text does not match an atomic operation, a pre-built standardized process knowledge graph can be retrieved, and the professional knowledge and normative constraints of the vertical industry provided by the standardized process knowledge graph can be utilized.

[0045] In this embodiment, the method for determining the standard step flow matching the business requirement description text by querying a pre-built standardized process knowledge graph can be as follows: Combining the business requirement description text with a preset query language (such as Cypher query language), entities and relationships are filtered from the standardized process knowledge graph to form an initial retrieval result set; a graph traversal algorithm is used to retrieve nodes and edges related to the initial retrieval result set to obtain a complete step flow; the step flow is then integrated into a standardized format to obtain the standard step flow matching the business requirement description text. For example, the standard JSON format for "click" is {"action": "move to an element and click","args": {"x": x, "y": y, "button": "left"}}.

[0046] It can be seen that if there are no entities and relationships in the standardized process knowledge graph that match the business requirement description text, that is, the initial search result set is empty, it indicates that there is no standard step flow in the standardized process knowledge graph that matches the business requirement description text.

[0047] Optionally, the construction steps of the standardized process knowledge graph include: Based on preset pattern information, a set of triples is extracted from the text corresponding to at least one standardized process. The pattern information includes entity type constraint information, entity relationship constraint information, and entity key attribute constraint information. For each entity, a first embedding vector of the entity and a second embedding vector of the relation corresponding to the entity are determined by a preset embedding algorithm, the triples associated with the entity, and a trained graph convolutional network; wherein, the graph convolutional network is determined by iterative training using a positive sample set composed of collected real triples and a negative sample set composed of fake triples constructed according to a preset construction strategy. For each triple in the set of triples, the evaluation function value of the triple is determined based on the first embedding vector of the head entity, the first embedding vector of the tail entity, and the second embedding vector of the relation, combined with a preset evaluation function. Based on the target triples whose evaluation function values ​​satisfy the quality assessment conditions, a standardized process knowledge graph is constructed.

[0048] In this embodiment, to construct a standardized process knowledge graph, the pattern information of the knowledge graph is first preset. Then, triples are extracted from the text corresponding to the standardized process based on the pattern information to form a triple set. The pattern information can be considered as the data structure of the semantic skeleton of the knowledge graph and the constraint benchmark for triple extraction.

[0049] The schema information can specifically include: a conceptual layer, a relational constraint layer, and an attribute definition layer. The conceptual layer defines the types of entities to be extracted in the standardized process, such as "operation carrier," "operation object," and "operation parameter." The relational constraint layer predefines the legal association types between entities, such as "operation behavior - click." It clarifies the mapping rules between "head entity type" and "tail entity type" for each type of relationship; for example, the head entity for "operation behavior - click" must be of the "operation carrier" class, and the tail entity must be of the "operation object" class. The attribute definition layer specifies the key attributes of entities; for example, "operation carrier - webpage" must include the "URL" attribute, providing a basis for accurate entity identification.

[0050] In this embodiment, the triple can be considered as structured data used to represent the semantic associations in the text corresponding to the standardized process. Its fixed structure is (head entity, relation, tail entity). The head and tail entities both conform to the entity type constraints defined in the conceptual layer of the schema information and the key attribute constraints defined in the attribute definition layer, meaning they conform to the types of entities to be extracted in the defined standardized process and carry key attribute information. The relation conforms to the inter-entity relation constraints defined in the relation constraint layer of the schema information, meaning it conforms to the semantic association type and forms a valid mapping with the types of the head and tail entities.

[0051] In this embodiment, after obtaining a set of triples consisting of a series of triples, to address the problem that the triple set may contain erroneous knowledge (such as entity type mismatch, semantic contradiction of relations, etc.), a method that integrates knowledge graph embedding algorithm and graph convolutional network local neighbor aggregation mechanism can be adopted to capture the structural features and semantic association features of entities and evaluate the triples to ensure the quality of the knowledge graph.

[0052] In this embodiment, the first embedding vector can be understood as the embedding vector corresponding to the entity, and the entity may include the head entity and the tail entity; the second embedding vector can be considered as the embedding vector corresponding to the relation.

[0053] As one implementation method, for each entity, the first embedding vector of the entity and the second embedding vector of the relationship corresponding to the entity can be determined by using a preset embedding algorithm, the triplet associated with the entity, and a trained graph convolutional network. This can be achieved by obtaining the triplet associated with each entity and classifying each triplet into two categories: inbound triplets and outbound triplets.

[0054] For example, a triple can be represented as ,in Represents the head entity, Represents the relationship between entities. Represents the tail entity. For a set of triples. For entities The incoming triplet and the outgoing triplet can be represented as follows: ; ; in, For entities The corresponding incoming triples; For entities The corresponding out-edge triplet.

[0055] Next, a fusion knowledge graph embedding algorithm can be used, leveraging the neighbor... Embedded and relationships Embedded Estimate Entity The embedding, and based on the aggregation idea of ​​graph convolutional networks and the residual connection idea of ​​ResNet, for entity... The Layer embedding It aggregates information from its neighbors, embeds the aggregated information with the entity itself, and obtains the entity in the first... The layer's embedding vectors avoid gradient vanishing and preserve the basic semantics.

[0056] For example, determine the entity In the Layer embedding vector The specific method can be expressed as: ; in, Typically an entity The degree; A shared layer-specific matrix; For the neighbors In the Layer embedding vector; For relationship In the Layer embedding vector. It is a ReLU activation function, usually Choose 1 or 2.

[0057] In this embodiment, after determining the entity embedding, the update of the relation embedding can be determined by inter-layer linear projection to obtain the relation. In the Layer embedding vector This enables synchronous updates of relation embedding and entity embedding, ensuring they adapt to each other within the same vector space. Specifically, for relations... The Layer embedding Through linear transformation matrix Projected to the Layer space. For example, defining relationships. In the Layer embedding vector The method can be expressed as: .

[0058] In this embodiment, by repeating the above update iteration steps, the embedding vector of each entity in the last layer space can be determined, denoted as the first embedding vector, and the embedding vector of the relationship corresponding to the entity in the last layer space can be determined, denoted as the second embedding vector.

[0059] Next, for each triple, an evaluation function value is determined based on the first embedding vector of the head entity, the first embedding vector of the tail entity, and the second embedding vector of the relation, combined with a preset evaluation function. This evaluation function value is used to measure the reasonableness of the triple. Optionally, the evaluation function can be determined using L2 distance, based on the idea that "the embedding vector of the head entity plus the embedding vector of the relation should be as close as possible to the embedding vector of the tail entity in space." For example, the evaluation function... Specifically, it can be expressed as: ; Among them, the The layer space can be considered as the last layer of space. For triples Middle head entity The first embedding vector, For triples Mid-tail entity The first embedding vector; For triples China-US relations The second embedding vector.

[0060] In this embodiment, the target triplet can be considered as a triplet whose evaluation function value satisfies the quality assessment conditions, that is, a reasonable and effective triplet that passes the quality assessment. In an optional embodiment, the target triplet can be determined based on the evaluation function value and the quality assessment conditions as follows: if the evaluation function value of the triplet is less than a preset evaluation threshold, then the evaluation function value is determined to satisfy the quality assessment conditions, and the triplet is determined to be the target triplet; otherwise, the evaluation function value of the triplet is determined not to satisfy the quality assessment conditions, and the triplet is ignored.

[0061] In this embodiment, the standardized process knowledge graph can be constructed based on the target triples as follows: CSV files are used for intermediate storage, and Python and open-source libraries are used to read and preprocess the CSV files. The Neo4j graph database is used as the storage medium for the knowledge graph. A connection is established with the database through the Neo4j Python driver library provided by Neo4j. Knowledge graph construction algorithms are used to create nodes and edges in Neo4j, forming a network knowledge structure to obtain the standardized process knowledge graph.

[0062] In this embodiment, the graph convolutional network is pre-trained. The training set used for training the graph convolutional network includes a positive sample set and a corresponding negative sample set. The positive sample set consists of real triples collected in practice. A false triplet is constructed for each real triplet based on a preset construction strategy to form the negative sample set. For example, the preset construction strategy can be: replacing only one of the head or tail entities, without replacing both simultaneously. For example, the loss function used for training the graph convolutional network... It can be represented as: ; in, This is the set of spurious triples, also known as the negative sample set. This is the set of triples, also known as the positive sample set; Positive sample The corresponding negative sample; To pre-determine the boundaries, .

[0063] The technical solution described in this embodiment employs a pre-defined knowledge graph embedding algorithm, using a single triple as the smallest learning unit to map entities and relations to a low-dimensional vector space, and learns the embedding representation through pre-defined vector constraint rules. Secondly, an entity association semantic mining module is constructed by introducing a Graph Convolutional Network (GCN). Using the topological structure of the knowledge graph as input, it mines the indirect semantic relationships between entities, supplementing the semantic gaps in single triple modeling. Furthermore, based on the simultaneous learning of entity and relation embeddings, a quantifiable triple quality evaluation function is constructed, improving the average recall rate of knowledge graph quality evaluation in real-world business scenarios.

[0064] d1) In the absence of the standard step flow, multiple agents determine the sub-tasks to be executed by time step and identify the sub-tasks to be executed as business operation tasks.

[0065] In this embodiment, if a standard step flow does not exist, it indicates that the business operation task corresponding to the business requirement description text is a complex, non-fixed process, or undefined composite requirement. In this case, to solve the problems of task matching failure and automated business process stagnation caused by the lack of automated fallback and complex requirement decomposition solutions when facing undefined complex composite tasks, as well as the problems of low execution efficiency and easy accuracy bottlenecks of single agents, accurate planning of business operation tasks can be achieved through multi-agent interaction.

[0066] As one implementation, the plurality of intelligent agents includes a master intelligent agent, an executive intelligent agent, an evaluation intelligent agent, and a memory intelligent agent.

[0067] The system comprises the following components: a master agent, an execution agent, and a memory agent. The master agent dynamically updates the sequence of subtasks to be executed based on the business requirements described in the business requirement description text and the search results. The execution agent determines the subtasks to be executed at the current time step and their corresponding execution action information based on the current interface state and the subtask sequence. The execution action information may include the thought process for selecting the subtask, the target action, and the expected effect of the action. The evaluation agent compares the expected effect of the action with the actual changes in the interface state to determine whether the subtask was successfully executed and generates diagnostic feedback. The memory agent extracts key information when a subtask is successfully executed and stores it as global memory for use in subsequent task planning and execution processes.

[0068] Accordingly, the process of determining the sub-tasks to be executed by multiple agents according to time steps can be further specified as the following steps: For the first time step, the main intelligent agent determines the sequence of subtasks to be executed for the current time step based on the business requirement description text and the interaction interface status information of the current time step.

[0069] For each time step except the first time step, the main intelligent agent determines the sequence of subtasks to be executed in the current time step based on the business requirement description text, the interactive interface status information of the current time step, and the sequence of subtasks to be executed, the sequence of completed subtasks, the execution action information, the evaluation feedback, and the accumulated memory of the previous time step. The execution agent determines the sub-tasks to be executed at the current time step based on the business requirement description text, the sequence of sub-tasks to be executed at the current time step, the status information of the interactive interface, and the evaluation feedback and accumulated memory of the previous time step.

[0070] Among them, the current time step Interactive interface status information It can be represented in the form of a high-dimensional tensor; for example, it can be represented as... Where H, W, and C represent the height, width, and channel dimension of the screenshot of the interactive interface state, respectively. Previous time step Sequence of subtasks to be executed It can be represented as an ordered sequence of multiple subtasks to be executed. For example, it can be specifically represented as: . A natural language text description of the discrete steps to achieve business requirements. A sequence of completed subtasks from the previous time step. It can be considered as a sequence of subtasks that have been successfully executed and evaluated, which can avoid redundant operations and track task progress.

[0071] The execution action information can be a structured tuple. For example, the execution action information from the previous time step can be specifically represented as follows: ,in, The thought process for selecting a specific subtask to be executed in the previous time step. The target action of the previous time step, The expected effect of the action after the target action is performed.

[0072] The evaluation feedback from the previous time step can be generated by observing the execution result of the target action in the previous time step, and can be a tuple. For example, the evaluation feedback from the previous time step... This can be specifically expressed as ,in, To determine the execution result, it can be represented as follows: , This is for diagnostic analysis results. The memory information from the previous time step can be considered as a collection of key information accumulated from the first time step to the previous time step. For example, the previous time step... Memory information It can be represented as: ,in, For the first The key information remembered at each time step for The memory information of time steps For the first The key information remembered at each time step.

[0073] In this embodiment, for the first time step, the main agent can decompose the business requirement description text into a coherent, ordered sequence of subtasks based on the current interactive interface state information, which serves as the sequence of subtasks to be executed. In subsequent time steps, the main agent can dynamically update the sequence of subtasks to be executed based on the latest execution status.

[0074] As described above, after obtaining the sequence of subtasks to be executed at the current time step, the executing agent can view the first preset number of subtasks in the sequence, and simultaneously analyze the business requirement description text and the interactive interface status information of the current time step to determine which subtask is most relevant to the current task and executable, and then designate that subtask as the subtask to be executed at the current time step. During this process, the evaluation feedback from the previous time step and accumulated memories are referenced to avoid repeating mistakes.

[0075] It should be noted that after determining the subtask to be executed at the current time step, the executing agent can output the execution action information for the current time step, including the thought process of selecting the current time step, the target action of the corresponding subtask to be executed, and the expected effect of the action, so as to provide support for the subsequent evaluation feedback of the current time step.

[0076] Optionally, the sequence of completed subtasks at each time step is determined by the main agent by identifying the subtasks that have been successfully executed in the sequence of subtasks to be executed at that time step; The execution action information for each time step is determined by the executing agent based on the sub-task to be executed. The execution action information includes the thought process for selecting the sub-task to be executed, the target action, and the expected effect of the action. The evaluation feedback for each time step is determined by the evaluation agent based on the business requirement description text, the interactive interface state information before the target action is executed, the interactive interface state information after the target action is executed, and the execution action information. The accumulated memory at each time step is determined by the memory agent based on the interactive interface state information of the previous time step, the interactive interface state information of the current time step, and the accumulated memory of the previous time step.

[0077] In this embodiment, the evaluation agent is a key component for ensuring the robustness of the multi-agent system and its ability to learn from errors. It is responsible for comparing the state transitions of the interactive interface with the intention (expected effect of the action) of the executing agent to evaluate the utility of the target action. Specifically, the evaluation agent analyzes the business requirement description text, the interactive interface state information before the target action is executed, the interactive interface state information after the target action is executed, and the execution action information at the current time step. If the changes in the interactive interface state before and after the target action are consistent with the thinking process of the executing agent, the evaluation agent will evaluate the effectiveness of the target action. and expectations If the progress of the tasks reflected in the data remains consistent, then it can be determined that... Conversely, if the interactive interface encounters an error, remains unchanged, or switches to an irrelevant state, the target action is considered to have failed. .like , It will provide diagnostic analysis, and this feedback will provide a basis and support for the main agent to dynamically update the sequence of subtasks to be executed.

[0078] To better understand this embodiment, an example of a multi-intelligent interaction process in a scenario where an employee applies for 3 days of annual leave through the enterprise OA system is given here. The specific process is shown in Table 1.

[0079] Table 1 In this embodiment, if the target action at the previous time step was successfully executed, that is, in the evaluation feedback... At the current time step, the main agent identifies completed subtasks from the previous time step's sequence of subtasks to be executed and moves them from the previous time step's sequence of subtasks to the sequence of completed subtasks, forming the initial subtask sequence corresponding to the current time step. Next, the remaining subtasks in the initial subtask sequence are reordered by priority to obtain the current time step's sequence of subtasks to be executed. If the target action fails to execute (… The main intelligent agent will utilize Revise the sequence of subtasks to be executed. Revision operations include reordering subtasks, modifying existing subtasks, inserting new subtasks, and / or reverting to the previous strategy.

[0080] In this embodiment, the memory agent is used to handle situations where the current interactive interface state information is needed later. The memory agent only activates when the target action is successfully executed ( When triggered, this information is identified and retained. The memory agent analyzes the state information transfer process of the interactive interface, filters out information important to the current task, and stores this information in a structured manner in the accumulated memory. Subsequently, the accumulated memory will be provided to the main agent and the executing agent to provide a basis for long-term planning and execution.

[0081] Understandably, multi-agent interaction is a periodic, state-driven process that begins with a description of business requirements and terminates when a task is completed or deemed impossible to complete. The master agent defines the initial sequence of subtasks to be executed. Then it enters an iterative loop. For each time step... The executing agent selects and executes a subtask from the sequence of subtasks to be executed, generates execution action information, and the interactive interface state information is then transferred to... The evaluation agent assesses the transfer process and generates evaluation feedback. If the target action is successfully executed, the memory agent may record and store relevant information in accumulated memory. Finally, the main agent can update the sequence of completed subtasks and the sequence of subtasks to be executed based on the evaluation feedback. By dynamically planning the workflow through multi-agent interaction, the challenges of breaking down long-cycle tasks and handling abnormal scenarios are solved. This structured, evaluative, and adaptive loop enables the handling of complex interaction sequences, the processing of unexpected events, and the robust advancement of the target business.

[0082] As a second optional embodiment of the present invention, when the business operation task is an atomic operation or a subtask to be executed, the step of determining the execution information corresponding to the business operation task based on the metadata can be further optimized to the following steps: Based on the pre-built task attribute library and the metadata, target attribute tags that match the business operation task are determined by keyword recognition or process template matching. The task attribute library includes a mapping table between operation task types and attribute tags. The attribute tags include local adaptation attributes, cloud adaptation attributes, high real-time attributes, and low real-time attributes. If the target attribute tag includes a local adaptation attribute, the execution end of the business operation task is determined to be a local terminal; If the target attribute tag includes a cloud adaptation attribute, the execution end of the business operation task is determined based on the current environment perception data; When the target attribute label includes a high real-time attribute and / or it is determined that the business operation task has a preceding process chain dependency, the execution sequence of the business operation task is determined to be synchronous execution; When the target attribute label includes a low real-time attribute and it is determined that the business operation task has no preceding process chain dependency, the execution sequence of the business operation task is determined to be asynchronous.

[0083] It should be noted that before processing business operation tasks, a task attribute library needs to be built to provide a basis for subsequent decision-making. The task attribute library can be used to predefine attribute tags for various operation tasks, forming a "operation task type - attribute tag" mapping table.

[0084] For example, the attribute tags corresponding to operation task types that do not depend on local software or hardware and can be completed independently over the network (such as sending emails and pushing messages) can be determined as cloud-adaptive attributes; the attribute tags corresponding to operation task types that require calling local software (such as Office and browsers), depend on local hardware (such as printers), and / or require operating on local interfaces (such as clicking on interfaces and saving files) can be determined as local-adaptive attributes; the attribute tags corresponding to operation task types such as instant message push can be determined as high-real-time attributes; and the attribute tags corresponding to operation task types such as non-urgent email sending can be determined as low-real-time attributes.

[0085] In this embodiment, keyword recognition or process template matching can be used to search for operation task types that match the business operation task in the task attribute library. The attribute tags corresponding to the operation task type are then determined based on a mapping table, and these attribute tags are identified as the target attribute tags for the business operation task. For example, the attribute tags for the business operation task "send sales report email" are cloud-adaptive attributes and low real-time attributes.

[0086] In this embodiment, decisions regarding the execution end and execution order can be made based on the target attribute tags. If the target attribute tags include local adaptation attributes, regardless of the current environmental awareness data, the execution end is preferentially determined to be the local terminal. When local resources are scarce, execution can wait for resources to be released before proceeding, thus avoiding execution failure due to the cloud terminal being unable to access local resources.

[0087] In this embodiment, if the target attribute tag includes a cloud adaptation attribute, the current environment awareness data can be obtained through methods such as data collection. The current environment awareness data can include network status data (such as bandwidth, latency, and the number of disconnections within a time period) and local resource data (such as CPU utilization, memory usage, and local software permissions). The method for determining the execution terminal of the service operation task based on the current environment awareness data can be as follows: Based on the current environment awareness data and corresponding preset evaluation rules, determine the network environment evaluation result; if the network environment evaluation result indicates a good network environment, then determine that the execution terminal of the service operation task is a cloud terminal; otherwise, determine that the execution terminal of the service operation task is a local terminal.

[0088] In this embodiment, supported by a task attribute library and current environment awareness data, multi-dimensional collaborative execution covering local and cloud environments, including synchronous and asynchronous operations, is achieved. On one hand, execution terminals are dynamically allocated based on the execution terminal adaptation attributes and execution environment awareness of the business operation tasks. On the other hand, the execution sequence is determined by combining the real-time requirements and process dependencies of the business operation tasks, avoiding task blocking, improving resource utilization, and ensuring that each step can efficiently and flexibly handle tasks, adapting to the standardized task characteristics of robotic process automation.

[0089] As a third optional embodiment of the present invention, when the business operation task is a standard step flow, the step of determining the execution information corresponding to the business operation task based on the metadata includes: Based on the pre-built task attribute library and the metadata, the target attribute tag sequence that matches the standard step flow is determined by keyword recognition or process template matching, with one target attribute tag corresponding to one standard step; For each target attribute tag in the target attribute tag sequence, the execution end and execution sequence of the corresponding standard step are determined based on the target attribute tag.

[0090] The target attribute label sequence can be considered as a sequence of target attribute labels arranged in the order of standard steps in the standard step flow, corresponding to the standard step flow.

[0091] In this embodiment, the execution end and execution sequence of each standard step can be determined based on the execution end adaptation attributes and environment awareness of each standard step in the standard step flow, as well as the real-time attributes and process dependencies. This ensures that each standard step can be processed efficiently and flexibly, thereby improving the execution efficiency and resource utilization of subsequent standard step flows.

[0092] As a fourth optional embodiment of the present invention, based on the above embodiments, the step of completing the business operation task according to the execution information can be specified as the following steps: Based on the execution terminal in the execution information, the business operation task is assigned to the corresponding terminal; Based on the execution sequence in the execution information, the terminal determines the task scheduling strategy corresponding to the business operation task, and executes the business operation task according to the task scheduling strategy. If the business operation task is a sub-task to be executed, return to re-execute the relevant steps for determining the business operation task corresponding to the business requirement description text, until the business process termination condition is met.

[0093] In this embodiment, if the execution end is a cloud terminal, and the execution sequence is asynchronous, the business operation task can be encapsulated into a cloud asynchronous task package and sent to the cloud task pool through an encrypted interface. The cloud allocates computing power according to the "first-come, first-served" principle. If the business operation task is urgent, its priority can be increased through a real-time tag. If the execution sequence is synchronous, the cloud can enable a fast response channel to prioritize the allocation of resources to the business operation task and send the execution result back to the local terminal in real time.

[0094] If the execution end is a local terminal, and the execution sequence is synchronous, the local robot process automation execution engine can be called to directly trigger the software and hardware interfaces, and the progress will be fed back in real time during the execution process. If the execution sequence is asynchronous, the business operation tasks are added to the local asynchronous queue, and the task execution order is dynamically adjusted according to the resource utilization rate to avoid affecting the local synchronous tasks.

[0095] Optionally, if multiple standard steps or business operation tasks in a standard workflow have partial dependencies, a combination of synchronous and asynchronous execution can be used. For example, in the task chain of "report generation - local saving - cloud sending," "report generation - local saving" is executed synchronously, while "local saving - cloud sending" is executed asynchronously. When multiple independent standard steps are executed simultaneously, a resource isolation strategy can be adopted. For local synchronous tasks, local memory resources are prioritized, and local asynchronous tasks are paused; for cloud asynchronous tasks, cloud resources are used independently without affecting local tasks.

[0096] To better understand the above embodiments, a specific example of a leave request scenario is given here, as shown in Table 2. The preset evaluation rules corresponding to the environmental awareness data may include bandwidth (≥10Mbps is good, <2Mbps is poor), latency (≤100ms is low latency, >500ms is high latency), memory utilization (≤50% is sufficient, >70% is strained), etc. It may also include considering the network environment as good if the bandwidth is good and the latency is low.

[0097] Table 2 Optionally, the execution terminal can perform data synchronization after the business operation task is completed. Specifically, after the business operation task is completed, the local terminal automatically synchronizes the task result data to the cloud database. Incremental synchronization can be used, synchronizing only changed data to reduce bandwidth consumption. After the business operation task is completed, the cloud terminal synchronizes the execution result to the local robotic process automation log system.

[0098] In this embodiment, when the business operation task is a subtask to be executed, after the subtask to be executed at the current time step is completed, an iteration is completed. The process can be returned to re-execute the steps related to determining the business operation task corresponding to the business requirement description text, determine the subtask to be executed at the next time step and the corresponding execution information, and execute the subtask to be executed according to the execution information until the business process termination condition is met. The business process termination condition can be the realization of the target business requirement corresponding to the business requirement description text, the reaching of the maximum execution time, or the target business requirement being determined to be unable to be completed, etc.

[0099] As a fifth optional embodiment of the present invention, based on the above embodiments, further optimizations can be made including: Based on the preset extraction frequency, obtain log data during the business process. The atomic operation library is updated based on the preset first statistical indicator, first triggering condition and first update logic, combined with the log data. The first statistical indicator includes the frequency of misses, atomic operation reuse rate, complexity of decomposing compound operations and atomic operation execution success rate. Based on the preset second statistical indicators, second triggering conditions, and second update logic, the standardized process knowledge graph is updated in conjunction with the log data. The second statistical indicators include task repetition frequency, multi-agent planning complexity, and workflow stability.

[0100] It is known that existing static library (including atomic operation library and standardized process knowledge graph) update methods cannot optimize knowledge associations according to continuous changes in business needs, nor can they incorporate new business knowledge in the domain in a timely manner, nor can they adapt to the evolution of interactive object behavior habits. Furthermore, for high-frequency, repetitive, stable, and high-success-rate complex tasks, the agent's dynamic planning process needs to be restarted repeatedly each time, resulting in redundant consumption in the planning stage and increased overall system operating costs, which does not meet the business requirements of high efficiency and low cost. Therefore, this embodiment uses the above steps to dynamically and adaptively update the atomic operation library and standardized process knowledge graph.

[0101] The log data during the business process can include atomic operation library matching logs corresponding to atomic operation libraries, knowledge graph retrieval logs corresponding to standardized process knowledge graphs, multi-agent interaction logs corresponding to multi-agents, and business operation task execution logs provided by the execution end.

[0102] In this embodiment, a structured data source can be formed based on the data source type, data content, and extraction frequency, and the entire chain of data in the planning and execution process can be uniformly collected, providing a foundation for the adaptive updating of the static atomic operation library and the standardized process knowledge graph.

[0103] Table 3 For example, Table 3 shows a correspondence between data source type, data content and extraction frequency.

[0104] In this embodiment, the optimization goal of the atomic operation library is to improve the hit rate of atomic operations and reduce the cost of decomposing complex tasks. This is achieved through statistical analysis of high-frequency requirements and operational patterns, enabling three types of updates: addition, optimization, and elimination. For updating the atomic operation library, a first statistical indicator can be defined. This first statistical indicator can be understood as an indicator used to determine the atomic operations that can be updated in the atomic operation library. Specifically, the miss frequency can be considered as the frequency of atomic operations not being matched in the business requirement description text within a preset time period (e.g., the past 30 days); the atomic operation reuse rate can be understood as the ratio of the number of times a certain atomic operation is called within a preset time period to the total number of calls to all atomic operations; the complexity of decomposing complex operations can be considered as the number of subtasks into which a certain complex task is decomposed by the main intelligent agent; and the atomic operation execution success rate can be understood as the ratio of the number of times a certain atomic operation is successfully executed to the total number of calls to that operation.

[0105] The first triggering condition can be considered as the condition used to trigger the update of the atomic operation library, which may include adding atomic operation triggering conditions, optimizing existing atomic operation triggering conditions, and eliminating redundant atomic operation triggering conditions. The first update logic can be understood as the logic used to update the atomic operation library.

[0106] In this embodiment, the method for updating the atomic operation library based on the preset first statistical indicator, first triggering condition, and first update logic, combined with the log data, can be as follows: If business requirement description texts that do not match atomic operations form clusters, then the triggering condition for adding a new atomic operation can be considered met, and the new operation is triggered. For example, if the business requirement description text "send email" appears 20 times in the past 30 days and does not match any atomic operations, semantic analysis is performed on this type of business requirement description text. If it is a simple composite operation composed of existing atomic operations, then a new combined atomic operation is added, such as "enter date in text box". If a new operation unit exists, the new atomic operation is entered into the atomic operation library. At the same time, the atomic operation matching rules are updated (synchronized to the training data of the word vector model).

[0107] If the success rate of a certain atomic operation is below a threshold, and common errors exist in the evaluation feedback from the evaluation agent, then it can be considered that the triggering condition for optimizing the existing atomic operation has been met, and the optimization operation is triggered. For example, the atomic operation "clicking an interface button" has a success rate of only 75%, and the failure reason is "button coordinate recognition deviation." The failure diagnosis for this atomic operation can be extracted. Locate the cause of the error and optimize the execution parameters or matching logic, such as changing the parameter of "click button" from fixed coordinates to control ID.

[0108] If the reuse rate of an atomic operation is below a threshold, then the condition for eliminating redundant atomic operations can be considered met, and an elimination operation can be triggered. For example, if the reuse rate of an atomic operation is less than 1% in the past 30 days, then the record of that atomic operation in the atomic operation library can be deleted, and the relevant configuration in the atomic operation matching rules can be cleaned up.

[0109] The second statistical indicator can be considered as an indicator used to screen composite tasks that can be used for updating the standardized process knowledge graph. The second triggering condition can be considered as the condition used to trigger the update of the standardized process knowledge graph. The second update logic can be understood as the logic used to update the standardized process knowledge graph.

[0110] In this embodiment, the optimization goal of the knowledge graph is to reduce resource consumption in the planning stage, reuse high-frequency standardized processes, and ensure that not all composite tasks need to be entered into the knowledge graph. Based on this, a second statistical indicator can be set, including task repetition frequency, multi-agent planning complexity, and workflow stability. Task repetition frequency can be considered as the number of times composite tasks with the same goal appear within a preset time period under the same business scenario. Multi-agent planning complexity can be evaluated by the number of planned sub-tasks and the planning time. Workflow stability can be evaluated by the proportion of sub-tasks whose order remains unchanged in multiple plans of the same task and the execution success rate of the task.

[0111] If the task repetition frequency, multi-agent planning complexity, and workflow stability all exceed preset thresholds, the composite task can be considered to meet the second triggering condition and thus be designated as a candidate task. Candidate tasks must clearly define subtask logic, dependencies, and execution parameters, transforming them into standardized processes to ensure accurate retrieval and execution after being entered into the standardized process knowledge graph. Based on the subtask sequence from the multi-agent planning logs, the execution entity is completed by combining it with the business requirement description text, and parameters are extracted from the execution logs, supplementing the parameter sources (e.g., user input, automatic system acquisition, etc.). Furthermore, referring to the construction steps of the standardized process knowledge graph, based on preset pattern information, triples are extracted from the text corresponding to the standardized process, and then the triples are evaluated to ensure the quality of the knowledge graph. Finally, based on triples whose evaluation function values ​​meet the quality evaluation conditions, a connection is established with the database using Neo4j, and nodes and edges are created and the network knowledge structure is updated in Neo4j.

[0112] Optionally, to avoid decreased system stability due to adaptive updates, a mechanism of verification before implementation can be adopted. The specific process is as follows: Verification Environment Setup: Build a test environment consistent with the production environment, import the updated atomic operation library, standardized process knowledge graph, and historical task data. Validity Verification: Use historical business requirement description texts within a preset time period (e.g., the last 7 days) to test the atomic operation hit rate and standardized process knowledge graph retrieval success rate, and perform corresponding updates, observing the changes in hit rate and execution success rate. Rollback Trigger Conditions: If the hit rate and execution success rate decrease instead of increase, or if a new error occurs causing system crash, trigger a rollback.

[0113] The above-described technical solution in this embodiment achieves dynamic optimization of the atomic operation library and iterative upgrade of the standardized process knowledge graph by using full-link data based on multi-agent collaborative planning and multi-dimensional collaborative execution. This solves the problem that static libraries cannot adapt to changes in business scenarios or the evolution of interactive object behavior habits, reduces the cost of multi-agent interaction, and improves the execution success rate at the execution end.

[0114] Figure 2 This is a schematic diagram of the structure of a service processing device provided in an embodiment of the present invention. Figure 2 As shown, the device includes: an acquisition module 21, a task determination module 22, an execution information determination module 23, and a task execution module 24, wherein... Module 21 is used to obtain the business requirement description text; The task determination module 22 is used to determine the business operation task corresponding to the business requirement description text. The business operation task includes atomic operations, standard step flow, or sub-tasks to be executed. The execution information determination module 23 is used to obtain the metadata of the business operation task and determine the execution information corresponding to the business operation task based on the metadata. The execution information includes the execution end and the execution sequence. The task execution module 24 is used to complete the business operation task based on the execution information.

[0115] The technical solution described in this embodiment obtains a business requirement description text; determines the corresponding business operation task, which includes atomic operations, standard step flows, or subtasks to be executed; obtains the metadata of the business operation task, and determines the execution information corresponding to the business operation task based on the metadata, including the execution end and execution sequence; and completes the business operation task based on the execution information. Using this device, by determining a business operation task that precisely matches the business requirement description text from three types of business operation tasks, redundant planning steps for high-frequency, simple requirements are reduced, enabling rapid response and resource consumption optimization for high-frequency, simple requirements. Furthermore, it solves the problem of task matching failure and automated business process stagnation caused by the lack of automated fallback and complex requirement decomposition solutions when facing undefined complex composite tasks, thus improving the automated closed-loop completion capability of the business process. In addition, by using the metadata of the business operation task to accurately determine the execution information, including the execution end and execution sequence, the rationality of execution end resource allocation is improved, thereby increasing the efficiency and resource utilization of automated business execution.

[0116] Furthermore, the task determination module 22 may specifically include: The matching unit is used to match the business requirement description text based on a pre-defined regular expression and determine whether the matching result meets the preset first matching condition. The hit result determination unit is used to determine the word vector of the business requirement description text if the conditions are met, and to determine the operation hit result based on the word vector and the pre-built atomic operation library. The first task determination unit is used to determine the business operation task based on the hit atomic operation when the operation hit result is a hit atomic operation. The second task determination unit is used to determine whether there is a standard step flow that matches the business requirement description text by querying a pre-built standardized process knowledge graph if the conditions are not met or the operation hit result is a no-hit atomic operation, and to determine the standard step flow as a business operation task if it exists. The third task determination unit is used to determine the sub-tasks to be executed by multiple agents according to time steps in the absence of the standard step flow, and to determine the sub-tasks to be executed as business operation tasks.

[0117] Furthermore, the task determination module 22 also includes a map construction unit, which can be specifically used for: Based on preset pattern information, a set of triples is extracted from the text corresponding to at least one standardized process. The pattern information includes entity type constraint information, entity relationship constraint information, and entity key attribute constraint information. For each entity, a first embedding vector of the entity and a second embedding vector of the relation corresponding to the entity are determined by a preset embedding algorithm, the triples associated with the entity, and a trained graph convolutional network; wherein, the graph convolutional network is determined by iterative training using a positive sample set composed of collected real triples and a negative sample set composed of fake triples constructed according to a preset construction strategy. For each triple in the set of triples, the evaluation function value of the triple is determined based on the first embedding vector of the head entity, the first embedding vector of the tail entity, and the second embedding vector of the relation, combined with a preset evaluation function. Based on the target triples whose evaluation function values ​​satisfy the quality assessment conditions, a standardized process knowledge graph is constructed.

[0118] Furthermore, the plurality of intelligent agents includes a master intelligent agent, an executive intelligent agent, an evaluation intelligent agent, and a memory intelligent agent; Accordingly, the third task determination unit may specifically include: The first sequence determination subunit is used to determine the sequence of subtasks to be executed at the current time step by the main intelligent agent based on the business requirement description text and the interaction interface status information of the current time step for the first time step. The second sequence determination subunit is used to determine the sequence of subtasks to be executed in the current time step for each time step except the first time step, through the main intelligent agent based on the business requirement description text, the interactive interface status information of the current time step, and the sequence of subtasks to be executed, the sequence of completed subtasks, the execution action information, the evaluation feedback and the accumulated memory of the previous time step. The subtask determination subunit is used to determine the subtask to be executed corresponding to the current time step by means of the execution agent, based on the business requirement description text, the sequence of subtasks to be executed at the current time step, the interactive interface status information, and the evaluation feedback and accumulated memory of the previous time step.

[0119] Furthermore, the sequence of completed subtasks at each time step is determined by the main agent by identifying the subtasks that have been successfully executed in the sequence of subtasks to be executed at that time step; The execution action information for each time step is determined by the execution agent based on the sub-task to be executed. The execution action information includes the sub-task to be executed, the target action, and the expected effect of the action. The evaluation feedback for each time step is determined by the evaluation agent based on the business requirement description text, the interactive interface state information before the target action is executed, the interactive interface state information after the target action is executed, and the execution action information. The accumulated memory at each time step is determined by the memory agent based on the interactive interface state information of the previous time step, the interactive interface state information of the current time step, and the accumulated memory of the previous time step.

[0120] Furthermore, the execution information determination module 23 can specifically be used for: When the business operation task is an atomic operation or a subtask to be executed, the target attribute tag matching the business operation task is determined by keyword recognition or process template matching based on the pre-built task attribute library and the metadata. The task attribute library includes a mapping table between operation task types and attribute tags. The attribute tags include local adaptation attributes, cloud adaptation attributes, high real-time attributes, and low real-time attributes. If the target attribute tag includes a local adaptation attribute, the execution end of the business operation task is determined to be a local terminal; If the target attribute tag includes a cloud adaptation attribute, the execution end of the business operation task is determined based on the current environment perception data; When the target attribute label includes a high real-time attribute and / or it is determined that the business operation task has a preceding process chain dependency, the execution sequence of the business operation task is determined to be synchronous execution; When the target attribute label includes a low real-time attribute and it is determined that the business operation task has no preceding process chain dependency, the execution sequence of the business operation task is determined to be asynchronous.

[0121] Furthermore, the execution information determination module 23 can also be used for: When the business operation task is a standard step flow, based on the pre-built task attribute library and the metadata, the target attribute tag sequence that matches the standard step flow is determined by keyword recognition or process template matching, and one target attribute tag corresponds to one standard step. For each target attribute tag in the target attribute tag sequence, the execution end and execution sequence of the corresponding standard step are determined based on the target attribute tag.

[0122] Furthermore, the task execution module 24 can specifically be used for: Based on the execution terminal in the execution information, the business operation task is assigned to the corresponding terminal; Based on the execution sequence in the execution information, the terminal determines the task scheduling strategy corresponding to the business operation task, and executes the business operation task according to the task scheduling strategy. If the business operation task is a sub-task to be executed, return to re-execute the relevant steps for determining the business operation task corresponding to the business requirement description text, until the business process termination condition is met.

[0123] Furthermore, the device also includes an update module, which can be specifically used for: Based on the preset extraction frequency, obtain log data during the business process. The atomic operation library is updated based on the preset first statistical indicator, first triggering condition and first update logic, combined with the log data. The first statistical indicator includes the frequency of misses, atomic operation reuse rate, complexity of decomposing compound operations and atomic operation execution success rate. Based on the preset second statistical indicators, second triggering conditions, and second update logic, the standardized process knowledge graph is updated in conjunction with the log data. The second statistical indicators include task repetition frequency, multi-agent planning complexity, and workflow stability.

[0124] The business processing apparatus provided in the embodiments of the present invention can execute the business processing method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the method.

[0125] Figure 3 A schematic diagram of an electronic device 30 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0126] like Figure 3As shown, the electronic device 30 includes at least one processor 31 and a memory, such as a read-only memory (ROM) 32 or a random access memory (RAM) 33, communicatively connected to the at least one processor 31. The memory stores computer programs executable by the at least one processor. The processor 31 can perform various appropriate actions and processes based on the computer program stored in the ROM 32 or loaded from storage unit 38 into the RAM 33. The RAM 33 can also store various programs and data required for the operation of the electronic device 30. The processor 31, ROM 32, and RAM 33 are interconnected via a bus 34. An input / output (I / O) interface 35 is also connected to the bus 34.

[0127] Multiple components in electronic device 30 are connected to I / O interface 35, including: input unit 36, such as keyboard, mouse, etc.; output unit 37, such as various types of monitors, speakers, etc.; storage unit 38, such as disk, optical disk, etc.; and communication unit 39, such as network card, modem, wireless transceiver, etc. Communication unit 39 allows electronic device 30 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0128] Processor 31 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 31 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 31 performs the various methods and processes described above, such as business processing methods.

[0129] In some embodiments, the business processing method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 38. In some embodiments, part or all of the computer program may be loaded and / or mounted on electronic device 30 via ROM 32 and / or communication unit 39. When the computer program is loaded into RAM 33 and executed by processor 31, one or more steps of the business processing method described above may be performed. Alternatively, in other embodiments, processor 31 may be configured to execute the business processing method by any other suitable means (e.g., by means of firmware).

[0130] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0131] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0132] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0133] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0134] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0135] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0136] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0137] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A business processing method, characterized in that, include: Obtain the business requirement description text; Determine the business operation task corresponding to the business requirement description text, wherein the business operation task includes atomic operations, standard step flow, or subtasks to be executed. Obtain the metadata of the business operation task, and determine the execution information corresponding to the business operation task based on the metadata. The execution information includes the execution terminal and the execution sequence. Based on the execution information, complete the business operation task.

2. The method according to claim 1, characterized in that, The step of determining the business operation task corresponding to the business requirement description text includes: The business requirement description text is matched based on a pre-defined regular expression to determine whether the matching result meets the preset first matching condition. If satisfied, the word vectors of the business requirement description text are determined, and the operation hit result is determined based on the word vectors and the pre-built atomic operation library. If the operation hit result is a hit atomic operation, the business operation task is determined based on the hit atomic operation; If the conditions are not met or the result of the operation is that no atomic operation is hit, then by querying the pre-built standardized process knowledge graph, it is determined whether there is a standard step flow that matches the business requirement description text, and if there is, the standard step flow is identified as the business operation task. In the absence of the standard step flow, multiple agents determine the sub-tasks to be executed by time steps, and the sub-tasks to be executed are identified as business operation tasks.

3. The method according to claim 2, characterized in that, The steps for constructing the standardized process knowledge graph include: Based on preset pattern information, a set of triples is extracted from the text corresponding to at least one standardized process. The pattern information includes entity type constraint information, entity relationship constraint information, and entity key attribute constraint information. For each entity, a first embedding vector of the entity and a second embedding vector of the relation corresponding to the entity are determined by a preset embedding algorithm, the triples associated with the entity, and a trained graph convolutional network; wherein, the graph convolutional network is determined by iterative training using a positive sample set composed of collected real triples and a negative sample set composed of fake triples constructed according to a preset construction strategy. For each triple in the set of triples, the evaluation function value of the triple is determined based on the first embedding vector of the head entity, the first embedding vector of the tail entity, and the second embedding vector of the relation, combined with a preset evaluation function. Based on the target triples whose evaluation function values ​​satisfy the quality assessment conditions, a standardized process knowledge graph is constructed.

4. The method according to claim 2, characterized in that, The plurality of intelligent agents includes a master intelligent agent, an executive intelligent agent, an evaluation intelligent agent, and a memory intelligent agent; Accordingly, the step of determining the sub-task to be executed by multiple intelligent agents according to time steps includes: For the first time step, the main intelligent agent determines the sequence of subtasks to be executed for the current time step based on the business requirement description text and the interaction interface status information of the current time step. For each time step except the first time step, the main intelligent agent determines the sequence of subtasks to be executed in the current time step based on the business requirement description text, the interactive interface status information of the current time step, and the sequence of subtasks to be executed, the sequence of completed subtasks, the execution action information, the evaluation feedback, and the accumulated memory of the previous time step. The execution agent determines the sub-tasks to be executed at the current time step based on the business requirement description text, the sequence of sub-tasks to be executed at the current time step, the status information of the interactive interface, and the evaluation feedback and accumulated memory of the previous time step.

5. The method according to claim 4, characterized in that, The sequence of completed subtasks at each time step is determined by the main agent by identifying the subtasks that have been successfully executed in the sequence of subtasks to be executed at that time step; The execution action information for each time step is determined by the execution agent based on the sub-task to be executed. The execution action information includes the sub-task to be executed, the target action, and the expected effect of the action. The evaluation feedback for each time step is determined by the evaluation agent based on the business requirement description text, the interactive interface state information before the target action is executed, the interactive interface state information after the target action is executed, and the execution action information. The accumulated memory at each time step is determined by the memory agent based on the interactive interface state information of the previous time step, the interactive interface state information of the current time step, and the accumulated memory of the previous time step.

6. The method according to claim 1, characterized in that, When the business operation task is an atomic operation or a subtask to be executed, determining the execution information corresponding to the business operation task based on the metadata includes: Based on the pre-built task attribute library and the metadata, the target attribute tags that match the business operation task are determined by keyword recognition or process template matching. The task attribute library includes a mapping table between operation task types and attribute tags. The attribute tags include local adaptation attributes, cloud adaptation attributes, high real-time attributes, and low real-time attributes. If the target attribute tag includes a local adaptation attribute, the execution end of the business operation task is determined to be a local terminal; If the target attribute tag includes a cloud adaptation attribute, the execution end of the business operation task is determined based on the current environment perception data; When the target attribute label includes a high real-time attribute and / or it is determined that the business operation task has a preceding process chain dependency, the execution sequence of the business operation task is determined to be synchronous execution; When the target attribute label includes a low real-time attribute and it is determined that the business operation task has no preceding process chain dependency, the execution sequence of the business operation task is determined to be asynchronous.

7. The method according to claim 1, characterized in that, When the business operation task is a standard step flow, determining the execution information corresponding to the business operation task based on the metadata includes: Based on the pre-built task attribute library and the metadata, the target attribute tag sequence that matches the standard step flow is determined by keyword recognition or process template matching, with one target attribute tag corresponding to one standard step; For each target attribute tag in the target attribute tag sequence, the execution end and execution sequence of the corresponding standard step are determined based on the target attribute tag.

8. The method according to claim 1, characterized in that, The step of completing the business operation task based on the execution information includes: Based on the execution terminal in the execution information, the business operation task is assigned to the corresponding terminal; Based on the execution sequence in the execution information, the terminal determines the task scheduling strategy corresponding to the business operation task, and executes the business operation task according to the task scheduling strategy. If the business operation task is a sub-task to be executed, return to re-execute the relevant steps for determining the business operation task corresponding to the business requirement description text, until the business process termination condition is met.

9. The method according to claim 1, characterized in that, Also includes: Based on the preset extraction frequency, obtain log data during the business process. The atomic operation library is updated based on the preset first statistical indicator, first triggering condition and first update logic, combined with the log data. The first statistical indicator includes the frequency of misses, atomic operation reuse rate, complexity of decomposing compound operations and atomic operation execution success rate. The standardized process knowledge graph is updated based on the preset second statistical indicators, second triggering conditions, and second update logic, combined with the log data. The second statistical indicators include task repetition frequency, multi-agent planning complexity, and workflow stability.

10. A business processing apparatus, characterized in that, include: The acquisition module is used to acquire the business requirement description text; The task determination module is used to determine the business operation task corresponding to the business requirement description text. The business operation task includes atomic operations, standard step flow, or subtasks to be executed. An execution information determination module is used to obtain the metadata of the business operation task and determine the execution information corresponding to the business operation task based on the metadata. The execution information includes the execution end and the execution sequence. The task execution module is used to complete the business operation task based on the execution information.

Images