Business process self-adaptive execution method and system based on dynamic dialogue state network

Abstract

The invention provides a service process adaptive execution method and system based on a dynamic dialogue state network, and relates to the technical field of service process distribution. The method comprises the following steps: firstly, acquiring current service data and process knowledge, and analyzing current round dialogue data to identify a dominant intention and an associated intention in the current dialogue data; then calculating a semantic distance and flow logic closeness between the dominant intention and the associated intention and generating a dynamic cognitive focus; comparing the dynamic cognitive focus with historical dialogue data to screen out historical dialogue fragments, and positioning a key decision sub-graph according to the historical dialogue fragments; and finally, optimizing the dynamic dialogue state network in real time according to the complexity of the key decision sub-graph and the emotional characteristics in the current dialogue data so as to adjust the execution logic in the business process knowledge graph and generate a business process execution scheme. According to the technical scheme provided by the invention, real-time, accurate and flexible adjustment of the business process in a complex interaction scene is realized.

Description

Technical Field

[0001] This application relates to the field of business process allocation technology, and in particular to a business process adaptive execution method and system based on a dynamic dialogue state network. Background Technology

[0002] As enterprises enter a more complex phase of digital transformation, business processes are becoming increasingly complex and dynamic. In particular, in scenarios involving multiple rounds of interaction and real-time decision-making, such as online customer service and intelligent approval, traditional fixed process models are no longer sufficient. These scenarios require processing systems to dynamically parse user intent and adjust predefined business process logic in real time during the dialogue.

[0003] To address this need, the current approach is a business process management system that combines large language models with intelligent agent technology. This system uses the semantic understanding capabilities of large models to parse user input and uses pre-defined intelligent agents to automate tasks or make approval decisions. For example, the system can directly generate or call predefined process templates based on natural language descriptions to achieve automatic process initiation and automated processing of key nodes.

[0004] However, existing solutions of this kind have significant drawbacks. Their decision-making logic heavily relies on static knowledge acquired through training, lacking the ability to accurately quantify and perform real-time computation of the deep semantic relationships and dynamic logical tightness between nodes in the business process knowledge graph. When handling multi-turn dialogues with frequent intent shifts, the system struggles to dynamically locate key decision paths and adjust the process based on the real-time dialogue state, resulting in insufficient adaptive adjustment capabilities and a tendency for process breakpoints or decision deviations. Summary of the Invention

[0005] This application provides a business process adaptive execution method and system based on a dynamic dialogue state network to solve the problems of weak dynamic context awareness and rigid business process adjustment in the prior art.

[0006] Firstly, this application provides a business process adaptive execution method based on a dynamic dialogue state network, including: Acquire current service data and process knowledge. The current service data includes current dialogue data and historical dialogue data. The process knowledge is a business process knowledge graph that includes process nodes, rules, and logical relationships. The current round of dialogue data is parsed using a pre-deployed dynamic dialogue state network to identify the dominant and associated intentions in the current dialogue data. Based on the business process knowledge graph, the semantic distance and process logic tightness between the dominant intent and the associated intent are calculated, and a dynamic cognitive focus is generated based on the semantic distance and the process logic tightness. The dynamic cognitive focus is compared with the historical dialogue data. Based on the comparison results, historical dialogue fragments are selected from the historical dialogue data. Based on the selected historical dialogue fragments, key decision subgraphs are located in the business process knowledge graph. Based on the complexity of the key decision subgraph and the emotional features in the dialogue data of the current round, the dynamic dialogue state network is optimized in real time, and the execution logic in the business process knowledge graph is adjusted using the optimized dynamic dialogue state network to generate a business process execution plan.

[0007] Optionally, the current round of dialogue data is parsed using a pre-deployed dynamic dialogue state network to identify the dominant and related intentions in the current dialogue data, including: Extract the user's statement in the current round and the previously proceeding dialogue statements from the current dialogue data; The user statement is parsed to obtain explicit query components; The user statement is subjected to a second parsing process to obtain implicit context components; The explicit query components, the implicit context components, and the previously executed dialogue statements are combined to form dialogue context information; Joint analysis is performed on the dialogue context information to calculate the association strength between the first candidate intent corresponding to the explicit query component and the second candidate intent corresponding to the implicit context component and the previously advanced dialogue statement; The first candidate intent and the second candidate intent are then integrated to generate a candidate intent set; In the set of candidate intentions, the candidate intentions whose association strength meets the first preset benchmark are determined as the dominant intentions; In the candidate intent set, intents whose semantic distance is within a preset range and whose association strength meets a second preset benchmark are determined as associated intents.

[0008] Optionally, based on the business process knowledge graph, the semantic distance and process logic tightness between the dominant intent and the associated intent are calculated, and a dynamic cognitive focus is generated based on the semantic distance and the process logic tightness, including: The dominant intent and the associated intent are respectively mapped to process nodes in the business process knowledge graph to obtain the dominant intent node and the associated intent node; Calculate the connection path between the dominant intent node and the associated intent node in the business process knowledge graph; The semantic distance between the dominant intent and the associated intent is determined based on the number of nodes traversed by the connection path. In the business process knowledge graph, identify the rule nodes that are commonly connected to the dominant intent node and the associated intent node, as well as the logical dependencies between the dominant intent node and the associated intent node; Based on the number of the commonly connected rule nodes and the type of the logical dependency relationship, the flow logic tightness between the dominant intent and the associated intent is calculated. By combining the semantic distance and the logical tightness of the process, the cognitive focus quantification value of all associated intentions is calculated; Compare all cognitive focus quantification values ​​and select the relevance intent with the largest cognitive focus quantification value as the focus intent; The focus intent, the semantic distance corresponding to the focus intent, and the tightness of the process logic are combined to obtain a dynamic cognitive focus.

[0009] Optionally, the dynamic cognitive focus is compared with the historical dialogue data, and historical dialogue segments are selected from the historical dialogue data based on the comparison results, including: Multiple historical dialogue fragments and corresponding historical focus information are obtained from the historical dialogue data. The historical focus information includes historical focus intent, historical semantic distance, and historical process logic tightness. From the dynamic cognitive focus, extract the current focus intent, current semantic distance, and current process logic tightness; Calculate the first similarity between the current focus intent and each of the historical focus intents, calculate the second similarity between the current semantic distance and each of the historical semantic distances, and calculate the third similarity between the current process logic tightness and each of the historical process logic tightness; For each historical dialogue segment, the first similarity, the second similarity, and the third similarity are merged to obtain the comprehensive matching degree between the historical dialogue segment and the dynamic cognitive focus; From the multiple historical dialogue segments, select the historical dialogue segments whose overall matching degree meets the preset threshold condition.

[0010] Optionally, based on the selected historical dialogue fragments, key decision subgraphs are located in the business process knowledge graph, including: Extract the processed historical intent from each of the selected historical dialogue segments; The historical intents are mapped onto the business process knowledge graph to obtain multiple corresponding historical intent nodes; In the business process knowledge graph, the path connecting the multiple historical intent nodes is determined, and an initial graph structure containing the multiple historical intent nodes is determined based on the path; Extract the current focus intent from the dynamic cognitive focus, and map the current focus intent to the business process knowledge graph to obtain the current focus intent node; In the business process knowledge graph, the connection relationship between the current focus intent node and each historical intent node in the initial graph structure is calculated; Based on the connection relationship, a graph range is defined in the business process knowledge graph, and the process nodes, rules and logical relationships involved in the graph range are determined as key decision subgraphs. The graph range includes the current focus intent node and at least one of the historical intent nodes.

[0011] Optionally, the dynamic dialogue state network is optimized in real time based on the complexity of the key decision subgraph and the emotional features in the dialogue data of the current round, including: The total number of process nodes and rules contained in the key decision subgraph is counted to obtain the first quantitative value; The second quantification value is obtained by analyzing the types and number of logical dependencies between each process node in the key decision sub-graph. The first quantized value and the second quantized value are combined to calculate the complexity quantification result of the key decision subgraph; Perform sentiment analysis on the current dialogue data to extract emotional state information and emotional fluctuation intensity information from the current dialogue data; Based on the complex quantification results, first optimization parameters are generated to perform first structural optimization on the dynamic dialogue state network. Based on the emotional state information and the emotional fluctuation intensity information, second optimization parameters are generated to perform a second structural optimization on the dynamic dialogue state network. The first optimization parameter and the second optimization parameter are applied to the dynamic dialogue state network to adjust the internal resource allocation weights of the dynamic dialogue state network, thereby completing the real-time optimization of the dynamic dialogue state network.

[0012] Optionally, the execution logic in the business process knowledge graph is adjusted using an optimized dynamic dialogue state network to generate a business process execution plan, including: Based on the optimized dynamic dialogue state network, the key decision subgraph is analyzed to determine the execution priority of each process node in the key decision subgraph. Based on the execution priority, adjust the logical connection relationship between the process nodes in the key decision sub-graph to form a preliminary logical sequence; The preliminary logical sequence is input into the optimized dynamic dialogue state network, and the connectivity and consistency of the preliminary logical sequence in the overall business process knowledge graph are evaluated through the optimized dynamic dialogue state network. Based on the evaluation results, the connection relationships between process nodes and the order of rule application in the preliminary logic sequence are corrected through the optimized dynamic dialogue state network to generate an optimized logic sequence. Based on the optimized logical sequence, corresponding process nodes, rules, and logical relationships are extracted from the business process knowledge graph and combined to generate a business process execution plan.

[0013] Secondly, this application provides a business process adaptive execution system based on a dynamic dialogue state network, comprising: The acquisition module is used to acquire current service data and process knowledge. The current service data includes current dialogue data and historical dialogue data, and the process knowledge is a business process knowledge graph that includes process nodes, rules and logical relationships. The identification module is used to parse the current round of dialogue data through a pre-deployed dynamic dialogue state network to identify the dominant intent and related intents in the current dialogue data; The calculation module is used to calculate the semantic distance and process logic tightness between the dominant intent and the associated intent based on the business process knowledge graph, and generate a dynamic cognitive focus based on the semantic distance and the process logic tightness. The positioning module is used to compare the dynamic cognitive focus with the historical dialogue data, filter historical dialogue fragments from the historical dialogue data according to the comparison results, and locate the key decision subgraph in the business process knowledge graph according to the filtered historical dialogue fragments. The optimization module is used to optimize the dynamic dialogue state network in real time based on the complexity of the key decision subgraph and the emotional features in the dialogue data of the current round, and to adjust the execution logic in the business process knowledge graph using the optimized dynamic dialogue state network to generate a business process execution plan.

[0014] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement a business process adaptive execution method based on a dynamic dialogue state network as described in the first aspect above.

[0015] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a business process adaptive execution method based on a dynamic dialogue state network as described in the first aspect.

[0016] This application first generates dynamic cognitive focus by quantifying the semantic and logical relationship between dominant intent and associated intent. Combined with historical dialogue pattern matching, it can accurately lock the core demand and context of the current dialogue even when the user intent jumps between multiple business processes. This enhances the dynamic perception and understanding of complex multi-turn dialogues and effectively solves the technical problems of weak dynamic context perception and easy process breakpoints in the existing technology.

[0017] Furthermore, based on the complexity of the key decision subgraphs identified in real time and the user's emotional characteristics, the dynamic dialogue state network is optimized. The optimized network is then used to dynamically adjust and optimize the execution logic of the business process knowledge graph. This enables the final business process execution plan to closely match the user's real-time cognitive focus and emotional state, achieving real-time and flexible self-adaptation of the business process. This fundamentally overcomes the shortcomings of rigid process adjustment and insufficient self-adaptation capabilities in existing technical solutions, and improves the accuracy and smoothness of business processing.

[0018] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of a business process adaptive execution method based on a dynamic dialogue state network provided in this application is shown; Figure 2 This paper presents a schematic diagram of the structure of a business process adaptive execution system based on a dynamic dialogue state network provided in this application. Figure 3 A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present application, the technical solution of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0022] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] Figure 1 This application provides a flowchart of a business process adaptive execution method based on a dynamic dialogue state network, such as... Figure 1 As shown, the method includes: Step 101: Obtain current service data and process knowledge. The current service data includes current dialogue data and historical dialogue data. The process knowledge is a business process knowledge graph that includes process nodes, rules, and logical relationships.

[0025] In this step, the current service data refers to the set of information related to the current user interaction collected from the ongoing online customer service dialogue. It is used to provide contextual basis for subsequent intent understanding and decision-making, and is obtained by listening to and recording the input and output streams of the dialogue interface in real time.

[0026] Process knowledge refers to a knowledge base defined in a structured form that describes the standard operating procedures and rules for specific business areas such as bank account opening and product after-sales service. It is used to provide a logical framework for dialogue and decision-making and is obtained through business expert analysis and knowledge engineer modeling.

[0027] Current dialogue data specifically refers to the dialogue record consisting of the latest round of user questions and system responses in the current service data. It is used to analyze the user's current needs in real time and is obtained by capturing the latest complete interaction round in the dialogue flow.

[0028] Historical dialogue data specifically refers to past dialogue records that occurred before the current round in the current service data and are related to the same session or the same user. It is used to provide historical background and pattern reference and is obtained by querying and extracting from the dialogue log database.

[0029] A business process knowledge graph is a specific form of storage and representation of process knowledge. It organizes and stores business knowledge in a graphical structure. In this graph, each step in the business process is represented as a process node, the execution conditions between steps are represented as rules, and the connection between steps is represented as logical relationships. It is used to transform process knowledge in text or rule form into a network structure that can be queried and reasoned by computers. It is constructed by taking original materials such as business process diagrams and rule documents through steps such as entity extraction and relationship definition.

[0030] In this step, a data listening and collection process is first started. By calling the application programming interface provided by the customer service platform, it continuously receives and records each round of user statements and system responses from the dialogue front end. A data collection thread is maintained for each independent service session. This thread not only captures the latest sentence sent by the user, which is the current dialogue data, but also automatically initiates a structured query to the platform's backend database based on the unique identifier of the session, such as a session ID. The purpose is to obtain all dialogue records generated before the current round of the session. After these historical records are extracted, they form historical dialogue data. The current dialogue data and historical dialogue data together constitute a complete set of information reflecting the current service status, namely the current service data. Secondly, a graph loading operation is performed in the parallel knowledge loading phase. Specifically, based on the business domain of this dialogue, such as credit card business or loan business, a data request is sent to an independent knowledge graph server. After receiving the request, the knowledge graph server retrieves the corresponding pre-built business process knowledge graph from its storage and transmits the complete graph data, including all process nodes, rules, and logical relationships, to the platform's backend via the network. The backend receives and parses this data in memory, reconstructing it into a graph structure object that can be directly accessed and manipulated by the program. Finally, through these two parallel operations, the current service data and the business process knowledge graph are prepared simultaneously, providing a solid information foundation and a rigorous reasoning framework for subsequent in-depth analysis and decision-making.

[0031] For example, in Bank A's online intelligent customer service system, user C is inquiring about credit card related services. First, when the process begins, a listener is started to capture the sentence user C just sent: "I want to check the balance of this card." This sentence is recorded as the current dialogue data. Second, based on the session ID of this dialogue, a query command is sent to the bank's dialogue log database. This successfully retrieves user C's statement a few minutes earlier that "my credit card seems to be lost" and the corresponding response from customer service. These earlier dialogue records are extracted as historical dialogue data. Both of these constitute the current service data. At the same time, a request is sent to the knowledge base server to load the knowledge graph of the bank card business domain. Then, the server responds to the request, sending a business process knowledge graph data package containing multiple process nodes such as loss reporting, balance inquiry, and transaction appeal, as well as the complex rules between them, to the backend, and loading it into memory.

[0032] This step, by systematically integrating real-time interactive data and structured business knowledge, lays a solid data and knowledge foundation for the entire adaptive processing method. It ensures that all subsequent intelligent analysis actions can be performed within a complete dialogue context and an accurate business logic framework, thereby avoiding decision-making biases caused by missing information or partial knowledge. This is the primary prerequisite for achieving accurate, flexible, and adaptive business processes.

[0033] Step 102: The current round of dialogue data is parsed through a pre-deployed dynamic dialogue state network to identify the dominant intent and related intents in the current dialogue data.

[0034] Optionally, step 102 may specifically include: Step 1021: Extract the user statement of the current round and the previously advanced dialogue statements from the current dialogue data.

[0035] Step 1022: Perform a first parsing process on the user statement to obtain explicit query components.

[0036] Step 1023: Perform a second parsing process on the user statement to obtain implicit context components.

[0037] Step 1024: Combine the explicit query component, the implicit context component, and the previously executed dialogue statement to form dialogue context information.

[0038] Step 1025: Perform joint analysis on the dialogue context information to calculate the association strength between the first candidate intent corresponding to the explicit query component and the second candidate intent corresponding to the implicit context component and the previously proceeding dialogue statement.

[0039] Step 1026: Integrate the first candidate intent with the second candidate intent to generate a candidate intent set.

[0040] Step 1027: In the candidate intent set, the candidate intent whose association strength meets the first preset benchmark is determined as the dominant intent.

[0041] Step 1028: In the candidate intent set, intents whose semantic distance is within a preset range and whose association strength meets the second preset benchmark are determined as associated intents.

[0042] In this step, the dynamic dialogue state network refers to a pre-trained neural network model that can understand and track changes in the dialogue process. It is used to parse user statements in real time and update the understanding of the core content of the dialogue. The network model is trained using a large amount of labeled dialogue history data.

[0043] The dominant intent refers to the most core and most likely request or goal extracted from the user's current discourse, representing the main business action that the user currently wants to complete.

[0044] Related intents refer to other possible requests or goals that are semantically close to the dominant intent, logically related to it, and mentioned or implied in the current conversation context. They are used to provide a more comprehensive understanding of user needs and decision-making alternatives.

[0045] The user's statement in the current round refers to the text content actively entered or spoken by the user in the current round of the conversation, which serves as the most direct basis for analyzing the user's immediate intent.

[0046] Previous dialogue statements refer to the dialogue content in the most recent rounds before the current round in which the two parties exchanged substantial information. They are used to provide the recent background necessary to understand the current statement and are obtained by extracting the dialogue from the previous few rounds immediately preceding the current round from historical dialogue data.

[0047] The first parsing process refers to a text processing technique that focuses on analyzing the surface grammar and keywords of sentences. It is used to extract explicit request components directly from the literal meaning of user statements, and obtains the results through dependency parsing and keyword extraction techniques in natural language processing.

[0048] Explicit query components refer to combinations of words or phrases that are directly extracted from user statements and explicitly express a specific request or question, serving as the basis for identifying the user's explicit intent.

[0049] The second parsing process refers to a text processing technique that focuses on analyzing the deep semantics, sentiment, and referential meaning of sentences. It is used to infer unstated but relevant background or potential demands from user statements, and obtains these through semantic role labeling and referential resolution techniques in natural language processing.

[0050] Implicit contextual components refer to background information, emotional tendencies, or potential goals that are inferred from user statements but are not directly stated but are closely related to the current dialogue context, and are used to supplement the complete dialogue scenario.

[0051] Dialogue context information refers to a structured information package that integrates the user's explicit requests, unspoken background information, and recent dialogue history to provide a comprehensive and coherent basis for decision-making in intent recognition.

[0052] The first candidate intent refers to a set of possible intent options matched or inferred from the business intent list based on the explicitly stated request components in the user's statement. It is used to cover the possibilities directly expressed by the user and is obtained by matching explicit query components with the business intent dictionary.

[0053] The second candidate intent refers to a set of possible intent options inferred from the list of business intents by combining unstated background information and recent conversation history, used to uncover potential or ongoing user needs.

[0054] A candidate intent set refers to a list containing all possible intent options, integrating various explicit and implicit requests from the user, and is used as a candidate pool to filter the final intent.

[0055] The first preset benchmark refers to a pre-set numerical threshold used to measure the strength of intent association, which is used to select the intent with the highest association strength as the dominant intent. This threshold is determined through model training.

[0056] The second preset benchmark refers to a pre-defined composite condition used to simultaneously measure the semantic closeness and relevance between intents. It is used to filter out intents that are both related to and different from the dominant intent as associated intents, and is obtained by determining the semantic distance range and relevance threshold through business experience.

[0057] In this step, the user statement that was just entered is first separated from the prepared current dialogue data and used as the user statement for the current round. At the same time, dialogue records with substantial content that are adjacent to the current round, usually the previous two or three rounds, are selected from the historical dialogue data and used as the previous progressing dialogue statements. Then, the natural language processing module is called to perform the first parsing process on the user statement of the current round. Dependency parsing technology is used to find the core verbs in the sentence and their directly related noun components. These combinations of explicit actions and objects are extracted to form explicit query components, such as extracting "query" and "bill" from "help me check my bill". In parallel, the same user statement is parsed a second time. Semantic role labeling and referential resolution technology are used to identify the unspoken but implied roles, emotions or referents in the sentence. For example, from "that money was deducted incorrectly", it is inferred that there is a potential appeal or question, and that the money specifically refers to a transaction mentioned in the historical dialogue. These inference results are organized into implicit context components. Next, the obtained explicit query components and implicit context components, along with the previously extracted previous dialogue statements, are packaged into a structured information unit, namely, dialogue context information. Simultaneously, the core processing unit of the dynamic dialogue state network begins operation, receiving this dialogue context information and performing joint analysis. First, the explicit query components are mapped to a set of possible first-candidate intentions, such as querying a bill or checking a balance. Then, the network analyzes the implicit context components and previous dialogue statements to infer a set of possible second-candidate intentions, such as inferring a complaint transaction based on historical transaction disputes. Then, the network evaluates the correlation strength between the first-candidate intentions and the second-candidate intentions through its complex internal weight calculations. This strength value reflects the rationality and closeness of these intentions appearing simultaneously in the current context. Finally, all calculated first-candidate intentions and second-candidate intentions are merged, and duplicates are removed to form a complete set of candidate intentions. Finally, according to preset rules, the dynamic dialogue state network traverses the candidate intent set to find the intent with the highest correlation strength calculation value among all intents. This highest value is then compared with the first preset benchmark. If it is higher than the first preset benchmark, the highest intent is determined as the dominant intent of this round of dialogue. Subsequently, further filtering is performed on the remaining candidate intents. The semantic distance between each remaining intent and the just determined dominant intent is calculated. This semantic distance reflects the similarity between the two intents in terms of business concepts. Intents whose semantic distance is within a preset reasonable range and whose own correlation strength calculation value also exceeds the second preset benchmark are selected and determined as related intents. For example, if the dominant intent is to report a lost credit card, its related intents may be to freeze an account and query recent transactions.

[0058] For example, following the specific implementation of the previous step, after user C said that his credit card seemed to be lost and that he wanted to check the balance of his card, he entered the sentence "Can I freeze this card now?" in the current round. Then, the user statement "Can I freeze this card now?" is extracted as the current round's statement, and the previous round's dialogue "I want to check the balance of my card" and the system's response are extracted as the previous proceeding dialogue statement. Then, through the first parsing process, the explicit query components "freeze" and "card" are extracted from the current statement. Simultaneously, through a second parsing process, combined with lost context from the past, it is inferred that implicit contextual components include an urgent need to prevent fraudulent transactions. These are then combined with previously generated dialogue statements to form complete dialogue context information. Subsequently, after dynamic dialogue state network analysis, a set of candidate intentions is generated, including freezing cards, reporting lost cards, and checking balances. It is then calculated that freezing cards has the highest relevance in the current context and exceeds the first preset benchmark, so it is identified as the dominant intention. Reporting lost cards is found to be semantically close to freezing cards, and its relevance also exceeds the second preset benchmark, so it is identified as a related intention. Finally, although checking balances has a reasonable relevance, its semantic distance from freezing cards is relatively far, so it may not be selected as a related intention.

[0059] This step, through progressive analysis of the dynamic dialogue state network, not only accurately captures the user's current main core needs, but also intelligently associates other possible needs that are closely related in business logic; it achieves a three-dimensional understanding of the intent of complex and disjointed dialogues, providing key and reliable input for subsequent precise decision-making focus positioning and dynamic process adjustment based on business process knowledge graphs.

[0060] Step 103: Based on the business process knowledge graph, calculate the semantic distance and process logic tightness between the dominant intent and the associated intent, and generate a dynamic cognitive focus based on the semantic distance and the process logic tightness.

[0061] Optionally, step 103 may specifically include: Step 1031: Map the dominant intent and the associated intent to process nodes in the business process knowledge graph, respectively, to obtain the dominant intent node and the associated intent node.

[0062] Step 1032: Calculate the connection path between the dominant intent node and the associated intent node in the business process knowledge graph.

[0063] Step 1033: Determine the semantic distance between the dominant intent and the associated intent based on the number of nodes traversed by the connection path.

[0064] Step 1034: In the business process knowledge graph, identify the rule nodes that are commonly connected to the dominant intent node and the associated intent node, as well as the logical dependency relationship between the dominant intent node and the associated intent node.

[0065] Step 1035: Calculate the flow logic tightness between the dominant intent and the associated intent based on the number of commonly connected rule nodes and the type of logical dependency relationship.

[0066] Step 1036: Combine the semantic distance and the process logic tightness to calculate the cognitive focus quantification value of all associated intentions.

[0067] Step 1037: Compare all cognitive focus quantification values ​​and select the related intent with the largest cognitive focus quantification value as the focus intent.

[0068] Step 1038: Combine the focus intent, the semantic distance corresponding to the focus intent, and the tightness of the process logic to obtain a dynamic cognitive focus.

[0069] In this step, semantic distance refers to the distance between the nodes corresponding to two intentions in the business process knowledge graph, and is used to measure the relevance of the two intentions at the business concept level.

[0070] Process logic tightness refers to the degree of coupling at the business execution level between the nodes corresponding to two intentions in the business process knowledge graph, which is reflected by shared rules and direct logical connections. It is used to measure the strength of the association between two intentions in the business operation process.

[0071] Dynamic cognitive focus refers to a structured information package that includes the relevant intents (focus intents) that are most relevant to the dominant intent and most likely to become the next point of attention in the current dialogue context, as well as the semantic distance and logical tightness between them, in order to accurately characterize the business demands that need to be focused on at the moment.

[0072] A process node refers to a specific, indivisible business operation step in a business process knowledge graph. It is used to construct a business process graph by abstracting each step in the business manual into a node when constructing the knowledge graph.

[0073] The dominant intent node refers to the process node in the business process knowledge graph that matches the dominant intent identified from the dialogue, and is used to anchor the user's linguistic intent to specific business operation steps.

[0074] A related intent node refers to the process node in the business process knowledge graph that matches a related intent identified from the dialogue. It is used to anchor the user's potential related intents to specific business operation steps.

[0075] A connection path refers to a set of continuous edges and intermediate nodes that connect two specific process nodes in a business process knowledge graph. It describes the possible route from one business step to another and is obtained by running a graph search algorithm on the graph to find the shortest path between the two nodes.

[0076] A rule node is a node in a business process knowledge graph that represents a business judgment condition or execution rule. It is used to connect different process nodes and control the flow of the process. It is obtained by abstracting business rules into nodes and connecting them with relevant process nodes when constructing the knowledge graph.

[0077] Cognitive Focus Quantification (CFF) is a comprehensive numerical score that quantifies the overall semantic and procedural relevance between a particular related intent and the current dominant intent. A higher CFF represents a closer relevance and a greater likelihood of it becoming the focus. It is used to objectively compare the priorities of different related intents. It is calculated by substituting the semantic distance value and procedural logic tightness value corresponding to a related intent into a preset mathematical formula.

[0078] The focus intent refers to the intent with the highest cognitive focus quantification value selected from all related intents. It represents the secondary or related business needs that most need attention and processing in the current dialogue state and is used to form the core of dynamic cognitive focus.

[0079] In this step, the obtained dominant intent and each associated intent are first used as query keywords and matched with the labels of all process nodes in the business process knowledge graph. When a node that is a perfect match or highly similar is found, the dominant intent and each associated intent are mapped to that process node, thus obtaining a dominant intent node and one or more associated intent nodes. Next, starting from the dominant intent node and ending at each associated intent node, the shortest path search algorithm and breadth-first search algorithm in graph theory are run on the network of the business process knowledge graph to find the shortest connection path connecting them. Then, the number of intermediate process nodes traversed by each connection path is counted, and this number is directly used as the semantic distance between the dominant intent and the associated intent. The longer the path and the more nodes, the farther apart the two intents are in terms of business concepts, and the greater the semantic distance. Then, for each pair of dominant intent nodes and related intent nodes, the business process knowledge graph is explored to find all rule nodes that are connected to both nodes simultaneously, and the number of these jointly connected rule nodes is counted. Simultaneously, the type of edges connecting these two nodes in the business process knowledge graph is analyzed to determine whether they are sequential dependencies, conditional branches, or parallel logical dependencies. Based on a pre-defined formula, the number of jointly connected rule nodes and the type weight of the logical dependency relationship are combined to calculate a process logic tightness value; the more rule nodes and the more direct the logical relationship, the higher the tightness value. Subsequently, the calculated semantic distance and process logic tightness are substituted into another pre-defined fusion formula to calculate a cognitive focus quantification value for each related intent. This formula is designed so that related intents with smaller semantic distances and greater process logic tightness have higher cognitive focus quantification values. Finally, the cognitive focus quantification values ​​of all related intents are compared, and the one with the largest value is selected as the focus intent. This focus intent, along with its calculated semantic distance and process logic tightness with the dominant intent, is packaged into a complete information structure, namely, the dynamic cognitive focus.

[0080] For example, following the specific implementation of the previous step, firstly, in the business process knowledge graph, the dominant intent to freeze the card is mapped to the application card freeze node, and the associated intent to report the loss of the card is mapped to the application card loss node. Secondly, the shortest path between these two nodes is searched, and it is found that an intermediate node is required to verify the customer's identity, thus determining the semantic distance to be 1. Next, it is identified that these two nodes jointly connect two rule nodes: identity verified and card status abnormal, and the logical relationship between them suggesting subsequent operations is analyzed, thereby calculating a high process logic tightness. Then, combining the small semantic distance and the high tightness, the cognitive focus quantification value of reporting the loss of the card is calculated to be the highest, so it is selected as the focus intent. Finally, a dynamic cognitive focus is generated, which includes the focus intent to report the loss of the card and its semantic distance and process logic tightness value with the dominant intent.

[0081] This step maps dialogue intents to a business process knowledge graph and performs quantitative analysis, accurately measuring the conceptual and logical relationships between intents and intelligently filtering out the most relevant focus intents. This achieves a transformation from discrete intents to focused business issues, providing a core basis for subsequently identifying key decision areas and driving adaptive adjustments to the process.

[0082] Step 104: Compare the dynamic cognitive focus with the historical dialogue data, filter historical dialogue fragments from the historical dialogue data based on the comparison results, and locate key decision subgraphs in the business process knowledge graph based on the filtered historical dialogue fragments.

[0083] Optionally, step 104 may specifically include: Step 1041: Obtain multiple historical dialogue fragments and corresponding historical focus information from the historical dialogue data. The historical focus information includes historical focus intent, historical semantic distance, and historical process logic tightness.

[0084] Step 1042: Extract the current focus intent, current semantic distance, and current process logic tightness from the dynamic cognitive focus.

[0085] Step 1043: Calculate the first similarity between the current focus intent and each of the historical focus intents, calculate the second similarity between the current semantic distance and each of the historical semantic distances, and calculate the third similarity between the current process logic tightness and each of the historical process logic tightnesses.

[0086] Step 1044: For each of the historical dialogue segments, the first similarity, the second similarity, and the third similarity are merged to obtain the comprehensive matching degree between the historical dialogue segment and the dynamic cognitive focus.

[0087] Step 1045: Select historical dialogue segments from the plurality of historical dialogue segments whose overall matching degree meets the preset threshold condition.

[0088] Step 1046: Extract the processed historical intent from each of the selected historical dialogue segments.

[0089] Step 1047: Map the historical intent to the business process knowledge graph to obtain multiple corresponding historical intent nodes.

[0090] Step 1048: In the business process knowledge graph, determine the path connecting the multiple historical intent nodes, and determine the initial graph structure containing the multiple historical intent nodes based on the path.

[0091] Step 1049: Extract the current focus intent from the dynamic cognitive focus, and map the current focus intent to the business process knowledge graph to obtain the current focus intent node.

[0092] Step 10410: In the business process knowledge graph, calculate the connection relationship between the current focus intent node and each historical intent node in the initial graph structure.

[0093] Step 10411: Based on the connection relationship, define a graph range in the business process knowledge graph, and determine the process nodes, rules and logical relationships involved in the graph range as key decision subgraphs. The graph range includes the current focus intent node and at least one of the historical intent nodes.

[0094] In this step, the key decision subgraph refers to a local area delineated from the complete business process knowledge graph. It contains the process nodes, rules, and their logical relationships that are most directly related to the current and historical dialogue focus, and serves as the core operational scope for subsequent adaptive process adjustments and optimizations.

[0095] Historical focus information refers to structured information similar to dynamic cognitive focus recorded after systematic analysis of previously completed dialogue segments. It includes historical focus intent, historical semantic distance, and historical process logic tightness, and is used as a reference benchmark for comparison with the current scenario.

[0096] The current focus intent refers to the intent extracted from the generated dynamic cognitive focus that represents the most relevant secondary demand at present. It is used to match the historical focus intent and is obtained by reading it from the dynamic cognitive focus information structure.

[0097] Current semantic distance refers to the semantic distance value between the current focus intent and the dominant intent extracted from the dynamic cognitive focus, which is used to match with historical semantic distance.

[0098] The current process logic tightness refers to the process logic tightness value extracted from the dynamic cognitive focus, which is the value between the current focus intention and the dominant intention.

[0099] The first similarity refers to the quantitative value of the closeness of the current focus intent and a historical focus intent in terms of textual semantics. It is used to measure the similarity of two intents in terms of literal meaning and connotation, and is obtained by calculating the cosine similarity between the text vector representations of the two intents.

[0100] The second similarity refers to the numerical closeness between the current semantic distance and a historical semantic distance. It is used to measure the consistency between two pairs of intentions in terms of conceptual proximity and is obtained by calculating the reciprocal of the absolute difference between the two distance values.

[0101] The third similarity refers to the numerical closeness between the current process logic tightness and the historical process logic tightness. It is used to measure the consistency of the business logic coupling strength between two pairs of intentions and is obtained by calculating the reciprocal of the absolute difference between the two tightness values.

[0102] The overall matching score is a comprehensive score that integrates the first similarity score, the second similarity score, and the third similarity score. It is used to evaluate the degree of matching between a historical dialogue segment and the current dynamic cognitive focus.

[0103] Historical intent refers to the list of intents that were identified and processed in history from selected historical dialogue fragments, which serves as the basis for constructing a historical experience map.

[0104] Multiple historical intent nodes refer to the set of process nodes corresponding to each intent in the extracted historical intent list after mapping to the business process knowledge graph. This is used to locate the steps involved in historical experience in the knowledge graph by matching and mapping historical intents with graph node labels.

[0105] The initial graph structure refers to a local network segment in the business process knowledge graph, consisting of multiple historical intent nodes and the edges connecting these nodes. It is used to represent the business process areas touched by similar historical dialogues and is obtained by finding and including the shortest path connecting these historical intent nodes and the nodes on it in the knowledge graph.

[0106] The current focus intent node refers to the corresponding process node after mapping the current focus intent to the business process knowledge graph. It is used to anchor the current focus to a specific business step and is obtained by matching and mapping the current focus intent with the graph node label.

[0107] In this step, the historical dialogue database is first accessed to read multiple historical dialogue fragments and pre-stored historical focus information for each fragment, including historical focus intent, historical semantic distance, and historical process logic tightness. Simultaneously, the current focus intent, current semantic distance, and current process logic tightness are extracted from the generated dynamic cognitive focus. Next, a multi-dimensional similarity calculation process is initiated. For each historical dialogue fragment, three similarities are calculated: using a word vector model in natural language processing, the current focus intent and historical focus intent are converted into numerical vectors, and the cosine of the angle between these two vectors is calculated to obtain the first similarity. The closer this value is to 1, the more similar the two intents are. Then, the absolute value of the difference between the current semantic distance and the historical semantic distance is calculated, and then mapped to a second similarity between 0 and 1 through a transformation function. The smaller the difference, the higher the similarity. Similarly, the absolute value of the difference between the current flow logic tightness and the historical flow logic tightness is calculated and mapped to a third similarity. Then, a preset weight configuration is used to perform a weighted sum of the three similarities to calculate a comprehensive matching degree for each historical dialogue segment. The comprehensive matching degree of all historical segments is compared with a preset threshold, and historical dialogue segments with a comprehensive matching degree higher than the threshold are selected. Furthermore, from each selected historical dialogue segment, a list of historical intents processed at that time is read, and these historical intents are used as query conditions to search in the business process knowledge graph to find the corresponding process node for each intent, thus obtaining multiple historical intent nodes; then, based on these historical intent nodes, a graph traversal algorithm is run in the business process knowledge graph to find the minimum connected subgraph connecting all these nodes. This minimum connected subgraph contains these nodes and all intermediate nodes and edges on the shortest path between them, forming the initial graph structure; Simultaneously, the current focus intent is also mapped onto the business process knowledge graph to obtain the current focus intent node. Then, the connection relationship between the current focus intent node and each historical intent node in the initial graph structure is analyzed in the graph, and the shortest path length and shared rule nodes between them are calculated. Finally, based on the tightness of these connections, for example, taking the current focus intent node as the center, all historical intent nodes with path lengths within a certain range and the path nodes between them are included, delineating a relatively compact graph range in the vast business process knowledge graph. This graph range is ultimately defined as the key decision subgraph, which includes the current focus intent node, the selected historical intent nodes, and other process nodes and rule nodes necessary to connect them.

[0108] For example, following the specific implementation of the previous step, firstly, the focus intent of the dynamic cognitive focus is to report a lost card. A search is conducted in the historical dialogue database to find two relevant historical fragments: fragment X's historical focus information is reporting a lost card, and fragment Y is applying for a new card. Secondly, the comprehensive matching degree between the current focus and these two historical fragments is calculated. It is found that fragment X has a much higher comprehensive matching degree than fragment Y because of its highly similar intent and its historical semantic distance and process logic tightness are very close to the current value. Therefore, fragment X is selected as a reference. Next, the historical intents processed at that time are extracted from fragment X, including verifying identity, freezing account, and reporting a lost card. These historical intents are mapped to the business process knowledge graph to obtain the corresponding process nodes such as identity verification, application for freezing, and application for reporting a lost card. Then, based on these process nodes, the shortest path connecting them is found in the knowledge graph, forming an initial graph structure containing these nodes and their connections. Simultaneously, the current focus intent of reporting a lost card is mapped to the same initial graph structure, resulting in the application to report a lost card node. Subsequently, the connection relationships between this node and each historical node in the initial graph structure are calculated, revealing direct connections to nodes such as application freezing and identity verification. Finally, based on these close connections, a scope is defined within the complete knowledge graph. This scope includes nodes such as identity verification, application freezing, application to report a lost card, and subsequent options for reporting a lost card, as well as the business rules between them. This defined local graph is the key decision subgraph for the current dialogue.

[0109] This step uses intelligent matching of historical experience to quickly locate the core decision-making area most relevant to the current context within a vast business process knowledge graph. This avoids the inefficiency of global search and allows for more precise focus on a limited range of highly relevant business process nodes and rules. This provides a clear and efficient scope for subsequent process optimization, improving the accuracy and response speed of decisions.

[0110] Step 105: Based on the complexity of the key decision subgraph and the emotional features in the dialogue data of the current round, optimize the dynamic dialogue state network in real time, and use the optimized dynamic dialogue state network to adjust the execution logic in the business process knowledge graph to generate a business process execution plan.

[0111] Optionally, step 105 may specifically include: Step 1051: Count the total number of process nodes and rules contained in the key decision subgraph to obtain the first quantification value.

[0112] Step 1052: Analyze the types and number of logical dependencies between process nodes in the key decision sub-graph to obtain a second quantification value.

[0113] Step 1053: Combine the first quantization value with the second quantization value to calculate the complexity quantization result of the key decision subgraph.

[0114] Step 1054: Perform sentiment analysis on the current dialogue data to extract the emotional state information and emotional fluctuation intensity information from the current dialogue data.

[0115] Step 1055: Based on the complexity quantification result, generate first optimization parameters for the first structural optimization of the dynamic dialogue state network.

[0116] Step 1056: Based on the emotional state information and the emotional fluctuation intensity information, generate second optimization parameters for the second structural optimization of the dynamic dialogue state network.

[0117] Step 1057: Apply the first optimization parameter and the second optimization parameter to the dynamic dialogue state network to adjust the internal resource allocation weights of the dynamic dialogue state network and complete the real-time optimization of the dynamic dialogue state network.

[0118] Step 1058: Based on the optimized dynamic dialogue state network, analyze the key decision subgraph to determine the execution priority of each process node in the key decision subgraph.

[0119] Step 1059: Adjust the logical connection relationship between process nodes in the key decision sub-graph according to the execution priority to form a preliminary logical sequence.

[0120] Step 10510: Input the preliminary logical sequence into the optimized dynamic dialogue state network, and evaluate the connectivity and consistency of the preliminary logical sequence in the overall business process knowledge graph through the optimized dynamic dialogue state network.

[0121] Step 10511: Based on the evaluation results, the connection relationships and rule application order between process nodes in the preliminary logical sequence are corrected through the optimized dynamic dialogue state network to generate an optimized logical sequence.

[0122] Step 10512: Based on the optimized logical sequence, extract the corresponding process nodes, rules and logical relationships from the business process knowledge graph and combine them to generate a business process execution plan.

[0123] In this step, the optimized dynamic dialogue state network refers to the new version of the network obtained by adjusting the internal parameters of the original dynamic dialogue state network according to the complexity of the current specific scenario and the user's emotions. It is used to process the current dialogue more accurately and guide the adjustment of the process logic.

[0124] A business process execution plan refers to a specific and actionable sequence of task steps that clearly defines which business nodes should be executed in the current dialogue context, in what order, and according to what rules, to directly guide the customer service system in completing the service.

[0125] The first quantification value refers to the total number of process nodes and rule nodes contained in the key decision subgraph. It is used to quantify the structural size of the subgraph and is obtained by traversing the subgraph and counting all process nodes and rule nodes.

[0126] The second quantification value refers to the number of different types of logical dependencies existing between different process nodes in the key decision subgraph. It is used to quantify the complexity of the internal logic of the subgraph and is obtained by analyzing the types of all edges in the subgraph and performing deduplication counting.

[0127] The complexity quantification result refers to a comprehensive numerical value that reflects the overall complexity of the key decision subgraph in terms of both structural size and logical complexity, and is used as a basis for adjusting the dialogue state network.

[0128] Emotional state information refers to the user's main emotional category identified from the dialogue text in the current round, such as calm, anxiety, anger, or satisfaction. It is used to understand the user's current psychological state in order to adjust the interaction strategy. It is obtained by analyzing the dialogue text using a pre-trained emotion classification model.

[0129] Emotional fluctuation intensity information refers to a quantitative value that reflects the intensity of a user's emotions in the current round of dialogue or the magnitude of emotional changes compared to previous rounds. It is used to determine the urgency of a user's emotions and is obtained by analyzing the confidence score difference output by the emotion classification model.

[0130] The first optimization parameter refers to a set of numerical instructions used to adjust the internal structure of the dynamic dialogue state network. The adjustment direction is to make the network more adaptable to processing highly complex business subgraphs. It is used to optimize the network related to business complexity. The complex quantification results are converted into network parameter adjustment quantities through a preset mapping function.

[0131] The second set of optimization parameters refers to another set of numerical instructions used to adjust the internal structure of the dynamic dialogue state network. The adjustment direction is to make the network more sensitive and appropriate to respond to the user's different emotional states. It is used to optimize the network in relation to the user's emotions. The emotional state information and emotional fluctuation intensity information are converted into network parameter adjustment values ​​through another preset mapping function.

[0132] Internal resource allocation weights refer to the proportion of importance assigned to different computing units or attention mechanisms within a dynamic dialogue state network during information processing, used to control the degree of attention the network pays to different parts of the input information.

[0133] Execution priority refers to a level or value assigned to each process node in the critical decision subgraph, representing its execution order. It is used to determine the initial order of these business steps and is calculated after analyzing the contextual relationships of the nodes in the subgraph through an optimized dynamic dialogue state network.

[0134] The preliminary logical sequence refers to a list of node execution order formed after the process nodes in the key decision sub-graph are initially sorted according to execution priority. It serves as the starting point for subsequent evaluation and optimization. It is obtained by arranging process nodes from high to low execution priority and taking into account the rule constraints between them.

[0135] The optimized logical sequence refers to a more reasonable and feasible list of final node execution order obtained after global consistency and fluency evaluation and correction based on the initial logical sequence. It is used to directly guide the generation of business process execution plans and is obtained by evaluating and iteratively correcting the initial sequence through an optimized dynamic dialogue state network.

[0136] In this step, the key decision subgraph is first traversed to count the total number of all process nodes and rule nodes, resulting in a first quantification value. Simultaneously, all edges connecting the process nodes in the key decision subgraph are analyzed to identify different types of logical dependencies, such as sequential execution, conditional branching, and parallel execution, and their number is counted, resulting in a second quantification value. Next, these two quantification values ​​are input into a preset formula to calculate a comprehensive complexity quantification result. In a parallel thread, the dialogue data text of the current round is input into a pre-trained sentiment analysis model. This model outputs the user's current emotional state information, such as anxiety, and also outputs sentiment fluctuation intensity information representing the current emotional intensity. Next, based on the complexity quantification results, a parameter generator produces the first set of optimization parameters. The logic is that the higher the complexity, the more attentional resources the network should invest in clarifying complex relationships. Simultaneously, based on emotional state information and emotional fluctuation intensity information, another parameter generator produces the second set of optimization parameters. The logic is that if the emotion is negative and the intensity is high, the optimization parameters should prioritize processing the business path most relevant to alleviating the emotion. Then, these two sets of optimization parameters are applied to the dynamic dialogue state network. In essence, according to the parameter guidance, the connection weights between neurons in each layer of the network are adjusted, that is, the internal resource allocation weights, thereby completing the real-time optimization of the network. Furthermore, the optimized dynamic dialogue state network receives structural information from the key decision subgraph as input. By analyzing the tightness of connections between nodes in the key decision subgraph, the strength of rule constraints, and the current dialogue context, it calculates an execution priority score for each process node. Then, based on this priority score, the nodes are arranged from high to low, and the rules that the nodes must follow are initially considered to form a preliminary logical sequence. This preliminary logical sequence, along with the global structural information of the entire business process knowledge graph, is then input into the optimized dynamic dialogue state network again. The task of the dynamic dialogue state network is to evaluate the feasibility of this preliminary logical sequence in the entire business graph: checking whether the preliminary logical sequence has logical conflicts with other nodes in the business graph that are not included in the subgraph, and whether the transitions between steps within the sequence are natural and consistent. Finally, the dynamic dialogue state network outputs an evaluation result, pointing out potential problems in the initial logical sequence. Based on this evaluation result, the initial logical sequence is adjusted, such as changing the order of certain nodes or inserting necessary steps that were previously ignored between specific nodes, thereby generating a better optimized logical sequence. Ultimately, based on this optimized logical sequence, like a navigator, the network extracts detailed operation descriptions, triggering rules, and connections between nodes from the business process knowledge graph, and combines and encapsulates this information to form a complete and step-by-step business process execution plan.

[0137] For example, following the specific implementation of the previous step, the key decision subgraph first includes nodes such as identity verification, application freezing, and application loss reporting. The total number of nodes and rules in the key decision subgraph is counted to obtain a first quantification value of 9. Sequential execution and conditional branching relationships are identified to obtain a second quantification value of 2. Based on this, a higher complexity quantification result is calculated. Next, the user's current statement, "Help me process this quickly, I'm in a hurry!", is analyzed. The sentiment analysis model determines the emotional state as anxiety with high fluctuation intensity. Then, based on the high complexity and high anxiety, two sets of optimization parameters are generated and applied to dynamically adjust the internal weights of the dialogue state network, completing real-time optimization. Then, the optimized network analysis of the key decision subgraph, combined with user anxiety, prioritized the application freeze node that directly handles risk, thus forming a preliminary logical sequence: identity verification -> application freeze -> application loss reporting. This preliminary logical sequence was then evaluated within the entire business process knowledge graph. It was found that jumping directly from application freeze to application loss reporting might overlook a necessary confirmation step. Therefore, the preliminary logical sequence was revised by inserting a confirmation node for the consequences of loss reporting, generating an optimized logical sequence: identity verification -> application freeze -> confirmation of consequences of loss reporting -> application loss reporting. Finally, based on this optimized logical sequence, specific operational instructions and rules for each step were extracted from the knowledge graph and combined to generate a business process execution plan that can be directly guided.

[0138] This step dynamically optimizes the dialogue understanding engine by quantifying business complexity and user emotions. The optimized engine then generates execution sequences that take into account both global logic and real-time user status within key business areas. This achieves a closed-loop process from contextual understanding to personalized, executable solutions, enabling business processes to adapt in real time and with flexibility, significantly improving service accuracy and user experience.

[0139] Figure 2 This application provides a schematic diagram of the structure of a business process adaptive execution system based on a dynamic dialogue state network, as shown below. Figure 2 As shown, the system includes: The acquisition module 21 is used to acquire current service data and process knowledge. The current service data includes current dialogue data and historical dialogue data, and the process knowledge is a business process knowledge graph that includes process nodes, rules and logical relationships. The identification module 22 is used to parse the current round of dialogue data through a pre-deployed dynamic dialogue state network to identify the dominant intent and related intent in the current dialogue data; The calculation module 23 is used to calculate the semantic distance and process logic tightness between the dominant intent and the associated intent based on the business process knowledge graph, and generate a dynamic cognitive focus based on the semantic distance and the process logic tightness. The positioning module 24 is used to compare the dynamic cognitive focus with the historical dialogue data, filter historical dialogue fragments from the historical dialogue data according to the comparison results, and locate the key decision subgraph in the business process knowledge graph according to the filtered historical dialogue fragments. The optimization module 25 is used to optimize the dynamic dialogue state network in real time based on the complexity of the key decision subgraph and the emotional features in the dialogue data of the current round, and to adjust the execution logic in the business process knowledge graph using the optimized dynamic dialogue state network to generate a business process execution plan.

[0140] Figure 2 The aforementioned adaptive execution system for business processes based on a dynamic dialogue state network can execute... Figure 1 The implementation principle and technical effects of the adaptive execution method for business processes based on a dynamic dialogue state network described in the illustrated embodiment will not be repeated here. The specific execution methods of each module and unit in the adaptive execution system for business processes based on a dynamic dialogue state network in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0141] In one possible design, Figure 2 The business process adaptive execution system based on a dynamic dialogue state network, as shown in the embodiment, can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0142] The processing component 32 is used for the above Figure 1 The above embodiment provides a business process adaptive execution method based on a dynamic dialogue state network.

[0143] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0144] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0145] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0146] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0147] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0148] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0149] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is a business process adaptive execution method based on a dynamic dialogue state network.

[0150] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0151] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0152] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A business process adaptive execution method based on a dynamic dialogue state network, characterized in that, include: Acquire current service data and process knowledge. The current service data includes current dialogue data and historical dialogue data. The process knowledge is a business process knowledge graph that includes process nodes, rules, and logical relationships. The current round of dialogue data is parsed using a pre-deployed dynamic dialogue state network to identify the dominant and associated intentions in the current dialogue data. Based on the business process knowledge graph, the semantic distance and process logic tightness between the dominant intent and the associated intent are calculated, and a dynamic cognitive focus is generated based on the semantic distance and the process logic tightness. The dynamic cognitive focus is compared with the historical dialogue data. Based on the comparison results, historical dialogue fragments are selected from the historical dialogue data. Based on the selected historical dialogue fragments, key decision subgraphs are located in the business process knowledge graph. Based on the complexity of the key decision subgraph and the emotional features in the dialogue data of the current round, the dynamic dialogue state network is optimized in real time, and the execution logic in the business process knowledge graph is adjusted using the optimized dynamic dialogue state network to generate a business process execution plan.

2. The method according to claim 1, characterized in that, The current round of dialogue data is parsed using a pre-deployed dynamic dialogue state network to identify the dominant and related intentions within the current dialogue data, including: Extract the user's statement in the current round and the previously proceeding dialogue statements from the current dialogue data; The user statement is parsed to obtain explicit query components; The user statement is subjected to a second parsing process to obtain implicit context components; The explicit query components, the implicit context components, and the previously executed dialogue statements are combined to form dialogue context information; Joint analysis is performed on the dialogue context information to calculate the association strength between the first candidate intent corresponding to the explicit query component and the second candidate intent corresponding to the implicit context component and the previously advanced dialogue statement; The first candidate intent and the second candidate intent are then integrated to generate a candidate intent set; In the set of candidate intentions, the candidate intentions whose association strength meets the first preset benchmark are determined as the dominant intentions; In the candidate intent set, intents whose semantic distance is within a preset range and whose association strength meets a second preset benchmark are determined as associated intents.

3. The method according to claim 1, characterized in that, Based on the business process knowledge graph, the semantic distance and process logic tightness between the dominant intent and the associated intent are calculated, and a dynamic cognitive focus is generated based on the semantic distance and the process logic tightness, including: The dominant intent and the associated intent are respectively mapped to process nodes in the business process knowledge graph to obtain the dominant intent node and the associated intent node; Calculate the connection path between the dominant intent node and the associated intent node in the business process knowledge graph; The semantic distance between the dominant intent and the associated intent is determined based on the number of nodes traversed by the connection path. In the business process knowledge graph, identify the rule nodes that are commonly connected to the dominant intent node and the associated intent node, as well as the logical dependencies between the dominant intent node and the associated intent node; Based on the number of the commonly connected rule nodes and the type of the logical dependency relationship, the flow logic tightness between the dominant intent and the associated intent is calculated. By combining the semantic distance and the logical tightness of the process, the cognitive focus quantification value of all associated intentions is calculated; Compare all cognitive focus quantification values ​​and select the relevance intent with the largest cognitive focus quantification value as the focus intent; The focus intent, the semantic distance corresponding to the focus intent, and the tightness of the process logic are combined to obtain a dynamic cognitive focus.

4. The method according to claim 1, characterized in that, The dynamic cognitive focus is compared with the historical dialogue data, and historical dialogue segments are selected from the historical dialogue data based on the comparison results, including: Multiple historical dialogue fragments and corresponding historical focus information are obtained from the historical dialogue data. The historical focus information includes historical focus intent, historical semantic distance, and historical process logic tightness. From the dynamic cognitive focus, extract the current focus intent, current semantic distance, and current process logic tightness; Calculate the first similarity between the current focus intent and each of the historical focus intents, calculate the second similarity between the current semantic distance and each of the historical semantic distances, and calculate the third similarity between the current process logic tightness and each of the historical process logic tightness; For each historical dialogue segment, the first similarity, the second similarity, and the third similarity are merged to obtain the comprehensive matching degree between the historical dialogue segment and the dynamic cognitive focus; From the multiple historical dialogue segments, select the historical dialogue segments whose overall matching degree meets the preset threshold condition.

5. The method according to claim 1, characterized in that, Based on the selected historical dialogue fragments, key decision subgraphs are located in the business process knowledge graph, including: Extract the processed historical intent from each of the selected historical dialogue segments; The historical intents are mapped onto the business process knowledge graph to obtain multiple corresponding historical intent nodes; In the business process knowledge graph, the path connecting the multiple historical intent nodes is determined, and an initial graph structure containing the multiple historical intent nodes is determined based on the path; Extract the current focus intent from the dynamic cognitive focus, and map the current focus intent to the business process knowledge graph to obtain the current focus intent node; In the business process knowledge graph, the connection relationship between the current focus intent node and each historical intent node in the initial graph structure is calculated; Based on the connection relationship, a graph range is defined in the business process knowledge graph, and the process nodes, rules and logical relationships involved in the graph range are determined as key decision subgraphs. The graph range includes the current focus intent node and at least one of the historical intent nodes.

6. The method according to claim 1, characterized in that, Based on the complexity of the key decision subgraph and the emotional features in the dialogue data of the current round, the dynamic dialogue state network is optimized in real time, including: The total number of process nodes and rules contained in the key decision subgraph is counted to obtain the first quantitative value; The second quantification value is obtained by analyzing the types and number of logical dependencies between each process node in the key decision sub-graph. The first quantized value and the second quantized value are combined to calculate the complexity quantization result of the key decision subgraph; Perform sentiment analysis on the current dialogue data to extract emotional state information and emotional fluctuation intensity information from the current dialogue data; Based on the complex quantification results, first optimization parameters are generated to perform first structural optimization on the dynamic dialogue state network; Based on the emotional state information and the emotional fluctuation intensity information, second optimization parameters are generated to perform a second structural optimization on the dynamic dialogue state network. The first optimization parameter and the second optimization parameter are applied to the dynamic dialogue state network to adjust the internal resource allocation weights of the dynamic dialogue state network, thereby completing the real-time optimization of the dynamic dialogue state network.

7. The method according to claim 1, characterized in that, The optimized dynamic dialogue state network is used to adjust the execution logic in the business process knowledge graph to generate a business process execution plan, including: Based on the optimized dynamic dialogue state network, the key decision subgraph is analyzed to determine the execution priority of each process node in the key decision subgraph. Based on the execution priority, adjust the logical connection relationship between the process nodes in the key decision sub-graph to form a preliminary logical sequence; The preliminary logical sequence is input into the optimized dynamic dialogue state network, and the connectivity and consistency of the preliminary logical sequence in the overall business process knowledge graph are evaluated through the optimized dynamic dialogue state network. Based on the evaluation results, the connection relationships between process nodes and the order of rule application in the preliminary logic sequence are corrected through the optimized dynamic dialogue state network to generate an optimized logic sequence. Based on the optimized logical sequence, corresponding process nodes, rules, and logical relationships are extracted from the business process knowledge graph and combined to generate a business process execution plan.

8. A business process adaptive execution system based on a dynamic dialogue state network, characterized in that, include: The acquisition module is used to acquire current service data and process knowledge. The current service data includes current dialogue data and historical dialogue data, and the process knowledge is a business process knowledge graph that includes process nodes, rules and logical relationships. The identification module is used to parse the current round of dialogue data through a pre-deployed dynamic dialogue state network to identify the dominant intent and related intents in the current dialogue data; The calculation module is used to calculate the semantic distance and process logic tightness between the dominant intent and the associated intent based on the business process knowledge graph, and generate a dynamic cognitive focus based on the semantic distance and the process logic tightness. The positioning module is used to compare the dynamic cognitive focus with the historical dialogue data, filter historical dialogue fragments from the historical dialogue data according to the comparison results, and locate the key decision subgraph in the business process knowledge graph according to the filtered historical dialogue fragments. The optimization module is used to optimize the dynamic dialogue state network in real time based on the complexity of the key decision subgraph and the emotional features in the dialogue data of the current round, and to adjust the execution logic in the business process knowledge graph using the optimized dynamic dialogue state network to generate a business process execution plan.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the adaptive execution method of business processes based on a dynamic dialogue state network as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The system stores a computer program, which, when executed by a computer, implements a business process adaptive execution method based on a dynamic dialogue state network as described in any one of claims 1 to 7.

Images