Business intelligence decision-making method, apparatus, device, and medium

Abstract

This invention relates to the field of intelligent decision-making technology, and discloses a business intelligent decision-making method, apparatus, device, and medium. The method includes: determining a computing resource allocation strategy based on the session characteristics of a business consultation session; splitting the business consultation session into business sub-sessions based on the computing resource allocation strategy; constructing a directed topology graph based on the logical dependencies of the business sub-sessions; generating a query statement based on the session semantics of the directed topology graph; retrieving candidate document fragments from multiple business knowledge document libraries based on the query statement; extracting logical constraints of the business sub-sessions and document summaries of the candidate document fragments; determining that there is no logical conflict between the business sub-session and the candidate document fragments when the document summaries satisfy the logical constraints; constructing a logical chain between the candidate document fragments and the business sub-sessions; and generating a session response text based on the logical chain. This invention can be applied to business system platforms in fintech, healthcare, and other fields, and can improve the accuracy of business decisions.

Description

Technical Field

[0001] This invention relates to the field of intelligent decision-making technology, and in particular to a business intelligent decision-making method, apparatus, equipment, and medium. Background Technology

[0002] With the widespread application of intelligent business consulting systems, scenarios where users conduct business consultations through natural language are increasing. Existing intelligent business decision-making methods typically treat the entire user consultation session as a whole, directly using large language models to generate decision responses through single-turn or multi-turn dialogues. However, in real-world business scenarios, user consultation sessions often contain multiple interrelated business sub-questions, with complex logical dependencies between these sub-questions, and different sub-questions may require consulting different business knowledge document libraries.

[0003] Existing intelligent decision-making technologies for business applications suffer from the following shortcomings. Firstly, existing technologies typically employ uniform inference resources across all sessions, failing to dynamically allocate computing resources based on the actual complexity of each session. This leads to wasted resources in simple sessions and insufficient inference capabilities in complex sessions. Secondly, existing technologies often adopt end-to-end generative decision responses, failing to break down complex consultation sessions into logically dependent business sub-sessions. This makes it difficult to guarantee logical consistency throughout the multi-step inference process, potentially causing decisions that contradict the user's actual business needs and reducing the reliability of the decision results.

[0004] For example, in healthcare settings, patient consultations often involve multiple interconnected sub-questions, such as symptom descriptions, medication history, and past medical history. These sub-questions are logically related and require reference to different document libraries, such as clinical guidelines, drug instructions, and medical insurance policies. Existing methods struggle to verify the logical consistency of multi-source knowledge, easily generating treatment recommendations that contradict the patient's actual condition.

[0005] For example, in fintech scenarios, user financial advice often involves multiple sub-questions such as risk preference assessment, product comparison, and compliance review. These sub-questions are conditionally dependent on each other and require searching different document libraries, such as product brochures, regulatory rules, and market data. Existing methods struggle to detect logical conflicts between search results and user needs during multi-step reasoning, easily generating financial decision-making recommendations that violate compliance requirements.

[0006] Therefore, how to implement a business intelligence decision-making method to improve the accuracy and reliability of business decisions is a technical problem that urgently needs to be solved. Summary of the Invention

[0007] This invention provides a business intelligent decision-making method, apparatus, equipment, and medium, the main purpose of which is to solve the problem of low accuracy in business decision-making.

[0008] Firstly, to achieve the above objectives, the present invention provides a business intelligence decision-making method, comprising: Obtain the business consultation sessions of the target user and extract the session features of the business consultation sessions; The computing power resource allocation strategy for the business consultation session is determined based on the session characteristics, and the computing power resources corresponding to the business consultation session are allocated according to the computing power resource allocation strategy. The business consultation session is divided into multiple business sub-sessions based on the allocated computing resources, and the logical dependencies between the multiple business sub-sessions are identified. Construct a directed topology graph of the business sub-sessions based on the logical dependencies; The session semantics contained in each business sub-session in the directed topology graph are parsed, and a query statement for each business sub-session is generated based on the session semantics. The query statement is used to search multiple preset business knowledge document libraries to obtain multiple candidate document fragments corresponding to each business sub-session; Extract the logical constraints of each business sub-session and the document summary of the corresponding candidate document fragment, and determine whether the document summary satisfies the logical constraints; When the document summary satisfies the logical constraints, it is determined that there is no logical conflict between the business sub-session and the corresponding candidate document fragment, and a logical chain is established between the candidate document fragment and the business sub-session. The session response text for the business consultation session is generated based on the logical chain.

[0009] Secondly, the present invention also provides a business intelligent decision-making device, comprising: The session feature extraction module is used to acquire the business consultation session of the target user and extract the session features of the business consultation session; The computing power resource allocation module is used to determine the computing power resource allocation strategy for the business consultation session based on the session characteristics, and allocate the computing power resources corresponding to the business consultation session according to the computing power resource allocation strategy. The logical relationship identification module is used to split the business consultation session into multiple business sub-sessions based on the allocated computing resources, and to identify the logical dependencies between the multiple business sub-sessions. A directed graph construction module is used to construct a directed topology graph of the business sub-session based on the logical dependencies; The query statement generation module is used to parse the session semantics contained in each business sub-session in the directed topology graph, and generate a query statement for each business sub-session based on the session semantics; The knowledge document retrieval module is used to search in multiple preset business knowledge document databases using the query statement to obtain multiple candidate document fragments corresponding to each business sub-session. The logical conflict judgment module is used to extract the logical constraints of each business sub-session and the document summary of the corresponding candidate document fragment, and to determine whether the document summary satisfies the logical constraints. The logical conflict determination module is used to determine that there is no logical conflict between the business sub-session and the corresponding candidate document fragment when the document summary satisfies the logical constraint conditions, and to establish a logical chain between the candidate document fragment and the business sub-session. The decision text generation module is used to generate the session response text of the business consultation session according to the logic chain.

[0010] Thirdly, the present invention also provides an electronic device, the electronic device comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the business intelligence decision-making method described above.

[0011] Fourthly, the present invention also provides a computer-readable storage medium storing at least one computer program, which is executed by a processor in an electronic device to implement the business intelligence decision-making method described above.

[0012] In this embodiment of the invention, by extracting session features and dynamically allocating computing resources, on-demand scheduling and refined management of computing resources are achieved, avoiding resource idleness or excessive contention. Based on computing resources, the entire session is divided into multiple sub-sessions and a directed topology graph is constructed, which can explicitly characterize the logical dependencies between subtasks. By recognizing session semantics to generate structured query statements and retrieving candidate fragments in parallel across multiple business knowledge document bases, natural language intent can be efficiently transformed into executable retrieval operations, reducing information gaps caused by the limitations of a single knowledge base. Furthermore, separate retrieval for each sub-session avoids redundant computation and interference caused by global retrieval, improving retrieval coverage and accuracy. Extracting the logical constraints of sub-sessions and summaries of candidate document fragments, and performing constraint satisfaction judgments, can filter irrelevant or conflicting knowledge fragments early on, reducing invalid computations in subsequent stages and forming a traceable binding path from user intent to reliable knowledge sources, providing an interpretable basis for the final response decision. This enhances the logical consistency, credibility, and robustness of business decision processing. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of an application environment for the business intelligent decision-making method in one embodiment of the present invention. Figure 2 A flowchart illustrating a business intelligent decision-making method provided in an embodiment of the present invention; Figure 3 A functional block diagram of a business intelligent decision-making device provided in an embodiment of the present invention; Figure 4 A schematic diagram of the structure of an electronic device for implementing a business intelligent decision-making method according to an embodiment of the present invention; Figure 5 This is another structural schematic diagram of an electronic device for implementing a business intelligent decision-making method according to an embodiment of the present invention.

[0015] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0016] It should be noted that in the technical solutions disclosed in this invention, the acquisition of user information (personal image data (e.g., facial videos or pictures, facial feature videos or pictures, etc.) and personal privacy information (e.g., name, ID number, occupation, address, etc.)) is all completed with the user's knowledge and consent, and the acquisition of the relevant user information is legal and compliant.

[0017] To enable those skilled in the art to better understand the technical solutions of this disclosure, and to fully understand and implement the process of how this disclosure applies technical means to solve technical problems and achieve corresponding technical effects, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. The embodiments of this disclosure and the various features within them can be combined with each other without conflict, and the resulting technical solutions are all within the protection scope of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort should fall within the protection scope of this disclosure.

[0018] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0019] This application provides a business intelligent decision-making method. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the device provided in this application: a server, a terminal, or other electronic equipment. In other words, the business intelligent decision-making method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0020] The business intelligent decision-making method of this invention can be applied to, for example... Figure 1 In this application environment, the client communicates with the server via a network. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will now be described in detail through specific embodiments.

[0021] Reference Figure 2 The diagram shown is a flowchart illustrating a business intelligence decision-making method according to an embodiment of the present invention. In this embodiment, the business intelligence decision-making method includes: S1. Obtain the business consultation session of the target user and extract the session characteristics of the business consultation session.

[0022] In this embodiment of the invention, the business consultation session refers to a complete interaction process conducted by a user around a specific business objective (such as product consultation, fault reporting, order inquiry, etc.), which usually includes multiple rounds of question and answer and information exchange. The session features refer to a set of quantifiable or categorizable attributes extracted from the business consultation session, including but not limited to session length, number of interaction rounds, keyword distribution, semantic topic, intent category, timestamp, user sentiment tendency, etc., which are used to describe the nature of the session and processing requirements.

[0023] In this embodiment of the invention, a complete business consultation session data stream is received from the interaction channel of the target user, and a session segmentation operation is performed on the data stream to convert the continuous dialogue stream into several independent business sub-session units based on semantic turning points and time intervals.

[0024] Multi-dimensional feature extraction is performed on each business sub-session, quantifying and representing it from multiple dimensions such as the session's intent type, the business domain involved, the depth of logical dependencies, the scale of candidate knowledge, and the complexity of the overall inference chain. The evaluation results of each dimension are organized into a structured session feature vector. These session features comprehensively depict the current business consultation session in terms of semantic complexity, knowledge coverage, and logical reasoning requirements, providing accurate decision-making basis for subsequent allocation of computing resources and formulation of retrieval strategies.

[0025] For example, in a healthcare scenario, the dialogue flow between a patient and an intelligent consultation assistant is acquired, and the conversation is segmented into independent business sub-conversations based on semantic transitions such as symptom description, medication inquiry, and examination result query. For the sub-conversation about whether to seek medical attention after experiencing chest tightness for three days, its intent is extracted as symptom assessment, involving the cardiovascular field, including slots for age and medical history, and requiring access to the clinical guideline knowledge base. The conversation characteristics are high-risk and of medium complexity.

[0026] For example, in a fintech scenario, the conversation flow between a user and a financial assistant is captured, and sub-conversations are segmented based on turning points such as product consultation, risk assessment, and return inquiry. For the sub-conversation "Does this financial product guarantee principal?", its intent is extracted as risk confirmation, involving fixed-income products, including amount and term slots, and requiring access to product brochures and regulatory compliance documents. The conversation characteristics are high compliance requirements and medium inference depth.

[0027] In this embodiment of the invention, by acquiring the business consultation sessions of the target user and extracting session features, unstructured natural language interactions can be transformed into quantifiable structured data, providing an objective basis for subsequent resource allocation and task breakdown.

[0028] S2. Determine the computing power resource allocation strategy for the business consultation session based on the session characteristics, and allocate the computing power resources corresponding to the business consultation session according to the computing power resource allocation strategy.

[0029] In this embodiment of the invention, the computing resources refer to the processing capabilities that can be used to perform computing tasks, including hardware resources such as processors, memory, storage, and network bandwidth, as well as their virtualization forms.

[0030] In this embodiment of the invention, a comprehensive analysis is performed on all the session characteristics of the business consultation session. These characteristics cover dimensions such as the intent complexity of the session, the number of business sub-sessions involved, the depth of logical dependencies, the size of candidate document fragments, and the reasoning span of the logical chain.

[0031] The aforementioned features are input into a pre-defined resource strategy mapping model. Based on a comprehensive evaluation of each feature dimension, this model matches the most suitable computing resource allocation strategy to the current session's complexity and execution requirements. This strategy clarifies the types and quantities of computing resources required by the session at each processing stage. According to this strategy, a dedicated computing resource quota is allocated to this business consultation session from the global computing resource pool. The required semantic encoding resources, vector retrieval resources, logical reasoning resources, etc., are bound and scheduled according to the strategy requirements, ensuring that the session receives computing power support commensurate with its complexity throughout the entire execution process, avoiding both resource waste and processing delays due to insufficient resources.

[0032] For example, in a healthcare scenario, the session was identified as involving clinical diagnostic consultation, including multiple rounds of symptom follow-up and medication logic constraints. The session was characterized by high risk, strong logical dependencies, and the need to access clinical guidelines and drug instruction manuals. Accordingly, the session was assigned to the highest priority resource queue, high-performance inference computing power was allocated to ensure low-latency response, dedicated GPU memory was isolated to ensure patient privacy data was not shared with other sessions, and compliance auditing computing power was reserved for verifying the medical accuracy of the output content.

[0033] For example, in a fintech scenario, if a session is identified as involving risk assessment of wealth management products, including returns inquiries and compliance checks, and the session is characterized by high compliance requirements, moderate inference depth, and the need to access product descriptions and regulatory documents, it is assigned to a high-priority queue but lower than real-time transaction tasks. Dedicated inference computing power deployed locally is used to meet the security requirement of data remaining within the data domain. Simultaneously, content review computing power is allocated to filter the generated results for compliance, ensuring no misleading statements appear. Furthermore, some computing power is automatically reclaimed during periods of low business activity to control costs.

[0034] In this embodiment of the invention, the computing resource allocation strategy is dynamically determined based on session characteristics, enabling on-demand scheduling and fine-grained allocation of computing resources, avoiding resource idleness or excessive occupation. This mechanism supports allocating more computing power to high-priority or high-complexity sessions, ensuring their response time and processing quality.

[0035] S3. Based on the allocated computing resources, the business consultation session is divided into multiple business sub-sessions, and the logical dependencies between the multiple business sub-sessions are identified.

[0036] In this embodiment of the invention, the computing power resources refer to the reserve capacity for processing computing tasks, including CPU, GPU, memory, network bandwidth, etc., which determine how many tasks the system can process simultaneously or sequentially and their complexity. The business sub-session refers to a smaller task segment broken down from a business consultation session according to computing power processing units. Each sub-session corresponds to a part of business logic that can be completed independently or relatively independently. The logical dependency relationship refers to the execution order or conditional triggering relationship between multiple business sub-sessions, such as the output of one sub-session being the input of another sub-session, or a sub-session must be completed before it can begin.

[0037] For example, in a fintech scenario, complex questions raised by users can be broken down into several sub-sessions with logical sequence. These sub-sessions should facilitate subsequent retrieval of insurance terms, regulatory rules, and claims cases. There should be clear dependencies between sub-sessions, and subsequent sessions can refer to the conclusions of previous questions. Each sub-session should ideally contain only one core decision point.

[0038] For example, a user's inquiry session might be titled "A customer has been involved in two traffic accidents and made one critical illness claim within a year. Can they still purchase critical illness insurance? What are the underwriting conditions?" This could be broken down into several sub-sessions, such as "Does the customer have a history of critical illness? Is the previous critical illness covered by the current critical illness insurance? Under the above conditions, does the customer meet the underwriting conditions of the current product?"

[0039] In this embodiment of the invention, identifying the logical dependencies between the multiple service sub-sessions includes: Obtain the session timestamp corresponding to each of the service sub-sessions, and sort all the service sub-sessions in ascending order according to the session timestamps to form target service sub-sessions arranged in chronological order. Extract the input parameter set and output parameter set of the target service sub-session; The intersection of the output parameter set of the current target business sub-session and the input parameter set of the next target business sub-session is checked. If an intersection exists, it is determined that there is a data dependency relationship between the current target business sub-session and the next target business sub-session. Obtain the business process identifier corresponding to the current target business sub-session, and query the set of preceding session identifiers and the set of following session identifiers corresponding to the business process identifier from the preset business process relationship database; If the previous current business sub-session of the current target business sub-session belongs to the preceding session identifier set, or the next target business sub-session belongs to the following session identifier set, then it is determined that there is a process sequence dependency relationship. The data dependencies and process sequence dependencies are merged and deduplicated to generate logical dependencies between the multiple business sub-sessions.

[0040] In this embodiment of the invention, a corresponding session timestamp is extracted from each business sub-session. This timestamp records the precise moment when the sub-session was created or executed. Using the timestamp as the sorting basis, all business sub-sessions are sorted in ascending order, reorganizing all sub-sessions according to their earliest to latest time sequence, generating a target business sub-session sequence strictly arranged chronologically. This sorting provides an ordered traversal basis for subsequent pairwise detection of dependencies between adjacent sessions, ensuring that dependency detection is always performed between sessions that are temporally adjacent.

[0041] Specifically, each sorted target business sub-session undergoes a parameter scan to identify all input parameters received during the session's execution, aggregating them into an input parameter set. Simultaneously, all output parameters generated and transmitted outwards after the session's completion are identified and aggregated into an output parameter set. The input parameter set characterizes which external data the session depends on to initiate, while the output parameter set characterizes which data the session provides downstream after completion. Together, they define the input and output boundaries of each session in the data flow, providing a clear comparison object for subsequent data dependency detection.

[0042] Specifically, the output parameter set of the current target business sub-session is compared element-wise with the input parameter set of the next target business sub-session in the time series. When there is at least one identical parameter in the two sets, it indicates that the output of the previous session is exactly the input required by the next session, and data has been passed from one session to the other. Based on this, it is determined that there is a data dependency between the two sessions.

[0043] Furthermore, using the business process identifier as a retrieval key, a search is performed in a pre-maintained business process relationship database to extract the set of preceding session identifiers and the set of following session identifiers associated with the process identifier. The set of preceding session identifiers records all sessions that should theoretically be executed before the current session in this business process, and the set of following session identifiers records all sessions that should theoretically be executed after the current session. Together, they define the standard execution order at the process level.

[0044] Furthermore, it checks whether the identifier of the previous target business sub-session is included in the set of preceding session identifiers, or whether the identifier of the next target business sub-session is included in the set of following session identifiers. If either condition is true, it indicates that the actual execution order of these two sessions is consistent with the standard order specified in the business process, and thus it is determined that there is a process order dependency between them.

[0045] This process involves aggregating all data dependencies and all process sequence dependencies into a single relation set. Duplicate dependencies are then deduplicated to ensure that any pair of sessions is recorded only once. The deduplicated result represents the complete logical dependencies between multiple business sub-sessions. This relation set includes both dependencies driven by data flow and dependencies driven by business process specifications, providing a unified dependency graph for subsequent session scheduling optimization and anomaly localization.

[0046] In this embodiment of the invention, by dynamically splitting business consultation sessions using computing resources, parallelization and load balancing of computing tasks can be achieved, significantly reducing single-point processing pressure and response latency. At the same time, identifying the logical dependencies between sub-sessions can effectively avoid data competition and deadlocks, which not only improves throughput and resource utilization in multi-round interaction scenarios, but also reduces session blocking and timeout risks, thereby enhancing the overall stability and scalability of large-scale concurrent business consultation systems.

[0047] S4. Construct a directed topology graph of the business sub-session based on the logical dependencies.

[0048] In this embodiment of the invention, constructing the directed topology graph of the business sub-session based on the logical dependencies includes: Construct an initially empty directed graph data structure, and extract the node set and directed edge set from the directed graph data structure; Each of the aforementioned business sub-sessions is added as a node instance to the node set; Each dependency record in the logical dependency relationship is treated as a directed edge and added to the set of directed edges; Based on the set of nodes and the set of directed edges, construct the initial directed graph of the business sub-session; Determine whether a cycle exists in the initial directed graph; If a loop exists, the loop is broken according to a preset loop resolution rule to obtain a directed topology graph.

[0049] In this embodiment of the invention, a directed graph data structure is created in memory initially, containing no nodes or edges. This structure consists of two disjoint sets: a node set to store all vertices in the graph, and a directed edge set to store all directed connections. Field access is performed on this empty structure to extract the node set and the directed edge set into two independent operable containers, preparing for the subsequent filling of business sub-session nodes and dependency edges.

[0050] Specifically, all business sub-sessions are traversed, and each business sub-session is encapsulated as a graph node instance. This node instance carries attribute information such as the session's identifier, execution status, and timestamp. Each node instance is then inserted into the node set one by one, so that each element in the node set corresponds one-to-one with a business sub-session, completing the initial filling of all vertices in the graph.

[0051] Specifically, each dependency record in the logical dependency relationship is traversed, and each record is instantiated as a directed edge. The directed edge carries the starting node identifier and the ending node identifier. The direction is determined by the order of the dependency relationship. Each directed edge is inserted into the directed edge set one by one, so that each edge in the edge set corresponds exactly to an identified dependency relationship, thus completing the initialization and filling of all connection relationships in the graph.

[0052] The process involves merging and assembling the already populated node set and the directed edge set. Each node instance in the node set is used as a vertex of the graph, and each directed edge in the directed edge set is used as a directed connection connecting the vertices, binding the two together into a complete initial directed graph. This directed graph graphically presents all business sub-sessions and their identified dependencies, but it has not yet undergone loop detection and may contain abnormal structures with circular dependencies.

[0053] Furthermore, a depth-first traversal or topological sorting test is performed on the initial directed graph. Starting from any node, the graph is traced along the direction of the directed edges to determine whether there is a closed path that starts from a certain node, passes through several directed edges, and then returns to the node itself. If at least one closed path is detected, it is determined that there is a cycle in the graph, which means that some business sub-sessions have formed a deadlock dependency of mutual waiting; if no closed path is detected, there is no cycle in the graph.

[0054] Furthermore, when a loop is detected, the system selects an applicable interruption strategy from a pre-defined loop resolution rule base. Based on the rule, it removes or reverses the direction of a directed edge within the loop, thereby breaking the loop's closure. After loop interruption, the system re-verifies that no closed paths exist in the graph, ultimately generating an acyclic directed topology graph. In this topology graph, the dependencies between any two nodes are unidirectionally reachable, providing a conflict-free dependency structure foundation for subsequent session scheduling, ordering, and parallel execution optimization.

[0055] In this embodiment of the invention, by constructing a directed topology graph, the logical dependencies between sub-sessions are explicitly transformed into a computable data structure, which facilitates automated scheduling and state management, and effectively improves the fault tolerance, execution efficiency and maintainability of large-scale business systems.

[0056] S5. Parse the session semantics contained in each business sub-session in the directed topology graph, and generate a query statement for each business sub-session based on the session semantics.

[0057] In this embodiment of the invention, parsing the session semantics contained in each service sub-session in the directed topology graph includes: Extract the session text content corresponding to each service sub-session in the directed topology graph; The conversation text content is preprocessed to obtain the target conversation text; Extract the semantic feature vector of the target conversation text; Calculate the target similarity between the semantic feature vector and each intent template vector in the preset business intent template library; The business intent tag corresponding to the business intent template with the highest target similarity that exceeds the preset similarity threshold is selected as the initial conversation semantics of the target conversation text; Extract key entity information from the target session text and use the key entity information as the semantic slot filling value for the initial session semantics; The initial session semantics are associated with and stored with the semantic slot filling value to obtain the session semantics contained in the business sub-session.

[0058] In this embodiment of the invention, each business sub-session node in the directed topology graph is traversed, and the complete session text content generated or received during execution is extracted from the attribute fields of that node. This text content records the complete interactive dialogue between the sub-session and the user or upstream and downstream sessions during actual operation, including the natural language expression input by the user and the response content returned by the system, which is the original corpus basis for subsequent semantic understanding.

[0059] Specifically, multiple rounds of preprocessing are performed on the extracted original conversation text, including removing meaningless system identifiers and special characters, standardizing the expression of synonyms, completing incomplete sentences caused by truncation, and filtering out greetings unrelated to business intent. The preprocessed text is reshaped into semantically focused and structurally clear target conversation text, eliminating noise interference in the original text and providing high-quality input corpus for accurate extraction of subsequent semantic features.

[0060] The target conversation text is input into a pre-trained semantic encoding model. This model performs layer-by-layer contextual understanding of the text, compressing and mapping the entire text into a fixed-dimensional semantic feature vector. This vector represents the core semantic meaning of the entire text as a point in a high-dimensional space, transforming the originally incalculable natural language into a numerical expression that can be mathematically calculated. This provides a unified computational carrier for subsequent similarity comparison with the intent template.

[0061] Furthermore, the semantic feature vector is compared with each intent template vector in the preset business intent template library one by one to calculate the similarity. This calculation quantifies the semantic fit between the target conversation text and each business intent by measuring the directional proximity of the two vectors in the semantic space. A similarity score is output for each pair of vectors. This score reflects the degree of semantic fit between the target conversation text and each preset business intent, providing a quantitative ranking basis for subsequent intent selection.

[0062] Furthermore, all similarity scores are sorted in descending order, and the business intent tag corresponding to the business intent template with the highest score is selected. This score is then compared to a preset similarity threshold. Only when the highest score exceeds this threshold is the matching result considered valid, and the corresponding business intent tag is determined as the initial conversational semantics of the target conversation text.

[0063] In this process, named entity recognition is performed on the target conversation text to locate and extract key entity information directly related to the business intent, such as product names, amounts, time points, and counterparty entities. This entity information is categorized according to a preset semantic slot framework, and each entity is filled into the slot corresponding to its semantic role, forming a set of structured slot filling values, which supplements the initial conversation semantics with specific business details.

[0064] This invention associates and binds initial session semantic tags with semantic slot fill values ​​according to a predetermined semantic framework. The intent tag serves as the semantic header, and the slot fill value serves as the semantic body, assembling them into a complete semantic structure containing the intent category and specific business parameters. This structure is persistently stored in the semantic attribute field of the business sub-session node, allowing any subsequent process to directly obtain its complete session semantics when accessing the session node without needing to recalculate the reasoning.

[0065] In this embodiment of the invention, initial session semantic tags and their corresponding semantic slot fill values ​​are extracted from the session semantics of each business sub-session. The intent tag is used as the core topic anchor of the query statement, and the slot fill values ​​are used as key constraint parameters to limit the topic. According to the preset natural language query template, the intent tag is embedded into the topic slot of the template, and the semantic slot fill values ​​are filled into the corresponding condition slots and constraint slots in the template according to their semantic roles, thus transforming the structured semantic framework into a complete query statement that conforms to the expression habits of natural language.

[0066] During the construction process, the slot filling values ​​are refined into language, transforming them from structured entity names into fluent natural language expressions. This ensures that the generated query statements accurately cover all the key points of conversational semantics while also conforming to the actual questioning methods used by users in real search scenarios.

[0067] In this embodiment of the invention, by identifying the conversational semantics of each business sub-session in the directed topology graph, the interaction intent of natural language can be transformed into a structured query statement, realizing an automated mapping from "understanding intent" to "direct retrieval / execution". This can effectively reduce semantic ambiguity and parsing overhead when subsequently calling the knowledge base or executing computational tasks, improve the accuracy and reusability of queries, and thus improve business processing efficiency and automation level.

[0068] S6. Using the query statement, search in multiple preset business knowledge document libraries to obtain multiple candidate document fragments corresponding to each business sub-session.

[0069] In this embodiment of the invention, the step of using the query statement to search in multiple preset business knowledge document libraries to obtain multiple candidate document fragments corresponding to each business sub-session includes: Obtain a list of library identifiers for multiple preset business knowledge document libraries. Based on the query semantic tags in the query statement and the list of library identifiers, filter out the target knowledge document libraries associated with the query semantic tags from the business knowledge document libraries to form a subset of document libraries to be retrieved. The query statement is expanded to generate an expanded query term set; For each knowledge document library in the subset of the document library to be retrieved, the extended query term set is matched with the document fragments in the knowledge document library to calculate the relevance score of each document fragment. Each document fragment is sorted from high to low according to the relevance score, and the top K document fragments in each knowledge document base are selected as the initial search results of the knowledge document base, where K is a preset positive integer; All the initial search results are merged and deduplicated to obtain multiple candidate document fragments corresponding to each of the business sub-sessions.

[0070] In this embodiment of the invention, a list of library identifiers for all preset business knowledge document libraries is obtained. Query semantic tags are parsed from the query statement. These tags represent the core business domain of the current query. The query semantic tags are compared one by one with the domain tags bound to each document library in the list of library identifiers. Document libraries whose domain tags are related to the query semantic tags are selected and formed into a smaller subset of the document libraries to be retrieved. This selection avoids indiscriminate retrieval in all document libraries, significantly reduces the scope of subsequent matching calculations, and ensures that the retrieval is only performed within the knowledge domain most relevant to the current query semantics.

[0071] This process involves semantic analysis of the original query to identify core business terms. Then, from a pre-defined business thesaurus and a network of hierarchical concepts, extended terms that are semantically similar to or hierarchically related to these core terms are retrieved. The original terms and all retrieved extended terms are combined into an extended query term set, covering various expressions that may involve the user's query intent. This extended processing allows the retrieval to move beyond the precise wording used by the user and recall document content expressed in different terms but with the same semantic meaning, effectively improving the retrieval recall rate.

[0072] Furthermore, an independent search and matching process is executed for each knowledge document repository within the subset of the document repository to be searched. Each extended term in the extended query term set is compared word-by-word with each document fragment in that repository for relevance. The frequency of each extended term in the document fragment, its positional weight, and its semantic fit with the fragment's topic are comprehensively considered. A comprehensive relevance score is then accumulated for each document fragment. This score quantifies the degree of matching between the document fragment and the user's query intent, providing a precise quantitative basis for subsequent ranking and filtering.

[0073] Furthermore, all document fragments within each knowledge document repository are sorted in descending order of their relevance scores to form a fragment priority sequence within that repository. Based on a preset positive integer truncation amount, the top-ranked document fragments are selected from the priority sequence of each document repository and used as the initial search results for that repository. This truncation operation ensures that each document repository contributes only its most relevant few fragments, thus preserving high-quality search content while avoiding diluting the overall search results quality by returning too many low-relevance fragments from a single document repository.

[0074] The process involves aggregating all initial search results from various knowledge document repositories into a single candidate set. Duplicate detection is then performed on document fragments with identical or highly similar content. When multiple repositories return fragments with substantially identical content, only the one with the highest relevance score is retained, and the rest are discarded. The final deduplicated set comprises multiple candidate document fragments for each business sub-session. This set covers highly relevant content from multiple knowledge domains without redundancy, providing concise and comprehensive knowledge materials for subsequent answer generation.

[0075] In this embodiment of the invention, by retrieving structured query statements in parallel across multiple business knowledge document bases, multiple candidate document fragments can be quickly retrieved for each business sub-session, significantly improving the coverage and efficiency of information acquisition. Simultaneously, retrieving candidate fragments separately for different sub-sessions facilitates subsequent accurate matching and sorting based on dependencies, avoiding redundant calculations and interfering information caused by global retrieval, and improving the accuracy and responsiveness of business decisions in complex business consultation scenarios.

[0076] S7. Extract the logical constraints of each business sub-session and the document summary of the corresponding candidate document fragment.

[0077] In this embodiment of the invention, extracting the logical constraints of each business sub-session and the document digest corresponding to the candidate document fragment includes: Identify the conditional logic structure in each of the business sub-sessions, and extract multiple conditional statement fragments of the business sub-session based on the conditional logic structure; The multiple conditional statement fragments are normalized to form standardized logical constraints. Feature extraction is performed on the candidate document fragments corresponding to each of the service sub-sessions to obtain candidate document features; The candidate document features are encoded into a document summary of the candidate document fragment.

[0078] In this embodiment of the invention, syntactic dependency analysis is performed on the conversation text of each business sub-session to identify the conditional logical connectors and logical relation markers contained therein, such as if, otherwise, and, or, thereby locating all logical structures in the session that have conditional judgment meanings. Using these logical connectors as segmentation points, each conditional logical structure is decomposed into independent conditional statement fragments. Each fragment contains a complete conditional judgment and its corresponding conclusion or action description, thereby extracting all scattered conditions throughout the session into independently processable structured units.

[0079] The process involves normalizing all extracted conditional statement fragments, converting semantically equivalent conditional statements with different expressions into a standardized conditional conclusion format. This eliminates ambiguity caused by differences in natural language expression, resolves vague references in conditional statements by replacing pronouns with explicit business entity names, and breaks down complex conditions into combinations of atomic conditions. Each normalized conditional statement fragment is then transformed into a standardized logical constraint with a unified format and precise semantics, which can be directly executed and judged by the subsequent inference engine.

[0080] Furthermore, feature extraction is performed on all candidate document fragments corresponding to each business sub-session. Key information representing the core content of each fragment is identified and extracted, including the fragment's keywords, involved business entities, core conclusive statements, and the most relevant contextual fragments to the query intent. This key information is organized according to preset feature dimensions to form candidate document features that accurately characterize the key points of the candidate document fragment. These features condense the core knowledge of the document fragment in a structured manner, providing a concise content carrier for subsequent summary generation.

[0081] Furthermore, candidate document features are taken as input and compressed and recombined using a sequence coding model, fusing scattered key information points into a semantically coherent and logically sound natural language text. This encoding process summarizes the core themes of the candidate document features, refines the expression of key entities and conclusions, eliminates redundant information, and ultimately outputs a document summary that fully covers the core points of the candidate document fragment. This summary serves as a condensed representation of the candidate document fragment, facilitating efficient matching and reasoning with the logical constraints of subsequent business sub-sessions.

[0082] In this embodiment of the invention, by extracting the logical constraints of each business sub-session, the contextual restrictions and input / output rules upon which the execution of the sub-session depends can be clearly defined, providing a calculable basis for subsequent semantic matching and result verification. At the same time, extracting document summaries of candidate document fragments can compress lengthy original documents into compact representations that retain core semantics, significantly reducing the storage overhead and computational complexity of subsequent processing steps. This enables efficient alignment and rapid filtering of constraints and knowledge fragments, thereby improving the accuracy and relevance of retrieval results and effectively reducing redundant computation and information noise during cross-sub-session processing.

[0083] S8. Determine whether the document summary satisfies the logical constraints.

[0084] In this embodiment of the invention, the set of all atomic propositions decomposed from the document digest of the candidate document fragment is used as the judgment basis, and all logical elements encapsulated by the standardized logical constraints of the business sub-session are used as the judgment criteria, and logical implication verification is performed one by one.

[0085] For each premise element in the logical constraint, a search is performed in the set of atomic propositions to determine if there is a semantically matching supporting proposition. If so, it is determined that the premise is covered by document knowledge. A deducibility test is then performed on the conclusion elements in the logical constraint to determine whether the supporting proposition can necessarily lead to the conclusion through logical reasoning.

[0086] The document summary is deemed to satisfy the logical constraints when all premises of all logical constraints are covered and all conclusions can be derived; if any premise is missing or any conclusion cannot be derived, it is deemed not to satisfy the logical constraints.

[0087] S9. If the document summary does not meet the logical constraints, return to step S6.

[0088] S10. When the document summary satisfies the logical constraint, it is determined that there is no logical conflict between the business sub-session and the corresponding candidate document fragment, and a logical chain between the candidate document fragment and the business sub-session is established.

[0089] In this embodiment of the invention, in a healthcare scenario, a patient asks, "Can amoxicillin and ibuprofen be taken together?" A candidate document summary is retrieved stating that the two can be used together. Comparing the summary with the session logic constraints reveals that the summary lacks a constraint regarding the risk of drug interactions, thus failing to meet the constraints. The process returns to step S6 for a second search. The second search yields another document summary that explicitly states there are gastrointestinal risks associated with the combined use of the two drugs, requiring intervals between doses. Step S10 determines that this summary meets all logic constraints, confirming no conflict. Step S11 constructs a logical chain based on this: the premise is that the patient is taking both drugs simultaneously; the supporting proposition is the interaction warning in the document; the inference is that intervals between doses are required; and the conclusion is that the user's current medication regimen carries risks.

[0090] In a fintech scenario, a user asks, "Does this product guarantee an 8% return on principal?" The retrieved candidate document summary states the expected return is 8%. Step S9 compares the summary and finds it lacks a statement denying the principal guarantee, thus failing to meet the constraint. The process returns to S6 for a second search. The second search yields a summary that clearly indicates the product is a non-principal-guaranteed floating-rate product. Step S10 determines the constraint is met, confirming no conflict. Step S11 constructs the logical chain: the premise is the user's question about principal protection; the supporting proposition is the non-principal-guaranteed nature described in the document; the inference is that there is no principal guarantee; the conclusion is that the product does not guarantee principal.

[0091] In this embodiment of the invention, establishing the logical chain between the candidate document fragment and the business sub-session includes: The logical constraints of each business sub-session are represented as logical elements to form logical units on the session side; The document summary of each candidate document fragment is decomposed into a set of atomic propositions; Calculate the semantic matching degree between each logical element in the logical unit and the set of atomic propositions; Based on the semantic matching degree, a first type of directed edge is established from the premise elements of the logical unit to the supporting propositions in the atomic proposition, and a second type of directed edge is established from the inference propositions in the atomic proposition to the conclusion elements of the logical unit. Connect the first type of directed edges with the second type of directed edges to form an initial logical chain; Redundancy pruning is performed on the initial logical chain to obtain the logical chain between the candidate document fragment and the business sub-session.

[0092] In this embodiment of the invention, the standardized logical constraints of each business sub-session are decomposed one by one, and each condition judgment, each causal relationship, and each exclusive constraint is mapped to an independent logical element. Each logical element is encapsulated into a propositional unit with a clear semantic role, such as a premise element, a conclusion element, a negation element, etc. All logical elements under the same business sub-session are aggregated and bound to form a complete session-side logical unit. This unit carries all the logical rules that the session must satisfy in business execution in a structured manner.

[0093] Semantic segmentation is performed on the document summary of each candidate document fragment, identifying each independent factual statement and inferential judgment in the summary. Each statement is broken down into the smallest indivisible semantic unit, i.e., an atomic proposition. Each atomic proposition expresses only a complete judgment or fact and no longer contains complex logic. All atomic propositions derived from the same document summary are compiled into an atomic proposition set, which lists all the knowledge assertions that the document fragment can provide in a flattened manner.

[0094] This process involves pairing each logical element in the conversational logical unit with each atomic proposition in the set of atomic propositions for the candidate document fragments, and performing semantic similarity calculations on each pair. This calculation measures the closeness of meaning between the two elements in the semantic space, determining whether the atomic proposition can support, refute, or supplement the logical element. A semantic matching score is output for each pair, quantifying the extent to which document knowledge covers the business logic, providing a precise matching basis for establishing subsequent logical connections.

[0095] Furthermore, based on the semantic matching score, pairings with a matching score exceeding a preset threshold are linked. When the logical element is a premise element and the atomic proposition is a supporting proposition, a first-type directed edge is established from the dependent premise of the logical unit to the supporting proposition in the atomic proposition, indicating that the document knowledge can be used as the basis for the business logic. When the atomic proposition is an inference proposition and the logical element is a conclusion element, a second-type directed edge is established from the inference proposition in the atomic proposition to the conclusion element, indicating that the document knowledge can deduce the business conclusion. The two types of directed edges respectively carry the two logical relationships of forward support and reverse inference.

[0096] Specifically, the first type of directed edge is used as the starting segment of the logical chain, and the second type of directed edge is used as the ending segment. The two types of directed edges are connected end-to-end according to the direction from business premises through document knowledge to business conclusions. The nodes between adjacent directed edges are checked to ensure that the end point of the previous edge is exactly the starting point of the next edge. In this way, the discrete directed edges are spliced ​​into a complete initial logical chain that starts from the dependency premises of the business sub-session, passes through the supporting propositions and inference propositions of candidate document fragments, and finally reaches the unverified conclusion of the business sub-session.

[0097] The process involves segment-by-segment review of the initial logical chain to identify redundant nodes and edges that do not substantially contribute to the logical deduction. Examples include supporting propositions completely covered by other edges and intermediate inferences that do not affect the final conclusion. These redundant elements are removed from the chain while ensuring that the pruned logical chain still fully covers all necessary reasoning paths from premise to conclusion. The pruned logical chain becomes the final logical chain between the candidate document fragment and the business sub-session. This chain is concise and complete, clearly demonstrating how document knowledge supports the complete reasoning process of business logic.

[0098] In this embodiment of the invention, when there is no logical conflict between the business sub-session and the candidate document fragment, a logical chain is constructed between the two, which can traceably bind the user's semantic intent with reliable knowledge sources, forming an interpretable path from question to answer, significantly improving the credibility and verifiability of the system output results, thereby enhancing the logical consistency, reasoning transparency and robustness of business intelligent decision-making.

[0099] S11. Generate the session response text for the business consultation session according to the logic chain.

[0100] In this embodiment of the invention, the final logic chain is used as the structured skeleton for response generation. Following the complete reasoning path from the dependency premises of the business sub-session through the supporting propositions of candidate document fragments to the verifiable conclusion of the business sub-session, the semantic key points of each step are extracted node by node. According to the expression habits of natural language, each reasoning step in the logic chain is transformed into a semantically coherent statement. Based on the causal and progressive relationships between the steps in the logic chain, these statements are organized sequentially into a complete response text.

[0101] During the generation process, the original expressions in the document summary are rewritten in a natural way to integrate them into the response context rather than being awkwardly pieced together. At the same time, the complete information of all key reasoning nodes in the logical chain is preserved, and the final output is a conversation response text that contains both document knowledge and fits the business consultation scenario.

[0102] For example, in a healthcare scenario, a patient asks: "Can amoxicillin and ibuprofen be taken together?" The retrieved candidate document abstracts indicate that combining these two drugs carries gastrointestinal risks and requires intervals between doses. The logical chain is: the premise is that the patient is taking both drugs simultaneously; the supporting proposition is the interaction warning in the document; the inference is that they need to be taken apart; and the conclusion is that the current regimen carries risks. Based on this, the response is: "Amoxicillin and ibuprofen can be used together, but it should be noted that taking them simultaneously may increase the risk of gastrointestinal discomfort. It is recommended to take them one to two hours apart. If you have a history of stomach problems, it is recommended to consult a doctor before taking the medication."

[0103] For example, in a fintech scenario, a user asks: "Does this wealth management product guarantee the principal?" The retrieved candidate document summary clearly states that the product is a non-principal-guaranteed, floating-rate product. The logical chain is as follows: the premise is that the user is asking whether the principal is guaranteed; the supporting proposition is the description of the non-principal-guaranteed nature in the document; the inference is that there is no guarantee of principal; the conclusion is that the product does not guarantee the principal. Based on this, the response is generated: This product does not guarantee the principal. It belongs to the non-principal-guaranteed, floating-rate wealth management product category. Returns will fluctuate with market changes. Past performance is not indicative of future results. We recommend that you carefully choose based on your own risk tolerance.

[0104] In this embodiment of the invention, generating session response text based on logical chains ensures that the output content is consistent with the verified knowledge path, avoiding the introduction of contradictory or irrelevant information. This method improves the accuracy and interpretability of responses, while supporting the automatic integration and orderly presentation of results from multiple sub-sessions, thereby effectively improving the decision-making quality and response reliability of business intelligence.

[0105] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0106] like Figure 3 The diagram shown is a functional block diagram of a business intelligent decision-making device provided in an embodiment of the present invention.

[0107] In this embodiment of the disclosure, a business intelligent decision-making device is provided, which corresponds one-to-one with the business intelligent decision-making method of the above embodiments. For example... Figure 3 As shown, this business intelligent decision-making device 100 can be installed in an electronic device. According to its functions, the business intelligent decision-making device 100 includes a session feature extraction module 101, a computing resource allocation module 102, a logical relationship recognition module 103, a directed graph construction module 104, a query statement generation module 105, a knowledge document retrieval module 106, a logical conflict judgment module 107, a logical conflict determination module 108, and a decision text generation module 109. Detailed descriptions of each functional module are as follows: The session feature extraction module 101 is used to acquire the business consultation session of the target user and extract the session features of the business consultation session; The computing power resource allocation module 102 is used to determine the computing power resource allocation strategy for the business consultation session based on the session characteristics, and allocate the computing power resources corresponding to the business consultation session according to the computing power resource allocation strategy. The logical relationship identification module 103 is used to split the business consultation session into multiple business sub-sessions based on the allocated computing power resources, and to identify the logical dependencies between the multiple business sub-sessions. The directed graph construction module 104 is used to construct a directed topology graph of the business sub-session based on the logical dependencies. The query statement generation module 105 is used to parse the session semantics contained in each business sub-session in the directed topology graph and generate a query statement for each business sub-session based on the session semantics. The knowledge document retrieval module 106 is used to retrieve multiple candidate document fragments corresponding to each business sub-session by using the query statement in multiple preset business knowledge document databases. The logical conflict judgment module 107 is used to extract the logical constraints of each business sub-session and the document summary of the corresponding candidate document fragment, and to determine whether the document summary satisfies the logical constraints. The logical conflict determination module 108 is used to determine that there is no logical conflict between the business sub-session and the corresponding candidate document fragment when the document digest satisfies the logical constraint conditions, and to establish a logical chain between the candidate document fragment and the business sub-session. The decision text generation module 109 is used to generate the session response text of the business consultation session according to the logic chain.

[0108] In one embodiment, when the logical relationship identification module 103 performs the task of identifying logical dependencies between multiple business sub-sessions, it is used to: Obtain the session timestamp corresponding to each of the service sub-sessions, and sort all the service sub-sessions in ascending order according to the session timestamps to form target service sub-sessions arranged in chronological order. Extract the input parameter set and output parameter set of the target service sub-session; The intersection of the output parameter set of the current target business sub-session and the input parameter set of the next target business sub-session is checked. If an intersection exists, it is determined that there is a data dependency relationship between the current target business sub-session and the next target business sub-session. Obtain the business process identifier corresponding to the current target business sub-session, and query the set of preceding session identifiers and the set of following session identifiers corresponding to the business process identifier from the preset business process relationship database; If the previous current business sub-session of the current target business sub-session belongs to the preceding session identifier set, or the next target business sub-session belongs to the following session identifier set, then it is determined that there is a process sequence dependency relationship. The data dependencies and process sequence dependencies are merged and deduplicated to generate logical dependencies between the multiple business sub-sessions.

[0109] In one embodiment, when the directed graph construction module 104 performs the construction of a directed topology graph of the service sub-session based on the logical dependencies, it is used to: Construct an initially empty directed graph data structure, and extract the node set and directed edge set from the directed graph data structure; Each of the aforementioned business sub-sessions is added as a node instance to the node set; Each dependency record in the logical dependency relationship is treated as a directed edge and added to the set of directed edges; Based on the set of nodes and the set of directed edges, construct the initial directed graph of the business sub-session; Determine whether a cycle exists in the initial directed graph; If a loop exists, the loop is broken according to a preset loop resolution rule to obtain a directed topology graph.

[0110] In one embodiment, the query statement generation module 105, when performing the parsing of the session semantics contained in each business sub-session in the directed topology graph, is used to: Extract the session text content corresponding to each service sub-session in the directed topology graph; The conversation text content is preprocessed to obtain the target conversation text; Extract the semantic feature vector of the target conversation text; Calculate the target similarity between the semantic feature vector and each intent template vector in the preset business intent template library; The business intent tag corresponding to the business intent template with the highest target similarity that exceeds the preset similarity threshold is selected as the initial conversation semantics of the target conversation text; Extract key entity information from the target session text and use the key entity information as the semantic slot filling value for the initial session semantics; The initial session semantics are associated with and stored with the semantic slot filling value to obtain the session semantics contained in the business sub-session.

[0111] In one embodiment, when the knowledge document retrieval module 106 executes the query statement to retrieve multiple candidate document fragments corresponding to each of the business sub-sessions, it is used to: Obtain a list of library identifiers for multiple preset business knowledge document libraries. Based on the query semantic tags in the query statement and the list of library identifiers, filter out the target knowledge document libraries associated with the query semantic tags from the business knowledge document libraries to form a subset of document libraries to be retrieved. The query statement is expanded to generate an expanded query term set; For each knowledge document library in the subset of the document library to be retrieved, the extended query term set is matched with the document fragments in the knowledge document library to calculate the relevance score of each document fragment. Each document fragment is sorted from high to low according to the relevance score, and the top K document fragments in each knowledge document base are selected as the initial search results of the knowledge document base, where K is a preset positive integer; All the initial search results are merged and deduplicated to obtain multiple candidate document fragments corresponding to each of the business sub-sessions.

[0112] In one embodiment, the logical conflict determination module 107, when executing the extraction of logical constraints for each of the business sub-sessions and the document summary corresponding to the candidate document fragments, is used to: Identify the conditional logic structure in each of the business sub-sessions, and extract multiple conditional statement fragments of the business sub-session based on the conditional logic structure; The multiple conditional statement fragments are normalized to form standardized logical constraints. Feature extraction is performed on the candidate document fragments corresponding to each of the service sub-sessions to obtain candidate document features; The candidate document features are encoded into a document summary of the candidate document fragment.

[0113] In one embodiment, when the logical conflict determination module 108 performs the process of establishing a logical chain between the candidate document fragment and the business sub-session, it is used to: The logical constraints of each business sub-session are represented as logical elements to form logical units on the session side; The document summary of each candidate document fragment is decomposed into a set of atomic propositions; Calculate the semantic matching degree between each logical element in the logical unit and the set of atomic propositions; Based on the semantic matching degree, a first type of directed edge is established from the premise elements of the logical unit to the supporting propositions in the atomic proposition, and a second type of directed edge is established from the inference propositions in the atomic proposition to the conclusion elements of the logical unit. Connect the first type of directed edges with the second type of directed edges to form an initial logical chain; Redundancy pruning is performed on the initial logical chain to obtain the logical chain between the candidate document fragment and the business sub-session.

[0114] In this invention, the specific limitations of a business intelligent decision-making device can be found in the above-described limitations of the business intelligent decision-making method, and will not be repeated here. Each module in the aforementioned business intelligent decision-making device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0115] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of the business intelligence decision-making method server-side.

[0116] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps of the business intelligence decision-making method on the client side.

[0117] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Obtain the business consultation sessions of the target user and extract the session features of the business consultation sessions; The computing power resource allocation strategy for the business consultation session is determined based on the session characteristics, and the computing power resources corresponding to the business consultation session are allocated according to the computing power resource allocation strategy. The business consultation session is divided into multiple business sub-sessions based on the allocated computing resources, and the logical dependencies between the multiple business sub-sessions are identified. Construct a directed topology graph of the business sub-sessions based on the logical dependencies; The session semantics contained in each business sub-session in the directed topology graph are parsed, and a query statement for each business sub-session is generated based on the session semantics. The query statement is used to search multiple preset business knowledge document libraries to obtain multiple candidate document fragments corresponding to each business sub-session; Extract the logical constraints of each business sub-session and the document summary of the corresponding candidate document fragment, and determine whether the document summary satisfies the logical constraints; When the document summary satisfies the logical constraints, it is determined that there is no logical conflict between the business sub-session and the corresponding candidate document fragment, and a logical chain is established between the candidate document fragment and the business sub-session. The session response text for the business consultation session is generated based on the logical chain.

[0118] In the several embodiments provided by this invention, it should be understood that the disclosed devices and apparatuses can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0119] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0120] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0121] In some embodiments of this example, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method described in the above embodiments.

[0122] The readable storage medium of the present invention stores a computer program, which, when executed by a processor of an electronic device, can perform the following: Obtain the business consultation sessions of the target user and extract the session features of the business consultation sessions; The computing power resource allocation strategy for the business consultation session is determined based on the session characteristics, and the computing power resources corresponding to the business consultation session are allocated according to the computing power resource allocation strategy. The business consultation session is divided into multiple business sub-sessions based on the allocated computing resources, and the logical dependencies between the multiple business sub-sessions are identified. Construct a directed topology graph of the business sub-sessions based on the logical dependencies; The session semantics contained in each business sub-session in the directed topology graph are parsed, and a query statement for each business sub-session is generated based on the session semantics. The query statement is used to search multiple preset business knowledge document libraries to obtain multiple candidate document fragments corresponding to each business sub-session; Extract the logical constraints of each business sub-session and the document summary of the corresponding candidate document fragment, and determine whether the document summary satisfies the logical constraints; When the document summary satisfies the logical constraints, it is determined that there is no logical conflict between the business sub-session and the corresponding candidate document fragment, and a logical chain is established between the candidate document fragment and the business sub-session. The session response text for the business consultation session is generated based on the logical chain.

[0123] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0124] Computer-readable storage media may also store at least one computer-executable program / instruction, such as computer-readable instructions. Computer-readable storage media include, but are not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Computer-readable storage media may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, a non-transitory computer-readable storage medium may be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions stored on the computer-readable storage medium, the various methods described above can be performed.

[0125] In addition, the computer device may include (but is not limited to) a data bus, an input / output (I / O) bus, a display, and input / output devices (e.g., keyboard, mouse, speakers, etc.).

[0126] In one embodiment, the at least one computer-executable instruction may also be compiled into or comprise a software product / computer program product, wherein one or more computer-executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described herein.

[0127] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Furthermore, any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory.

[0128] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0129] In the embodiments provided in this disclosure, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0130] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

[0131] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

Claims

1. A business intelligence decision-making method, characterized in that, The method includes: Obtain the business consultation sessions of the target user and extract the session features of the business consultation sessions; The computing power resource allocation strategy for the business consultation session is determined based on the session characteristics, and the computing power resources corresponding to the business consultation session are allocated according to the computing power resource allocation strategy. The business consultation session is divided into multiple business sub-sessions based on the allocated computing resources, and the logical dependencies between the multiple business sub-sessions are identified. Construct a directed topology graph of the business sub-sessions based on the logical dependencies; The session semantics contained in each business sub-session in the directed topology graph are parsed, and a query statement for each business sub-session is generated based on the session semantics. The query statement is used to search multiple preset business knowledge document libraries to obtain multiple candidate document fragments corresponding to each business sub-session; Extract the logical constraints of each business sub-session and the document summary of the corresponding candidate document fragment, and determine whether the document summary satisfies the logical constraints; When the document summary satisfies the logical constraints, it is determined that there is no logical conflict between the business sub-session and the corresponding candidate document fragment, and a logical chain is established between the candidate document fragment and the business sub-session. The session response text for the business consultation session is generated based on the logical chain.

2. The business intelligent decision-making method as described in claim 1, characterized in that, The construction of the directed topology graph of the business sub-session based on the logical dependencies includes: Construct an initially empty directed graph data structure, and extract the node set and directed edge set from the directed graph data structure; Each of the aforementioned business sub-sessions is added as a node instance to the node set; Each dependency record in the logical dependency relationship is treated as a directed edge and added to the set of directed edges; Based on the set of nodes and the set of directed edges, construct the initial directed graph of the business sub-session; Determine whether a cycle exists in the initial directed graph; If a loop exists, the loop is broken according to a preset loop resolution rule to obtain a directed topology graph.

3. The business intelligent decision-making method as described in claim 1, characterized in that, The parsing of the session semantics contained in each service sub-session in the directed topology graph includes: Extract the session text content corresponding to each service sub-session in the directed topology graph; The conversation text content is preprocessed to obtain the target conversation text; Extract the semantic feature vector of the target conversation text; Calculate the target similarity between the semantic feature vector and each intent template vector in the preset business intent template library; The business intent tag corresponding to the business intent template with the highest target similarity that exceeds the preset similarity threshold is selected as the initial conversation semantics of the target conversation text; Extract key entity information from the target session text and use the key entity information as the semantic slot filling value for the initial session semantics; The initial session semantics are associated with and stored with the semantic slot filling value to obtain the session semantics contained in the business sub-session.

4. The business intelligent decision-making method as described in claim 1, characterized in that, The step of using the query statement to search through multiple preset business knowledge document libraries to obtain multiple candidate document fragments corresponding to each business sub-session includes: Obtain a list of library identifiers for multiple preset business knowledge document libraries. Based on the query semantic tags in the query statement and the list of library identifiers, filter out the target knowledge document libraries associated with the query semantic tags from the business knowledge document libraries to form a subset of document libraries to be retrieved. The query statement is expanded to generate an expanded query term set; For each knowledge document library in the subset of the document library to be retrieved, the extended query term set is matched with the document fragments in the knowledge document library to calculate the relevance score of each document fragment. Each document fragment is sorted from high to low according to the relevance score, and the top K document fragments in each knowledge document base are selected as the initial search results of the knowledge document base, where K is a preset positive integer; All the initial search results are merged and deduplicated to obtain multiple candidate document fragments corresponding to each of the business sub-sessions.

5. The business intelligent decision-making method as described in claim 1, characterized in that, The extraction of the logical constraints for each of the business sub-sessions and the document digests corresponding to the candidate document fragments includes: Identify the conditional logic structure in each of the business sub-sessions, and extract multiple conditional statement fragments of the business sub-session based on the conditional logic structure; The multiple conditional statement fragments are normalized to form standardized logical constraints. Feature extraction is performed on the candidate document fragments corresponding to each of the service sub-sessions to obtain candidate document features; The candidate document features are encoded into a document summary of the candidate document fragment.

6. The business intelligent decision-making method as described in claim 1, characterized in that, The step of establishing the logical chain between the candidate document fragment and the business sub-session includes: The logical constraints of each business sub-session are represented as logical elements to form logical units on the session side; The document summary of each candidate document fragment is decomposed into a set of atomic propositions; Calculate the semantic matching degree between each logical element in the logical unit and the set of atomic propositions; Based on the semantic matching degree, a first type of directed edge is established from the premise elements of the logical unit to the supporting propositions in the atomic proposition, and a second type of directed edge is established from the inference propositions in the atomic proposition to the conclusion elements of the logical unit. Connect the first type of directed edges with the second type of directed edges to form an initial logical chain; Redundancy pruning is performed on the initial logical chain to obtain the logical chain between the candidate document fragment and the business sub-session.

7. The business intelligent decision-making method as described in claim 1, characterized in that, The identification of logical dependencies between the multiple service sub-sessions includes: Obtain the session timestamp corresponding to each of the service sub-sessions, and sort all the service sub-sessions in ascending order according to the session timestamps to form target service sub-sessions arranged in chronological order. Extract the input parameter set and output parameter set of the target service sub-session; The intersection of the output parameter set of the current target business sub-session and the input parameter set of the next target business sub-session is checked. If an intersection exists, it is determined that there is a data dependency relationship between the current target business sub-session and the next target business sub-session. Obtain the business process identifier corresponding to the current target business sub-session, and query the set of preceding session identifiers and the set of following session identifiers corresponding to the business process identifier from the preset business process relationship database; If the previous current business sub-session of the current target business sub-session belongs to the preceding session identifier set, or the next target business sub-session belongs to the following session identifier set, then it is determined that there is a process sequence dependency relationship. The data dependencies and process sequence dependencies are merged and deduplicated to generate logical dependencies between the multiple business sub-sessions.

8. A business intelligent decision-making device, characterized in that, The device includes: The session feature extraction module is used to acquire the business consultation session of the target user and extract the session features of the business consultation session; The computing power resource allocation module is used to determine the computing power resource allocation strategy for the business consultation session based on the session characteristics, and allocate the computing power resources corresponding to the business consultation session according to the computing power resource allocation strategy. The logical relationship identification module is used to split the business consultation session into multiple business sub-sessions based on the allocated computing resources, and to identify the logical dependencies between the multiple business sub-sessions. A directed graph construction module is used to construct a directed topology graph of the business sub-session based on the logical dependencies; The query statement generation module is used to parse the session semantics contained in each business sub-session in the directed topology graph, and generate a query statement for each business sub-session based on the session semantics; The knowledge document retrieval module is used to search in multiple preset business knowledge document databases using the query statement to obtain multiple candidate document fragments corresponding to each business sub-session. The logical conflict judgment module is used to extract the logical constraints of each business sub-session and the document summary of the corresponding candidate document fragment, and to determine whether the document summary satisfies the logical constraints. The logical conflict determination module is used to determine that there is no logical conflict between the business sub-session and the corresponding candidate document fragment when the document summary satisfies the logical constraint conditions, and to establish a logical chain between the candidate document fragment and the business sub-session. The decision text generation module is used to generate the session response text of the business consultation session according to the logic chain.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the business intelligence decision-making method as described in any one of claims 1 to 7.

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

Images