Search

CN-122022746-A - Enterprise management transaction cooperative processing method and system integrating AI intelligent cooperation and decision

CN122022746ACN 122022746 ACN122022746 ACN 122022746ACN-122022746-A

Abstract

The invention discloses an enterprise management transaction collaborative processing method and system integrating AI intelligent collaborative and decision-making, wherein the method comprises the following steps of S1, data base construction and multi-source data fusion, S2, AI model system customized construction, S3, management transaction access and intention recognition, S4, AI collaborative decision-making and man-machine collaborative processing, S5, decision-making execution monitoring and effect evaluation, and S6, closed loop optimization iteration. According to the invention, through a mixed model system of a general model, exclusive fine tuning and a scene plug-in and an exclusive knowledge base, AI decisions are attached to actual business logic and adapt to differential demands, and meanwhile, hierarchical man-machine cooperation, decision reasoning visualization, full-link closed-loop optimization and monitoring and early warning are relied on, so that the practicability and the floor property of the cooperative decisions are greatly improved, the labor burden is reduced, and the core problems of insufficient fusion, difficult decision floor and the like in the prior art are effectively solved.

Inventors

  • ZHANG SHUYING

Assignees

  • 烟台松鼠网络科技有限公司

Dates

Publication Date
20260512
Application Date
20260203

Claims (8)

  1. 1. The enterprise management transaction cooperative processing method integrating AI intelligent cooperation and decision-making is characterized by comprising the following steps: The method comprises the steps of S1, constructing a data base and multi-source data fusion, namely, interfacing an enterprise internal and external business system through a data acquisition unit, acquiring structured, semi-structured and unstructured multi-source data in real time, cleaning, standardizing and associated modeling the acquired data through a data treatment unit, and constructing a business data associated network; S2, customizing and constructing an AI model system, namely, taking a general large model as a base, utilizing an enterprise exclusive knowledge base and historical service data through a fine tuning and optimizing unit, carrying out fine tuning on the base model by combining with an RAG technology, and injecting enterprise service characteristics to generate an exclusive AI model; S3, managing transaction access and intention recognition, wherein an enterprise user accesses a management transaction through a scene application module, and the intention recognition unit analyzes transaction appeal, service types, priorities and associated service flows based on an NLP algorithm, precisely matches corresponding scene algorithm plug-ins and exclusive AI models, and ensures that an AI decision direction is consistent with a service target; The AI collaborative decision-making and man-machine collaborative processing comprises the steps that a task scheduling unit allocates AI resources according to transaction information, invokes corresponding models and plugins, combines an exclusive knowledge base and real-time service data, generates decision suggestions and synchronously outputs reasoning processes and data supporting basis; S5, decision execution monitoring and effect evaluation, wherein an execution monitoring unit tracks the decision execution progress and the node state in real time, automatically triggers early warning and pushes the early warning to a corresponding responsible person when the execution deviates from a target or is abnormal through visual billboard synchronous data; And S6, performing closed-loop optimization iteration, namely feeding back execution data, an evaluation result and manual adjustment opinion to an AI model system, optimizing exclusive model parameters and a scenerization plug-in algorithm, updating an exclusive knowledge base, and repeating the steps S1-S5 to realize continuous iteration of the model, the flow and the decision capability and continuously improving the fusion depth and the decision landing effect of the AI and the service.
  2. 2. The method for collaborative processing of enterprise management transactions incorporating AI intelligent collaboration and decision-making according to claim 1, wherein the specific logic steps of S1 are as follows: S101, synchronously accessing an internal and external business system of an enterprise, an internal docking ERP, OA, CRM system and an external docking supply chain collaboration platform and an industry database by adopting a multi-protocol docking mode through a data acquisition unit, adopting a differential acquisition strategy for different types of data, wherein structured data are directly acquired through the database, semi-structured data are acquired through an API (application program interface), unstructured data are acquired through a crawler technology and a document analysis tool, so that real-time capturing of the whole data is realized, and the acquisition frequency can be dynamically configured according to the importance of the data; S102, cleaning the data acquired in the S101 by a data management unit, removing abnormal values, missing values and repeated data, performing standardized conversion, unifying data formats and units, and uniformly converting the data with different sources and different formats into a JSON-LD format after standardized processing to ensure data interoperability; S103, constructing a business data association network based on a knowledge graph technology, mining semantic association and logic relation among different module data, calculating association strength among data entities by adopting a cosine similarity algorithm, and determining entity connection relation, wherein the used formula is as follows: ; Wherein the method comprises the steps of The feature vectors of the two data entities respectively, For the eigenvalue of the corresponding dimension of the vector, n is the eigenvector, when cosine similarity When the method is used, the existence of strong association between two entities is judged, association edges are constructed, association types are marked, and finally a visual business data association network is formed; s104, converting unstructured knowledge of enterprise industry rules, business processes and historical decision cases into structured knowledge items after OCR recognition and semantic word segmentation, realizing knowledge vectorization storage through a vector database, and converting the knowledge items into vector representations with fixed dimensions by adopting a Sentence-BERT model to ensure efficient knowledge retrieval, wherein the used formula is as follows: ; Wherein the method comprises the steps of As a vector of the kth knowledge, A function is generated for the Sentence-BERT vector, For the text content of the kth piece of knowledge, Training parameters for the model; And after vector storage, millisecond-level retrieval based on semantic similarity is supported, business knowledge support is provided for subsequent AI model fine adjustment and decision-making reasoning, and the whole construction of the data base is completed.
  3. 3. The method for collaborative processing of enterprise management transactions incorporating AI intelligent collaboration and decision-making according to claim 1, wherein the specific logic steps of S2 are as follows: S201, selecting a lightweight general large model as a basic capability base, deploying the model in an enterprise private cloud environment, reducing hardware resource occupation by a model quantization technology, reserving core AI capabilities of semantic understanding, data analysis and content generation, adapting to basic analysis requirements of multiple types of management transactions, and guaranteeing that model call delay is less than or equal to 200ms; S202, based on RAG technology and supervised learning, fine tuning a basic model by utilizing an enterprise exclusive knowledge base and historical service data, injecting service rules and personalized requirements, optimizing model parameters by adopting a cross entropy loss function, and ensuring that the fine-tuned model output is attached to enterprise service logic, wherein the used formula is as follows: ; Where N is the number of fine-tuned samples, C is the number of output classes, For the true label of the ith sample in class c, Predicting the probability that the ith sample belongs to the c class for the model, and iteratively optimizing the loss function through a gradient descent algorithm until the loss value converges to generate an enterprise exclusive AI model; S203, developing a proprietary algorithm plug-in for high-frequency management scenes of personnel, finance, projects and supply chains, integrating the proprietary algorithm plug-in into a model system by adopting a hot plug architecture, and linking with a proprietary AI model through a standardized interface to ensure the accuracy of scene decision.
  4. 4. The method for collaborative processing of enterprise management transactions incorporating AI intelligent collaboration and decision-making according to claim 1, wherein the specific logic steps of S3 are as follows: S301, enterprise users access management transactions through a scene application module, an API interface, enterprise WeChat and spike integrated entrance multiple channels, and a system automatically captures transaction submission information, including transaction content, submitters, affiliated departments and emergency degree basic attributes, and generates unique transaction IDs for whole-flow tracking; s302, the intention recognition unit performs semantic analysis on transaction contents based on an NLP algorithm, extracts key information of core appeal, service types and priorities, and converts model output into confidence degrees of various intentions through a softmax function, wherein the used formulas are as follows: ; Wherein the method comprises the steps of For confidence, x is the transaction text feature vector, For the purpose of the j-th class, For scoring the j-th type of intentions by the model, M is the total number of the types of intentions, selecting the intentions with the highest confidence as a final analysis result, and triggering manual auxiliary confirmation when the confidence is lower than a threshold; S303, based on the analyzed service type, intention and priority, the system automatically matches the corresponding scenerizing algorithm plug-in and the exclusive AI model to generate a transaction processing capacity association table, for example, a project progress optimizing transaction matching project progress predicting plug-in and the exclusive AI model, and the high priority transaction preferentially allocates AI resources to ensure processing timeliness.
  5. 5. The method for collaborative processing of enterprise management transactions incorporating AI intelligent collaboration and decision-making according to claim 1, wherein the specific logic steps of S4 are as follows: S401, a task scheduling unit adopts SERVERLESS architecture, automatically allocates CPU, GPU and memory resources according to transaction types, priorities and resource occupation demands, schedules corresponding models and plug-ins to process multiple transactions in parallel, avoids resource conflict through a resource scheduling algorithm, ensures that the high-priority transaction resource occupation rate is more than or equal to 60%, and the low-priority transaction does not occupy core resources; S402, calling a matched exclusive AI model and a scene plugin, carrying out multidimensional analysis and generating decision suggestions by combining an exclusive knowledge base and real-time service data, and simultaneously outputting a decision reasoning process comprising data sources, algorithm logic and service rule bases, ensuring that decisions can be interpreted, determining each analysis dimension weight by adopting a hierarchical analysis method, and improving the scientificity of the decisions, wherein the used formulas are as follows: ; Wherein the method comprises the steps of For the weight of the i-th analysis dimension, The importance ratio of the dimension i to the dimension j is calculated, and n is the number of the analysis dimensions; S403, dividing processing authorities according to transaction values and complexity, wherein low-value repeated transactions are automatically executed by an AI and an execution log is generated, medium-value transactions are executed after manual review confirmation by the AI in a decision suggestion and reasoning process, the review opinions are recorded in real time, high-value complex transactions are decided by manual leading, and the AI provides data support, multi-scheme comparison and risk early warning to ensure that decisions meet actual demands of enterprises.
  6. 6. The method for collaborative processing of enterprise management transactions incorporating AI intelligent collaboration and decision-making according to claim 1, wherein the specific logic steps of S5 are as follows: s501, an execution monitoring unit tracks decision execution progress, node completion conditions and resource consumption data through a buried point technology, displays the decision execution progress, node completion conditions and resource consumption data in real time through a visual BI (business information board) and synchronizes related business system data, so that consistency of the execution data and the business data is ensured, the monitoring frequency is 1 minute/time, and full-visual execution state is realized; S502, setting an execution deviation threshold, automatically triggering early warning when the execution progress is more than or equal to 20% and the resource consumption exceeds the budget by more than or equal to 15%, pushing the early warning to a corresponding responsible person through a short message and an enterprise WeChat, simultaneously providing initial analysis and intervention suggestion of an abnormal cause, supporting quick treatment, and adopting the formula for calculating the execution deviation as follows: ; Wherein the method comprises the steps of In order to execute the deviation rate, the actual value is the current execution data, the planned value is the preset target value, when Triggering corresponding level early warning when the corresponding dimension threshold value is exceeded; And S503, after the decision execution is completed, the effect evaluation unit automatically performs statistical analysis and generates an evaluation report based on a preset index system, and quantifies the AI collaborative decision value to provide data support for subsequent optimization.
  7. 7. The method for collaborative processing of enterprise management transactions incorporating AI intelligent collaboration and decision-making according to claim 1, wherein the specific logic steps of S6 are as follows: S601, collecting decision execution data, effect evaluation results, manual review opinions and abnormal treatment record multi-source feedback data, storing the multi-source feedback data into a data base according to categories, and taking the multi-source feedback data as an optimization basis; S602, a model iteration unit periodically optimizes exclusive AI model parameters based on feedback data, updates a scenerized plug-in algorithm, and re-fine-adjusts a model by adopting a cross entropy loss function to improve decision accuracy; And S603, automatically or manually adjusting the collaborative process nodes, the authority allocation, the resource scheduling strategy and the early warning threshold value based on the effect evaluation result, adapting to the enterprise organization architecture, the service scale and the market environment change, and repeating the steps S1-S5 to form a full-link closed loop of data-model-decision-execution-feedback-optimization, so as to continuously promote the collaborative processing effect.
  8. 8. An enterprise management transaction cooperative processing system integrating AI intelligent cooperation and decision making is used for realizing the method of any of claims 1-7, and is characterized by comprising a data base module, an AI model system module, an AI intelligent force module, a scene application module and a monitoring feedback and closed loop optimization module; The data base module is used for realizing fusion management and exclusive knowledge storage of multi-source service data, breaking a data island, providing high-quality data support for method execution, and comprises a data acquisition unit, a data management unit and an exclusive knowledge base unit, wherein the data acquisition unit is used for acquiring internal and external multi-source data of an enterprise in real time through RESTful API interfaces, direct connection of the database and a third party system integration mode, and comprises structured data, semi-structured data and unstructured data, and simultaneously supports seamless connection with the existing CRM, ERP, OA system of the enterprise to ensure comprehensiveness and instantaneity of data acquisition; The AI model system module is used for constructing a mixed model system of a general large model, an enterprise exclusive fine tuning model and a scene algorithm plugin, realizing the deep coupling of AI and business, improving the decision accuracy, and comprises a basic model unit, a fine tuning optimization unit and a scene plugin unit, wherein the basic model unit adopts the light general large model as a basic capability base, provides core AI capability for semantic understanding, data analysis and content generation, adapts to basic analysis requirements of multiple types of management businesses, finely tunes the basic model through RAG technology and supervision learning based on enterprise exclusive knowledge base and historical business data, injects enterprise business rules, flow characteristics and personalized requirements, generates an enterprise exclusive AI model, ensures the output of the model to be attached to actual business logic of an enterprise, reduces the phenomenon of illusion; The AI intelligent force module is used as a bridge between a model system and an application layer and is used for realizing business encapsulation, collaborative scheduling and man-machine collaborative processing of AI capacity, and comprises an intention recognition unit, a task scheduling unit and a man-machine collaborative unit, wherein the intention recognition unit is used for accurately recognizing core appeal, business types and priorities of management transactions based on an NLP algorithm, associating corresponding business processes and model plug-ins and ensuring that an AI decision direction is consistent with a business target; the task scheduling unit adopts SERVERLESS architecture to construct intelligent scheduling engine, automatically allocates AI model resources according to transaction type, priority and resource occupation condition, schedules corresponding scene plugins to realize multi-transaction parallel processing, supports dynamic adjustment of flow nodes, adapts organization architecture and service priority change, and is used for constructing hierarchical man-machine collaboration mechanism, dividing processing authorities according to transaction value and complexity, wherein low-value repeated transactions are automatically processed and executed by AI, medium-value transactions are output by AI to carry out decision suggestion with reasoning process, and are executed after manual rechecking confirmation; the scene application module is used for packaging the AI collaborative decision-making capability into scene application and adapting to the collaborative requirements of various management transactions of enterprises; The monitoring feedback and closed-loop optimization module is used for realizing monitoring, evaluation and iteration of the decision-making full link, and improving the decision-making floor property and the continuous optimization capability.

Description

Enterprise management transaction cooperative processing method and system integrating AI intelligent cooperation and decision Technical Field The invention relates to the technical field of enterprise management transaction collaborative processing, in particular to an enterprise management transaction collaborative processing method and system integrating AI intelligent collaboration and decision. Background Along with the rapid development of artificial intelligence technology, more and more enterprises begin to integrate the AI technology into the collaborative processing of management matters so as to improve decision-making efficiency and optimize operation flow, and the conventional collaborative processing method of the enterprise management matters integrating the AI intelligent collaboration and decision-making is used for analyzing management data through a general AI model and outputting decision-making suggestions so as to assist the enterprises to complete the collaboration of multi-module matters such as personnel, finance, projects, supply chains and the like; the method has the defects that 1, AI and business fusion depth is insufficient, a general AI model adopted by the prior method is not deeply coupled with business rules, management processes, organization structures and personalized requirements of industries where enterprises are located, macroscopic conclusions can be output only based on the general algorithm, logic and requirements of specific business scenes of the enterprises cannot be accurately adapted, the AI capability and the business processes are disjointed, collaborative processing of complex business scenes is difficult to support, 2, the practicability and the floor property of collaborative decisions are poor, the decision suggestions output by the prior method lack of matching with the execution processes, authority systems and resource allocation of the enterprises, and the decision suggestions can fall to the floor only by a large number of manual secondary adjustments, even due to lack of interpretability and execution monitoring, the decisions cannot be effectively converted into actual actions, not only are collaborative efficiency not improved, but also artificial burden is increased, and the method and the system for collaborative processing of the enterprise management business which is integrated with AI intelligent collaboration and decision are provided by the method and the system. Disclosure of Invention Based on the technical problems in the background technology, the invention provides an enterprise management transaction collaborative processing method and system integrating AI intelligent collaboration and decision. The enterprise management transaction collaborative processing method integrating AI intelligent collaboration and decision-making provided by the invention comprises the following steps: The method comprises the steps of S1, constructing a data base and multi-source data fusion, namely, interfacing an enterprise internal and external business system through a data acquisition unit, acquiring structured, semi-structured and unstructured multi-source data in real time, cleaning, standardizing and associated modeling the acquired data through a data treatment unit, and constructing a business data associated network; S2, customizing and constructing an AI model system, namely, taking a general large model as a base, utilizing an enterprise exclusive knowledge base and historical service data through a fine tuning and optimizing unit, carrying out fine tuning on the base model by combining with an RAG technology, and injecting enterprise service characteristics to generate an exclusive AI model; S3, managing transaction access and intention recognition, wherein an enterprise user accesses a management transaction through a scene application module, and the intention recognition unit analyzes transaction appeal, service types, priorities and associated service flows based on an NLP algorithm, precisely matches corresponding scene algorithm plug-ins and exclusive AI models, and ensures that an AI decision direction is consistent with a service target; The AI collaborative decision-making and man-machine collaborative processing comprises the steps that a task scheduling unit allocates AI resources according to transaction information, invokes corresponding models and plugins, combines an exclusive knowledge base and real-time service data, generates decision suggestions and synchronously outputs reasoning processes and data supporting basis; S5, decision execution monitoring and effect evaluation, wherein an execution monitoring unit tracks the decision execution progress and the node state in real time, automatically triggers early warning and pushes the early warning to a corresponding responsible person when the execution deviates from a target or is abnormal through visual billboard synchronous data; And S6, performing closed-loop optimiza