Business system building method and system

Abstract

The application discloses a business system construction method and system, and relates to the field of business process management.The scheme identifies related entities such as roles, demands and strategic targets through a bidirectional non-discriminative coding model, performs semantic coding and element classification by using an attention polarity model, and generates a stakeholder map in combination with a rule engine; the demand weight is calculated by means of a graph neural network, a weighted strategic vector is generated by embedding a semantic vector, process nodes are extracted by an AI model and a process element table is generated in association; a final process diagram is generated through state coding and a near-end strategy optimization learning network, a programming environment is determined based on demands, a code segment is matched to generate a business system, and the whole-process automation from strategic analysis to business system generation is realized, so that the business system development efficiency is improved and the technical threshold is reduced.

Description

Technical Field

[0001] This invention relates to the field of business process management, specifically to a method and system for building a business system. Background Technology

[0002] In enterprise digital transformation, business system development needs to accurately align with strategic goals and be efficiently implemented. Traditional methods rely on manually breaking down strategic elements, manually designing business processes, and writing code, leading to problems such as distorted strategic communication, inefficient process design, and long system implementation cycles.

[0003] A Chinese patent application with publication number CN119886777A discloses a process link evaluation method based on graph neural networks. The method analyzes the bottlenecks and redundancies of process nodes through graph neural networks to evaluate the rationality of process design.

[0004] Existing technologies only focus on the static evaluation of process steps, without addressing the automated decomposition and semantic transmission of strategic objectives, and without solving the problem of intelligent mapping from process nodes to code segments. Summary of the Invention

[0005] This invention addresses the shortcomings of existing technologies by proposing a business system construction method and system to achieve intelligent business system construction based on automated decomposition and semantic transmission analysis of strategic objectives.

[0006] The technical solution to achieve the purpose of this invention is as follows:

[0007] The method for building a business system includes the following steps:

[0008] A bidirectional non-discriminatory coding model is used to identify relevant entities in the document. An attention polarity model is used to semantically encode strategic objectives and aggregate and classify them into internal or external elements based on attention weights. A list of strategic elements is established by combining polarity analysis. A rule engine is used to connect roles and needs to strategic objectives to generate a stakeholder graph. Relevant entities include roles, needs, and strategic objectives.

[0009] The demand weight of each demand in the stakeholder graph is calculated by using a graph neural network. The demand semantic vector of the demand supporting each strategic goal is embedded into the strategic semantic vector of the strategic goal in the strategic element list according to the corresponding demand weight, and a weighted strategic vector is generated. Based on the stakeholder graph constraining the GPT4-Turbo model, the process nodes of the business documents are extracted and associated to generate a process element table. The strategic semantic vector and the demand semantic vector are generated by the strategic goal and demand input preset language model.

[0010] Each process element in the process element table is state-coded, a near-end strategy optimization learning network is adopted, action space filtering is set based on stakeholder graph, and process elements are connected based on strategic transmission reward guidance to generate the final-level process diagram. The state coding is implemented based on the demand weight table and the weighted strategic vector set.

[0011] Based on customer needs, the programming language and compilation environment are determined. The process semantic vectors of the process nodes in the final-level flowchart are matched with the code semantic vectors of the code segments in the code library using a logical verification function. The matching code segment is then filled into each process node to generate the business system. The logical verification function is equal to the norm of the intersection of the process semantic vector and the code semantic vector divided by the norm of the code semantic vector.

[0012] Furthermore, the bidirectional non-discriminating coding model identifies strategic objectives and related entities in the underlying documents, including the following steps:

[0013] Use a tokenizer to split each document into short sentences and then into sub-words, adding start tags and split tags at the beginning of the document and the end of the short sentences respectively;

[0014] Insert filler words before the segmentation markers of all clauses to make the number of words in each clause the same;

[0015] Based on the vocabulary, sub-words are mapped to number vectors. Based on the order of sub-words in the document, position vectors are generated and superimposed with the sub-word number vectors to generate the sub-word embedding vectors.

[0016] The embedding vectors of each word in the short sentence are concatenated and input into the Transformer encoder. After processing through 12 layers of continuous multi-head self-attention mechanism, residual connection, layer normalization and feedforward neural network, a context representation vector is generated.

[0017] The context representation of each word in the context representation vector is transformed into a probability distribution of each word through linear modulation and Softmax function by a linear classifier. The BIN label with the highest probability is selected and assigned to the word. The BIN label adopts the prefix + type format. Entity A has two BIN labels, BA and IA, which represent the start word and continuation word of entity A, respectively. Non-entity only has label NN.

[0018] Based on the principle of continuous merging, the sub-words of the beginning words and continuation words of short sentences that are consecutive and whose BIN tags correspond to the same entity are merged to identify related entities in the document.

[0019] Furthermore, the principle of successive mergers includes the following steps:

[0020] Perform a start test on the first word in the short sentence. If the label of the first word is NN, skip it and perform a new start test on the second word. If the label of the first word represents the start word of the entity, perform a continuation test on the second word.

[0021] If the label of the second sub-word represents a continuation word of the entity and is the same as the entity of the first sub-word, then the two sub-words are merged and a continuation check is performed on the third sub-word;

[0022] If the label of the second sub-word is NN, then the first sub-word is treated as an entity and a new starting point test is performed on the third sub-word;

[0023] If the label of the second sub-word represents the start word of the entity, the first sub-word is treated as an entity alone and the continuation test of the second sub-word is treated as a new start test, and a new continuation test is performed on the third sub-word;

[0024] Continue until all words in the short sentence have been checked.

[0025] Furthermore, the attention polarity model establishes a list of strategic elements, including the following steps:

[0026] Strategic objectives are transformed into strategic semantic vectors through a pre-defined language model;

[0027] The product of the strategic semantic vector and the internal and external query vectors is transformed into internal and external weights through a self-attention mechanism.

[0028] The product of the concatenated strategic semantic vector and the internal and external weights is transformed into the internal and external probability distributions of the strategic objective through linear modulation and the Softmax function;

[0029] By convolving strategic semantic vectors with three convolutional kernels of different window sizes and using PReLU function and max pooling for local feature enhancement, three semantic features of different scales are generated and concatenated into the fully connected layer. The polarity probability distribution is output based on the Softmax function.

[0030] By combining internal and external probability distributions with polarity probability distributions, a SWOT element probability distribution of strategic objectives is constructed. An edge is established between the SWOT element with the highest probability and the strategic objective, and the value of the edge is equal to the highest probability in the SWOT element probability distribution. Strategic objectives are grouped according to each SWOT element, and a list of strategic elements is reconstructed.

[0031] Furthermore, a rules engine is used to match and connect the roles and needs of relevant entities with strategic objectives, generating a stakeholder graph, including the following steps:

[0032] Establish entity chain rules, establish the attribution relationship of role demand chain and the support relationship of demand strategic goal chain, and establish two-way correspondence rules, that is, roles and demands, and demands and strategic goals are all in many-to-many form;

[0033] Each relevant entity is transformed into a bag-of-words feature vector and a corresponding semantic vector using a TF-IDF converter and a pre-defined language model, and then concatenated into an entity description vector.

[0034] Each role and need, as well as each need and strategic goal, is arranged and combined separately to construct role-need pairs and need-strategy goal pairs. Cosine similarity is calculated for each pair. For roles and needs with cosine similarity greater than or equal to the similarity threshold, belonging edges and supporting edges are established respectively to generate a stakeholder graph. The belonging edge value and the supporting edge value are equal to the corresponding cosine similarity.

[0035] Specifically, the graph neural network counts the number of roles belonging to the edge of each demand in the stakeholder graph as the out-degree. It divides the out-degree of each demand by the out-degree of all demands and the demand weight that generates each demand. It encodes each strategic objective in the strategic element list into a strategic semantic vector through a preset language model. It retrieves the demands that can support each strategic objective from the stakeholder graph and converts them into demand semantic vectors through the preset language model. It performs a weighted sum based on the corresponding demand weights and superimposes it with the strategic semantic vector of each strategic objective to generate a weighted strategic vector for each strategic objective.

[0036] Specifically, in the GPT4-Turbo model, process node role matching rules are pre-established. All relevant role requirement chains are introduced into each process node and role matching pair and integrated to build a template rule base. The natural language processing capability of the GPT4-Turbo model is used to identify process nodes in business documents and match them with the template rule base to obtain the role requirement chains that match the process nodes. Through a preset language model, the text description of the process nodes and the requirements in each role requirement chain are transformed into process semantic vectors and requirement semantic vectors and the cosine similarity is calculated. The role requirement chain corresponding to the requirement semantic vector with the largest cosine similarity is organized with the process node as process elements. The process elements of all process nodes are combined to build a process element table.

[0037] Furthermore, each process element in the process element table is converted into a corresponding process state vector through state coding, including the following steps:

[0038] The text descriptions and roles of process nodes in each process element are transformed into process semantic vectors and role semantic vectors using a preset language model;

[0039] Based on the stakeholder map and demand weight table, the demand and corresponding demand weight related to the roles in the process elements are obtained. The role-related demand is generated into a demand semantic vector through a preset language model. The demand is weighted and summed based on the corresponding demand weights and the role semantic vector of the role is superimposed to generate a weighted role vector.

[0040] Retrieve the strategic objectives that the needs of process elements can support from the stakeholder map and obtain the corresponding weighted strategic vectors. Based on the weighted strategic vectors obtained by weighted aggregation of the supporting boundary values ​​between the corresponding needs and strategic objectives, generate the strategic association demand vector for each process element.

[0041] By combining the process semantic vector, weighted role vector, and strategic related requirement vector of each process element, a process status vector for each process element is generated.

[0042] Furthermore, a near-end strategy is employed to optimize the learning network and generate a final-level flowchart that meets the requirements of the business documentation, including the following steps:

[0043] The business strategic objectives in the business documents are converted into business strategic semantic vectors through a pre-set language model. The strategic transmission reward is designed to be equal to the cosine similarity between the current process state vector and the business strategic semantic vector multiplied by the transmission coefficient. The transmission coefficient is equal to the sum of the edge values ​​of the belonging edges and / or supporting edges that the role corresponding to the current process state vector passes through in the stakeholder graph to reach the business strategic objective.

[0044] Based on the near-end strategy optimization algorithm, action space filtering is set based on the graph structure of the stakeholder graph to generate a strategy network. The strategy network is used to select the next process element. Action space filtering refers to eliminating process elements with roles or needs that do not have direct or indirect connections with the roles and needs in the current process element when selecting the next process element.

[0045] Select a process state vector of a process element as the starting point, and the policy network selects the next process element to connect based on the near-end policy optimization algorithm and the strategic transmission reward. Repeat this step until it stops.

[0046] If the stop is due to having traversed all process elements, then a final-level flowchart is generated;

[0047] If the stoppage is due to spatial filtering resulting in no selectable process elements, discard the flowchart under construction and reselect a new process element as the starting point.

[0048] Obtain all final-level flowcharts that traverse all process elements, and output the final-level flowchart that has the highest cumulative strategic transmission reward.

[0049] A business system building system is used to execute the business system building method, including a graph construction module, a process extraction module, a process generation module, and an integration building module;

[0050] The graph construction module uses a bidirectional non-discriminatory coding model to identify relevant entities in the document, uses an attention polarity model to semantically encode strategic objectives, and aggregates and categorizes them into internal or external elements based on attention weights. It combines polarity analysis to establish a list of strategic elements and uses a rule engine to connect roles and needs to strategic objectives to generate a stakeholder graph. Relevant entities include roles, needs, and strategic objectives.

[0051] The process extraction module calculates the demand weight of each demand in the stakeholder graph using a graph neural network. It then embeds the demand semantic vectors supporting strategic objectives into the strategic semantic vectors of each strategic objective in the strategic element list according to their corresponding demand weights, generating a weighted strategic vector. Based on the stakeholder graph constraining the GPT4-Turbo model, it extracts the process nodes of business documents and associates them to generate a process element table. The strategic semantic vector and demand semantic vector are generated by a preset language model of strategic objectives and demand inputs.

[0052] The process generation module encodes the status of each process element in the process element table, uses a near-end strategy to optimize the learning network, sets action space filtering based on stakeholder graphs, and connects process elements based on strategic transmission and reward guidance to generate the final-level process diagram. The status encoding is implemented based on the demand weight table and the weighted strategic vector set.

[0053] The integrated setup module determines the programming language and compilation environment based on customer needs. It filters and matches the process semantic vectors of the process nodes in the final-level flowchart with the code semantic vectors of the code segments in the code library using a logical verification function. It fills in the matching code segment for each process node to generate the business system. The logical verification function is equal to the norm of the intersection of the process semantic vector and the code semantic vector divided by the norm of the code semantic vector.

[0054] Compared with existing technologies, this invention automatically identifies entities through a bidirectional, non-discriminatory coding model, and utilizes an attention polarity model and graph neural networks to achieve semantic encoding of strategic objectives, calculation of demand weights, and aggregation of elements, generating a strategic element list and stakeholder map. It extracts process nodes using GPT4-Turbo model constraints and generates final-level flowcharts using a near-end strategy optimization learning network. Based on a logical verification function, it achieves automated matching of process semantic vectors and code semantic vectors, forming an end-to-end closed loop of "strategy-process-code". This solution builds a cross-modal semantic mapping mechanism through the integration of multiple technologies, overcoming the bottleneck of manual intervention and systematically solving the technical bottlenecks of cross-level element transmission and cross-modal data mapping. It achieves full-process automation from strategic analysis to business system generation, significantly improving the efficiency of strategy implementation and the accuracy of system development. Attached Figure Description

[0055] Figure 1 Build method flowcharts for business systems;

[0056] Figure 2 Flowchart of the processing steps for the bidirectional indistinguishable coding model;

[0057] Figure 3 This is a schematic diagram of the attention polarity model;

[0058] Figure 4 Build a system diagram for the business system. Detailed Implementation

[0059] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0060] Example 1

[0061] like Figure 1 As shown, this invention discloses a method for building a business system, including the following steps:

[0062] A bidirectional non-discriminatory coding model is used to identify different relevant entities in each document. An attention polarity model is used to semantically encode each strategic objective and calculate internal and external attention weights through learnable query vectors. Internal or external elements are aggregated and classified, and a list of strategic elements is established by combining polarity analysis. A rule engine is used to match the roles and needs of relevant entities with strategic objectives to generate a stakeholder graph. Relevant entities include specific roles, needs, and strategic objectives, such as "suppliers", "order tracking", and "increasing market share".

[0063] The demand weight of each demand in the stakeholder graph is calculated by graph neural network, and a demand weight table is constructed. The demand semantic vectors supporting each strategic goal in the strategic element list are embedded in the strategic semantic vector according to the corresponding demand weights to construct a weighted strategic vector set. The entity chain rule constraint of the stakeholder graph is introduced into the GPT4-Turbo model to extract process nodes in business documents and associate them with the relevant role demand chains to generate a process element table. The strategic semantic vector and demand semantic vector are generated by inputting the strategic goal and demand into a preset language model, which includes the BERT model and the RoBERTa model.

[0064] Each process element in the process element table is converted into a corresponding process state vector through state encoding. A near-end strategy optimization learning network is adopted. Action space filtering is set based on the graph structure of the stakeholder graph to prohibit cross-chain jumps. State transition actions are learned based on strategic transmission rewards to connect different process state vectors, generating a final-level process diagram that meets the requirements of the business document. The state encoding is implemented based on the demand weight table and the weighted strategic vector set.

[0065] Based on customer needs, the programming language and compilation environment are determined, the final-level flowchart is obtained, and the text description of each process node and the functional label of each code segment in the code library are converted into process semantic vector and code semantic vector respectively based on the preset language model. The matching degree between process nodes and code segments is calculated based on the logical verification function, and the code segment with the highest matching degree is selected to fill the process node to generate the business system. The logical verification function is equal to the intersection norm of the process semantic vector and the code semantic vector divided by the norm of the code semantic vector.

[0066] like Figure 2 As shown, the bidirectional non-discriminating coding model further identifies strategic objectives and related entities in the underlying documents, including the following steps:

[0067] The WordPiece word segmenter is used to split each document into short sentences, and each short sentence is further split into sub-words. A start tag is added at the beginning of the document, and a split tag is added at the end of each short sentence.

[0068] Find the maximum number of subwords in a sentence, and insert filler words between the end of all sentences with a subword count less than the maximum number of subwords and the corresponding delimiter so that the number of subwords in each sentence is equal to the maximum number of subwords;

[0069] Each subword is mapped to a number vector based on the vocabulary. The number vector for filling subwords is an all-zero vector. Sine and cosine positional encoding is used to generate position vectors based on the order of each subword in the document and superimposes them with the subword number vectors to fuse semantic and positional information, thereby generating the embedding vector of each subword.

[0070] The embedding vectors of words in each short sentence are concatenated and fed into a 12-layer Transformer encoder for multi-scale analysis. Each layer includes a multi-head self-attention mechanism and a feedforward neural network. The multi-head self-attention mechanism concatenates the results of the 12 self-attention mechanisms and projects them back to the original dimension before inputting the self-attention mechanism through linear weights. After residual connection and layer normalization, it enters the feedforward neural network for feedforward processing. After passing through the 12-layer Transformer encoder, the context representation vector is output.

[0071] The context representation of each word in the context representation vector is linearly modulated by a linear classifier and transformed into a probability distribution for each word by a Softmax function. The BIN label with the highest probability is selected and assigned to the word. The probability distribution records the probability of a word belonging to each BIN label. The BIN label adopts a prefix + type format. For any specific entity A, there are two BIN labels, BA and IA, which represent the start word and continuation word of entity A, respectively. For non-entities, there is only label NN. Entity A is a strategic goal, a specific role, a specific need, or a specific strategic goal. The BIN format makes the entity boundary clear and separable, reducing the problem of boundary ambiguity.

[0072] Based on the principle of continuous merging, the sub-words of the beginning words and continuation words of short sentences that are consecutive and whose BIN tags correspond to the same entity are merged to identify the relevant entities of each short sentence in the document.

[0073] Furthermore, the principle of successive mergers includes the following steps:

[0074] For the first word in each short phrase, perform a start test. If the label of the first word is NN, skip it and perform a new start test for the second word. If the label of the first word represents the start word of an entity, perform a continuation test for the second word.

[0075] If the label of the second sub-word represents a continuation word of an entity and is the same as the entity of the first sub-word, then the two sub-words are merged and a continuation test is performed on the third sub-word. If the label of the second sub-word is NN, then the first sub-word is treated as an entity alone and a new start test is performed on the third sub-word. If the label of the second sub-word represents the start word of an entity, then the first sub-word is treated as an entity alone and the continuation test of the second sub-word is treated as a new start test, and a new continuation test is performed on the third sub-word.

[0076] Continue until all words in the short sentence have been checked.

[0077] like Figure 3 As shown, the attention polarity model further establishes a list of strategic elements, including the following steps:

[0078] Strategic objectives are semantically encoded using a pre-defined language model and converted into strategic semantic vectors.

[0079] The product of the strategic semantic vector and the internal query vector, and the product of the strategic semantic vector and the external query vector are transformed into internal weights and external weights through the self-attention mechanism. The internal query vector and the external query vector are learnable parameters in the attention polarity model. The internal weights and external weights reflect the semantic relevance between the strategic semantic vector and the internal and external elements, respectively. The internal elements include the strengths (S) and weaknesses (W) in the SWOT analysis, and the external elements include the opportunities (O) and challenges (T) in the SWOT analysis.

[0080] The product of the concatenated strategic semantic vector and the internal and external weights is transformed into the internal and external probability distributions of the strategic objective through linear modulation and the Softmax function. The internal and external probability distributions record the probabilities of the strategic objective with internal and external elements.

[0081] By sliding three convolutional kernels with different window sizes on the strategic semantic vector to capture local semantic features at different scales, the PReLU function is used to enhance the nonlinear expressive power of the three convolutional results, and dimensionality reduction based on max pooling is performed to enhance the most salient local features, generating three semantic features at different scales, with window sizes of 1, 3 and 5 respectively.

[0082] Three semantic features of different scales are concatenated and fused and reduced in dimensionality through a fully connected layer. The polarity probability distribution is output based on the Softmax function. The polarity probability distribution records the probability that strategic objectives belong to positive polarity and negative polarity. Among them, advantages S and opportunities O are positive elements, and disadvantages W and challenges T are negative elements.

[0083] Based on probability and statistics, we construct the SWOT element probability distribution of strategic objectives by combining internal and external probability distributions and polar probability distributions. We establish an edge between the SWOT element with the highest probability and the strategic objective, and the value of the edge is equal to the highest probability in the SWOT element probability distribution.

[0084] Strategic objectives are selected and grouped according to strengths (S), weaknesses (W), opportunities (O), and challenges (T), and the grouping results are reorganized in a structured table to generate a list of strategic elements.

[0085] Furthermore, a rules engine is used to match and connect the roles and needs of relevant entities with strategic objectives, generating a stakeholder graph, including the following steps:

[0086] Establish entity chain rules, establish the attribution relationship of role demand chain, that is, demand is put forward by role, establish the support relationship of demand strategic goal chain, that is, satisfying demand can partially or fully achieve strategic goal;

[0087] Establish a two-way correspondence rule, where roles and needs, and needs and strategic goals are all in a many-to-many form. The two-way correspondence rule means that a role can raise multiple needs at the same time, a need can also be raised by different roles, a need can support multiple strategic goals, and a strategic goal can also be supported by multiple needs at the same time.

[0088] Each relevant entity is converted into a bag-of-words feature vector using a TF-IDF converter. Each role, requirement, and strategic goal is converted into a role semantic vector, requirement semantic vector, and strategic semantic vector using a pre-defined language model. The bag-of-words feature vector and the corresponding semantic vector of each relevant entity are concatenated to form an entity description vector.

[0089] Permuting and combining each role and requirement to construct role-requirement pairs, calculating the cosine similarity of each role-requirement pair, establishing attribution edges between roles and requirements whose cosine similarity is greater than or equal to the similarity threshold, permuting and combining each requirement and strategic goal to construct requirement-goal pairs and calculating the cosine similarity, establishing support edges between requirements and strategic goals whose cosine similarity is greater than or equal to the similarity threshold, and the attribution edge value and support edge value are equal to the corresponding cosine similarity.

[0090] Once the ownership edges and support edges are fully established, a stakeholder graph is generated.

[0091] Specifically, the graph neural network starts with each demand in the stakeholder graph, retrieves the number of roles with belonging edges to each demand as the out-degree of each demand, divides the out-degree of each demand by the out-degree of all demands, and generates the demand weight of each demand. The demand weight reflects the importance of each demand. The demand weights of each demand are combined to construct a demand weight table. Each strategic objective in the strategic element list is encoded into a strategic semantic vector through a preset language model. Demands that can support each strategic objective are retrieved from the stakeholder graph and converted into demand semantic vectors through a preset language model. Based on the corresponding demand weights in the demand weight table, a weighted sum is calculated and superimposed on the strategic semantic vector of each strategic objective to generate a weighted strategic vector for each strategic objective. The weighted strategic vector set is then constructed by combining these elements.

[0092] Specifically, in the GPT4-Turbo model, process node role matching rules are pre-established. In each process node and role matching pair, all relevant role requirement chains from the stakeholder graph are introduced, and a template rule library is constructed by integrating them in JSON format. The natural language processing capabilities of the GPT4-Turbo model are used to identify process nodes in business documents and match them in the template rule library to obtain the role requirement chains that match the process nodes. Through a preset language model, the text description of the process node and the requirements in each role requirement chain are converted into process semantic vectors and requirement semantic vectors, respectively. The cosine similarity between the process semantic vector and each relevant requirement semantic vector is calculated. The role requirement chain corresponding to the requirement semantic vector with the largest cosine similarity and the process node are organized into process elements according to JSON format. The process elements of all process nodes are combined to construct a process element table.

[0093] Furthermore, each process element in the process element table is converted into a corresponding process state vector through state coding, including the following steps:

[0094] The text descriptions and roles of the process nodes in each process element are converted into process semantic vectors and role semantic vectors respectively using a preset language model;

[0095] Based on the stakeholder graph, the requirements related to roles in the process elements are obtained, and the corresponding requirements weights are obtained from the requirements weight table. Based on the graph neural network, the requirements related to roles are weighted and summed according to the corresponding requirements weights, and then superimposed on the role semantic vector of the role to generate a weighted role vector.

[0096] In the stakeholder map, retrieve the strategic objectives that the requirements of the process elements can support and obtain the corresponding weighted strategic vectors. Based on the supporting boundary values ​​between the corresponding requirements and strategic objectives, perform secondary weighted aggregation on all the obtained weighted strategic vectors to generate the strategic related requirement vector for each process element.

[0097] By concatenating the process semantic vector, weighted role vector, and strategic related requirement vector of each process element, a process state vector for each process element is generated. The process state vector indirectly reflects the state distribution of strategic goals that can be achieved by roles in each process element when choosing different requirements at process nodes.

[0098] Furthermore, a near-end strategy is employed to optimize the learning network and generate a final-level flowchart that meets the requirements of the business documentation, including the following steps:

[0099] The business strategic objectives in the business documents are converted into business strategic semantic vectors through a pre-defined language model. The strategic transmission reward is designed to be equal to the cosine similarity between the current process state vector and the business strategic semantic vector multiplied by the transmission coefficient. The transmission coefficient is equal to the shortest path length of the role corresponding to the current process state vector in the stakeholder graph to reach the business strategic objective. The path length is equal to the sum of the edge values ​​of the belonging edges and / or supporting edges traversed by the role to reach the business strategic objective.

[0100] Define the action as selecting the next process element and linking it to the current process element;

[0101] Based on the traditional near-end policy optimization algorithm, action space filtering is set based on the graph structure of the stakeholder graph to generate a policy network. The policy network is used to guide the adaptive execution of the optimal action. Action space filtering refers to eliminating process elements that do not have a direct or indirect connection with the roles and needs in the current process element when selecting the next process element.

[0102] Select the process state vector of a process element as the starting point. The policy network is based on the near-end policy optimization algorithm and uses strategic transmission reward as expert guidance to generate the best action to reach the next process element. Repeat this step until all process elements are traversed or the next process element cannot be selected based on spatial filtering. If all process elements are traversed, a final-level flowchart is generated. If the next process element cannot be selected, the flowchart under construction is discarded and an unselected process element is reselected as the starting point.

[0103] Obtain all final-level flowcharts that traverse all process elements, and select the final-level flowchart with the highest cumulative strategic transmission reward for output. This final-level flowchart is the one that best meets the business requirements.

[0104] Example 2

[0105] like Figure 4 As shown, the present invention also discloses a business system building system for executing the business system building method, including a graph construction module, a process extraction module, a process generation module and an integration building module;

[0106] The graph construction module uses a bidirectional non-discriminatory coding model to identify different related entities in each document, uses an attention polarity model to semantically encode each strategic objective and calculates internal and external attention weights through learnable query vectors, aggregates and classifies them into internal or external elements, and combines polarity analysis to classify and establish a list of strategic elements. The rule engine uses a rule engine to match and connect the roles and needs of related entities to strategic objectives to generate a stakeholder graph.

[0107] The process extraction module calculates the demand weight of each demand in the stakeholder graph using a graph neural network, constructs a demand weight table, and embeds the demand semantic vectors supporting each strategic objective into the strategic semantic vector of each strategic objective in the strategic element list according to the corresponding demand weights, constructing a weighted strategic vector set. It introduces entity chain rule constraints from the stakeholder graph into the GPT4-Turbo model, extracts process nodes from business documents and associates them with the relevant role demand chains, generating a process element table.

[0108] The process generation module converts each process element in the process element table into a corresponding process state vector through state encoding. It adopts a near-end strategy optimization learning network, sets action space filtering based on the graph structure of the stakeholder graph to prohibit cross-chain jumps, learns state transition actions based on strategic transmission rewards to connect different process state vectors, updates network parameters based on the near-end strategy optimization algorithm, and generates a final-level process diagram that meets the requirements of the business document.

[0109] The integrated setup module determines the programming language and compilation environment based on customer needs, obtains the final-level flowchart, and converts the text description of each process node and the functional label of each code segment in the code library into process semantic vectors and code semantic vectors respectively based on the preset language model. It calculates the matching degree between process nodes and code segments based on the logical verification function, selects the code segment with the highest matching degree to fill the process node, and generates the business system.

[0110] This invention discloses a method and system for building a business system. It automatically identifies entities using a bidirectional, non-discriminating coding model, and utilizes an attention polarity model and graph neural networks to achieve semantic encoding of strategic objectives, calculation of demand weights, and aggregation of elements, generating a strategic element list and stakeholder graph. It extracts process nodes using GPT4-Turbo model constraints and generates final-level flowcharts using a near-end strategy optimization learning network. Based on a logical verification function, it achieves automated matching of process semantic vectors and code semantic vectors, forming an end-to-end closed loop of "strategy-process-code". This solution builds a cross-modal semantic mapping mechanism through multi-technology integration, overcoming the bottleneck of manual intervention and systematically solving the technical bottlenecks of cross-level element transmission and cross-modal data mapping. It achieves full-process automation from strategic analysis to business system generation, significantly improving the efficiency of strategy implementation and the accuracy of system development.

[0111] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for building a business system, characterized in that, Includes the following steps: A bidirectional non-discriminatory coding model is used to identify roles, needs, and strategic objectives in documents. An attention polarity model is used to semantically encode strategic objectives and aggregate and classify them into internal or external elements. A list of strategic elements is established by combining polarity analysis. A rule engine is used to connect roles and needs to strategic objectives to generate a stakeholder map. The system generates corresponding semantic vectors for roles, requirements, and strategic objectives based on a preset language model. It calculates the requirement weight of each requirement in the stakeholder graph and embeds the requirement semantic vectors supporting each strategic objective into the strategic semantic vectors of the strategic objectives in the strategic element list according to their corresponding requirement weights. This generates a weighted strategic vector. The system then constrains the AI ​​model based on the stakeholder graph, extracts process nodes from business documents, and associates them to generate a process element table. The process elements in the process element table are state-encoded based on the demand weight table and the weighted strategic vector of each strategic objective. A near-end strategy optimization learning network is adopted, action space filtering is set based on the stakeholder graph, and process elements are connected based on strategic transmission reward guidance to generate the final-level process diagram. The business system is generated by matching the process nodes in the final-level flowchart with the code segments in the code library based on the logical verification function. The logical verification function is equal to the norm of the intersection of the process semantic vector of the process node and the code semantic vector of the code segment divided by the norm of the code semantic vector.

2. The business system construction method as described in claim 1, characterized in that, The method of identifying roles, needs, and strategic objectives in documents using a bidirectional non-discriminatory coding model includes the following steps: The document is split into short sentences and then further split into sub-words using a word segmenter. A delimiter is added to the end of each short sentence, and padding sub-words are inserted before the delimiter to make each short sentence have the same number of sub-words. Based on the vocabulary, sub-words are mapped to number vectors. Based on the order of sub-words in the document, position vectors are generated and superimposed with the sub-word number vectors to generate the sub-word embedding vectors. The embedding vectors of each word in the short sentence are concatenated and input into the Transformer encoder. After processing through a multi-head self-attention mechanism, residual connections, layer normalization, and feedforward neural network, a context representation vector is generated. The context representation of each word in the context representation vector is transformed into a probability distribution of each word through linear modulation and Softmax function by a linear classifier, and the BIN label with the highest probability is selected and assigned to the word. The BIN label of entity A has BA and IA representing the start word and continuation word of entity A, respectively, while non-entity only has label NN. Based on the principle of continuous merging, the sub-words of the first word and the continuation word that are consecutive in sequence in the short sentence and whose BIN tags correspond to the same entity are merged. The entities include roles, needs and strategic goals.

3. The business system construction method as described in claim 2, characterized in that, The principle of continuous consolidation includes the following steps: If the BIN tag test result based on the first sub-word is a non-entity or entity start word, then a new test or a continuation test should be performed directly on the BIN tag of the second sub-word. If the BIN tag of the second sub-word is a non-entity, then a new check is performed directly on the BIN tag of the third sub-word; If the BIN tag of the second sub-word is a continuation of the entity and is the same as the entity of the first sub-word, then merge the two sub-words and continue to check the continuation of the BIN tag of the third sub-word; If the BIN tag of the second sub-word is the start word of the entity, the continuation test of the second sub-word is treated as a new test, and a new continuation test is performed on the third sub-word; Continue until all words in the short sentence have been checked.

4. The business system construction method as described in claim 1, characterized in that, The strategic element list is established using an attention polarity model, including the following steps: Strategic objectives are transformed into strategic semantic vectors through a pre-defined language model; The product of the strategic semantic vector and the internal and external query vectors is transformed into internal and external weights through a self-attention mechanism. The product of the concatenated strategic semantic vector and the internal and external weights is transformed into the internal and external probability distributions of the strategic objective through linear modulation and the Softmax function; Convolutional kernels with different window sizes are used to convolve with strategic semantic vectors, and semantic features of different scales are generated by sequentially passing them through the PReLU function and max pooling. After concatenation, the features are output as a polarity probability distribution through a fully connected layer and the Softmax function. By combining internal and external probability distributions with polarity probability distributions, a SWOT element probability distribution of strategic objectives is constructed. An edge is established between the SWOT element with the highest probability and the strategic objective, and the value of the edge is equal to the highest probability in the SWOT element probability distribution. Strategic objectives are grouped according to each SWOT element to generate a list of strategic elements.

5. The business system construction method as described in claim 1, characterized in that, The process of using a rules engine to connect roles and needs with strategic objectives to generate a stakeholder graph includes the following steps: Establish the attribution relationship of the role requirement chain and the supporting relationship of the requirement strategic goal chain, and determine that the relationship between roles and requirements, and between requirements and strategic goals, is many-to-many. Each relevant entity is transformed into a bag-of-words feature vector and a corresponding semantic vector using a TF-IDF converter and a pre-defined language model, and then concatenated into an entity description vector. Each role and need, as well as each need and strategic goal, is arranged and combined separately to construct role-need pairs and need-strategy goal pairs. Cosine similarity is calculated for each pair. For roles and needs with cosine similarity greater than or equal to the similarity threshold, belonging edges and supporting edges are established respectively to generate a stakeholder graph. The belonging edge value and the supporting edge value are equal to the corresponding cosine similarity.

6. The business system construction method as described in claim 1, characterized in that, The near-end strategy is used to optimize the learning network for generating the final-level flowchart, including the following steps: The business strategic objectives in the business documents are converted into business strategic semantic vectors through a pre-set language model. The strategic transmission reward is designed to be equal to the cosine similarity between the current process state vector and the business strategic semantic vector multiplied by the transmission coefficient. The transmission coefficient is equal to the sum of the edge values ​​of the belonging edges and / or supporting edges that the role corresponding to the current process state vector passes through in the stakeholder graph to reach the business strategic objective. Based on the near-end strategy optimization algorithm, spatial filtering is performed based on the stakeholder graph. When selecting the next process element, process elements with roles or requirements that do not have direct or indirect connections with the roles and requirements in the current process element in the stakeholder graph are eliminated. Select a process state vector of a process element as the starting point, and select the next process element to connect based on the near-end strategy optimization algorithm and strategic transmission reward. Repeat this step until it stops. If the stop is due to all process elements having been traversed, a final-level flowchart is generated. If the stop is due to spatial filtering resulting in no process elements, the flowchart under construction is discarded and a new process element is selected as the starting point. Select the output that maximizes the cumulative strategic transmission reward from the generated final-level flowchart.

7. The business system construction method as described in claim 1, characterized in that, The process element encoding based on the demand weight table and the weighted strategic vector of each strategic objective to the process element table includes the following steps: The text descriptions and roles of process nodes in each process element are transformed into process semantic vectors and role semantic vectors using a preset language model; Based on the stakeholder map and demand weight table, the demand and corresponding demand weight related to the roles in the process elements are obtained. The role-related demand is generated into a demand semantic vector through a preset language model. The demand is weighted and summed based on the corresponding demand weights and the role semantic vector of the role is superimposed to generate a weighted role vector. Retrieve the strategic objectives supported by the needs in the process elements from the stakeholder map and obtain the corresponding weighted strategic vectors. Based on the weighted strategic vectors obtained by weighted aggregation of the supporting boundary values ​​between the corresponding needs and strategic objectives, generate the strategic association demand vector for each process element. By combining the process semantic vector, weighted role vector, and strategic related requirement vector of each process element, a process status vector for each process element is generated.

8. The business system construction method as described in claim 1, characterized in that, In the stakeholder graph, the number of roles belonging to the edge of each demand is counted as the out-degree. The out-degree of each demand is divided by the out-degree of all demands and the demand weight that generates each demand. Each strategic objective in the strategic element list is encoded into a strategic semantic vector through a preset language model. The demands supporting each strategic objective are retrieved from the stakeholder graph and converted into demand semantic vectors through the preset language model. The demands are weighted and summed based on the corresponding demand weights and superimposed with the strategic semantic vector of each strategic objective to generate a weighted strategic vector for each strategic objective.

9. The business system construction method as described in claim 1, characterized in that, In the AI ​​model, process node role matching rules are pre-established. All relevant role requirement chains are introduced into each process node and role matching pair and integrated to build a template rule base. The natural language processing capability of the AI ​​model is used to identify process nodes in business documents and match them with the template rule base to obtain the role requirement chains that match the process nodes. The text description of the process nodes and the requirements in each role requirement chain are converted into process semantic vectors and requirement semantic vectors through a preset language model, and the cosine similarity is calculated. The role requirement chain corresponding to the requirement semantic vector with the largest cosine similarity is organized into process elements with the process nodes. The process elements of all process nodes are combined to build a process element table. The AI ​​model used is the GPT4-Turbo model.

10. A business system construction system, characterized in that, It includes a graph construction module, a process extraction module, a process generation module, and an integration and setup module; The graph construction module uses a bidirectional non-discriminatory coding model to identify roles, needs, and strategic goals in documents. It uses an attention polarity model to semantically encode strategic goals and aggregate and classify them into internal or external elements. It combines polarity analysis to establish a list of strategic elements and uses a rule engine to connect roles and needs to strategic goals to generate a stakeholder graph. The process extraction module generates corresponding semantic vectors for roles, requirements, and strategic support based on a preset language model, calculates the requirement weight of each requirement in the stakeholder graph, embeds the requirement semantic vectors of the requirements supporting each strategic goal into the strategic semantic vectors of the strategic goals in the strategic element list according to the corresponding requirement weights, generates a weighted strategic vector, and extracts process nodes from business documents and associates them to generate a process element table based on the stakeholder graph constrained AI model. The process generation module encodes the process elements in the process element table based on the demand weight table and the weighted strategic vector of each strategic objective. It adopts a near-end strategy optimization learning network, sets action space filtering based on stakeholder graph, and connects process elements based on strategic transmission reward guidance to generate the final-level process diagram. The integrated building module matches the process nodes in the final-level flowchart with the code segments in the code library based on the logical verification function to generate the business system. The logical verification function is equal to the norm of the intersection of the process semantic vector of the process node and the code semantic vector of the code segment divided by the norm of the code semantic vector.

Images