Business rule mining and business process construction method based on prompt learning

Abstract

The invention discloses a business rule mining and process construction method based on prompt learning. The method comprises a rule mining module, a relation identification module and a process construction module. Firstly, a Prompt is constructed, a business rule is defined in a two-tuple format, a big language model and a context learning technology are combined, rules in a business text are recognized, and key information is extracted. Then, in a relation identification module, according to the combination of conditions and actions, the logic relation between the rules is judged; and finally, in the process construction module, integrating the results of the first two modules, determining the dependency relationship between the rules through dependency relationship prompt Prompt and action value analysis, and generating a complete process model. According to the method, the process is constructed in a structured and intelligent mode, the dependency relationship analysis is combined, the generated process model is visual and explanatory, and support is provided for process management and optimization of a complex business scene.

Description

Technical Field

[0001] The present invention relates to the fields of natural language processing (NLP) and business process management (BPM), and in particular to a method for business rule mining and business process construction based on prompt learning. Background Art

[0002] With the rapid development of information technology and the deepening of digital transformation of enterprises, intelligent tools play an increasingly important role in enterprise operations. These tools help enterprises improve operational efficiency, optimize resource allocation and enhance customer service experience by combining big data analysis, artificial intelligence technology and automation. Among them, the multi-round dialogue system, as a key technology that can automatically complete interaction with users, has been widely used in customer service, internal management, business process automation and other scenarios, and has become an important support for enterprises to improve service levels and realize business intelligence.

[0003] Traditional multi-round dialogue systems mainly rely on predefined rules or templates to respond to user needs by designing a series of fixed logical paths. However, although this approach has certain applicability in simple and repetitive task scenarios, its limitations are very obvious when dealing with complex and changing business needs. On the one hand, the scalability and flexibility of the rules are restricted, making it difficult to meet the needs of diverse business scenarios; on the other hand, in the face of dynamically changing user input and task requirements, the system cannot adjust the dialogue strategy in real time, which can easily lead to dialogue interruptions or a decline in user experience.

[0004] The introduction of large language models (LLMs) has brought new development opportunities for multi-round dialogue systems. Large language models based on deep learning technology have powerful semantic understanding and generation capabilities, and can achieve excellent performance in natural language processing tasks. However, although LLMs have significant advantages at the semantic level, they still have significant limitations in business process planning and rule execution. Specifically, in the absence of clear guidance, large language models find it difficult to plan process logic that meets business needs, and are prone to generating inaccurate or incomplete task paths. According to research data, only about 12% of generated plans can be accurately executed in complex scenarios, which greatly limits their application effect in actual business scenarios.

[0005] In response to the above challenges, how to efficiently mine key business rules from the company's business documents and build business processes based on these rules that can accurately guide and optimize conversational interactions has become a technical problem that the industry urgently needs to solve. In practical applications, business documents contain a large amount of potential rule information, which is usually described in natural language and includes both conditional constraints and task execution logic. By efficiently converting this information into structured rule data, it can provide clear guidance for multi-round dialogue systems, thereby significantly improving the system's execution capabilities and user experience. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a method for business rule mining and business process construction based on prompt learning. This method automatically extracts business rules from business documents by combining the semantic understanding ability of a large language model with prompt learning technology, and generates a business process model with logical consistency and execution feasibility. Through the technical framework of rule mining, relationship identification and process generation, the present invention significantly improves the execution capability and interactive experience of the dialogue system in complex business scenarios.

[0007] A method for business rule mining and business process construction based on prompt learning, comprising the following steps:

[0008] S1: Business rule mining: first construct a prompt for business rule mining, obtain the text of the business rules, and extract the business rules in the form of tuples from the text of the business rules through the prompts for business rule mining combined with the large language model. The format of the tuple is <condition, action>, where the condition is used to describe the preconditions for triggering the business rules, and the action defines the execution operation associated with the condition; further preferably, by constructing a prompt template Prompt for the business rules and combining the powerful natural language processing capabilities of the large language model, the business rules defined in the form of tuples can be efficiently extracted from the business text. The format of the tuple is <condition, action>, where the condition is used to describe the preconditions for triggering the business rules, and the action defines the execution operation associated with the condition;

[0009] S2: Business rule relationship identification: The business rules in the form of two tuples extracted in step S1 and the association judgment prompt are input into the large language model, and the association relationship between the conditions and actions in the two tuples is judged and analyzed by the large language model. If there is an association relationship, the requirements of the parallel structure are met, and the business rules in the form of two tuples that meet the requirements of the parallel structure are taken as the business rules that meet the parallel structure, and then step S3 is entered; further preferably, based on the business rules in the form of two tuples extracted in step (S1), by constructing an association judgment prompt Prompt, the many-to-one association relationship between the conditions and actions between the rules is further identified. Through this process, it can be judged whether the requirements of the parallel structure can be met between multiple conditions;

[0010] S3: Business process construction: using dependency prompts and contextual learning technology, analyze the dependencies between the actions in the binary business rules obtained in step S1 and the business rules satisfying the parallel structure obtained in step S2, and generate a business process model, wherein the dependencies between the actions refer to the execution order of the actions; further preferably, combining the business rules obtained in step (S1) and the business rules satisfying the parallel structure obtained in step (S2), using dependency prompts and contextual enhanced learning technology, deeply analyze the dependencies between the rules. On this basis, a business process model is generated, which is not only logically complete, but also clear in execution. This model covers sequential structure, parallel structure and selection structure, comprehensively reflects the execution logic and process path of business rules, and ensures the clarity and efficiency of business processes.

[0011] S4: Apply the business process model to financial services equipment.

[0012] The method of the present invention is mainly divided into three modules: a business rule mining module, a business rule relationship identification module, and a business process construction module. First, a prompt is constructed in the first module, and the business rules are defined in the format of a binary. Subsequently, the rules in the business text are identified using a large language model and context learning technology, and key information is mined. Next, the logical relationship between the business rules is further determined in the second module, which mainly determines the parallel relationship between the business rules through the combination of conditions and actions. Finally, the output results of the first two modules are integrated in the third module, the dependency prompt is used, and the dependency between the business rules is determined according to the action value. Through the above steps, a complete process model is finally generated.

[0013] In step S1, a prompt for business rule mining is first constructed to obtain the text of the business rule. The business rule mining prompt is combined with a large language model to extract the bigram business rule from the text of the business rule, specifically including:

[0014] (1.1) Through conceptual definition, the core elements of business rules, including conditions and actions, are clarified, and then the prompts for business rule mining are constructed. The text of business rules is parsed through the optimized large language model to obtain business rules in the form of tuples. The conditions include slot type, logical judgment and reference value. The slot type includes currency and customer type. The logical judgment is to describe the logical relationship between the slot type and the reference value, and the value range is ["contains", "includes", "equal to", "less than", "greater than", "less than or equal to", "greater than or equal to"]. The reference value refers to the text used to describe the correspondence between the slot type and the business rule. The reference value includes two types: numeric and enumeration. Logical judgment: describes the logical relationship between the slot type and the reference value.

[0015] Further preferably, the step S1 includes the following sub-steps:

[0016] (1.1) Design business rule extraction prompt: Construct a prompt described in natural language, clarify the task objectives, guide the model to play the role of a logic reasoning professor, and parse business texts from a professional perspective. Through concept definition, clarify the core elements of the rules, including conditions, actions and their components (slot type, logical judgment, reference value).

[0017] (1.2) Optimize Prompt and introduce contextual learning: Provide examples to help the model understand the task context and enhance its rule extraction capabilities. Concatenate the business text with the optimized Prompt to ensure accurate extraction of rule tuples.

[0018] The slot type includes but is not limited to information categories such as currency, customer type, etc., and the reference value includes two types: digital type and enumeration type, which are used to describe the corresponding relationship between the slot type and the specific business text to ensure the accuracy of rule extraction.

[0019] In step S2, the business rules in the form of bigrams and the associated judgment prompts extracted in step S1 are input into the large language model, and the association relationship between the conditions and actions in the bigrams is judged and analyzed by the large language model, specifically including:

[0020] (2.1) Separate the conditions and actions in the binary business rules to form independent nodes, and generate all possible condition and action combination rules through the association between conditions and actions;

[0021] (2.2) Construct an association judgment prompt, input the association judgment prompt and all possible condition and action combination rules obtained in step (2.1) into the large language model, and identify the association relationship between the condition and the action.

[0022] Further preferably, the step S2 includes the following sub-steps:

[0023] (2.1) Extraction of association between conditions and actions: Separate the conditions and actions in the business rules to form independent nodes, and generate all possible condition and action combination rules through the condition and action combination generator Prompt.

[0024] (2.2) Many-to-one relationship judgment and parallel relationship construction: By constructing an association judgment prompt, the many-to-one association relationship between conditions and actions is identified, and the parallel relationship is extracted.

[0025] The step S3 comprises the following sub-steps:

[0026] (3.1) Integrate parallel and selection structures: Combine the outputs of the first two modules to integrate the parallel and selection structures in the business rules. In this process, the selection structure is directly extracted from the output of the first module, reflecting the selection path of decision points and conditions.

[0027] (3.2) Sequential relationship determination: Use the dependency prompt to determine the sequential relationship of the integrated results. By analyzing the dependency relationship between action values, the execution order of each business rule is determined to ensure the rationality of the business process logic.

[0028] (3.3) Build a complete process model: Based on the determined sequence relationship and integrated structure, build a complete business process model to ensure that all activities in the process are executed in the correct order and under the correct conditions.

[0029] The dependency prompt Prompt can accurately determine the dependency between rules by combining the semantic understanding ability of the large language model, ensuring that the generated process model is executable and efficient in actual business scenarios.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] The present invention significantly improves the accuracy of business rule mining and the logic of process modeling through a large language model based on prompt learning and context learning technology. This method can efficiently extract business rules defined in the form of tuples from business texts, and intelligently generate a complete process model containing sequential structures, parallel structures, and selection structures, ensuring rigorous logic and feasible execution. Compared with traditional technologies, this patent has significant advantages such as high degree of automation, strong adaptability, and excellent scalability. It can be widely used in complex business scenarios such as finance, medical care, and supply chain, providing intuitive and efficient technical support for process management and optimization.

[0032] The method of the present invention can realize the mining of business rules and the construction of business processes in a structured and intelligent way. By applying prompt learning technology, the accuracy of business rule identification and the logic of process construction are significantly improved. Combined with the dependency analysis of business rules, the business process generated by the present invention is intuitive and highly interpretable, providing strong support for process management and optimization in complex business scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A schematic diagram of the method provided by the present invention;

[0034] Figure 2 A technical roadmap provided for the present invention; DETAILED DESCRIPTION

[0035] The present invention is described in detail below in conjunction with the accompanying drawings.

[0036] The present invention provides a method for mining business rules and building business processes based on prompt learning, such as Figure 1 As shown, the following steps are included:

[0037] Step 1: Design and optimize business rule mining prompt

[0038] The first step of the specific implementation of the present invention is to design and optimize the business rule mining prompt, and efficiently extract the business rule tuples that meet the specifications from the business text through the prompt learning technology. This process not only ensures the accurate extraction of rules, but also significantly improves the ability of the generative large model to understand the task, providing a reliable data foundation for subsequent business process modeling.

[0039] Step 1.1: Design of business rule extraction prompt

[0040] In order to realize the automatic identification and extraction of business rules, a prompt described in natural language was first constructed to clarify the task objectives and specific operation requirements. The core of the prompt design is to guide the model to play the role of a professor of logical reasoning, parse the business text from a professional perspective, deeply explore the logical relationships and implicit rules therein, and convert them into structured two-tuple representations. By constructing the concept definition of specific tasks, the task background and objectives are further clarified, making the rule extraction process more systematic and efficient. The specific design content is as follows:

[0041] ●Rule description: Business rules are expressed in the form of two tuples, with the structure of:

[0042] Condition: The prerequisite for triggering the rule, which consists of three components:

[0043] ■Slot type: refers to the noun information extracted from the text, such as currency, customer type, etc.;

[0044] ■Logical judgment: describes the relationship between slot type and reference value, supporting logic such as "contains", "equal to", and "less than";

[0045] ■Reference value: The specific value corresponding to the slot type, which must be a phrase explicitly mentioned in the text.

[0046] Action: The operation to be performed after the condition is met. It must be a complete sentence in the text.

[0047] ●Format requirement: In order to ensure the uniformity and operability of the output results, the format of the tuple is <condition, action>.

[0048] In step 1, the prompt for obtaining the business step is designed as follows:

[0049] Role

[0050] You are a professor of logical reasoning, able to infer the mutual dependence and logical relationship between different things. You are good at summarizing facts and analyzing situations. You are loyal to reality and creative.

[0051] definition

[0052] Definition 1: Business text can be seen as consisting of many business rules. It is a natural language description of a business scenario or process, and is usually used to help understand and standardize business operations and management requirements in a specific field.

[0053] Definition 2: Business rules are clear instructions that describe the operating procedures, conditions, and restrictions that must be followed in various business scenarios. They are all the rules contained in the business text. We can represent these business rules in the business text in the form of a tuple, in the format of <condition, action>, where:

[0054] 1. Condition: It consists of three attributes, namely <slot type, logical judgment, reference value>, which are used to describe the conditions for triggering business rules. Among them:

[0055] Slot type: refers to the classification used to capture and store user input information in the dialogue system. Each slot type corresponds to a specific information category. Requirements: The slot type must be a noun that exists in the business text. The slot cannot be a verb.

[0056] Logical judgment: describes the logical relationship between the slot type and the reference value. The value range is ["includes", "equal to", "less than", "greater than", "less than or equal to", "greater than or equal to"].

[0057] Reference value: refers to the specific data corresponding to the slot type in the business text.

[0058] The reference value must not be repeated. The reference value type is:

[0059] Enumeration type: used for specific enumeration values, such as currency types including USD, RMB, EUR, etc. Customer types include domestic individuals and overseas individuals.

[0060] Numeric type: used for numerical comparison. For example, the reference value of the amount is 50,000 US dollars, and the logical judgment of the amount is greater than or less than.

[0061] 2. Action: The specific business operation that needs to be performed after the condition is met. Requirements: The action must be a sentence that exists in the business text. Please do not generate sentences that are irrelevant to the business text.

[0062] implement

[0063] You need to analyze, reason and execute strictly according to the definition.

[0064] Next, I will provide you with a piece of business text input. You need to extract the business rule tuples contained in the business text I provide and send it to me in the corresponding output format in the example.

[0065] Input:

[0066] Step 1.2: Optimize Prompt with Contextual Learning Technology

[0067] Although large language models have demonstrated strong natural language understanding and generation capabilities through large-scale pre-training, they may show limitations when dealing with business rules in specific fields. This is because these models are mainly exposed to general data rather than business rules in specific fields during the pre-training process. Therefore, when facing specific business scenarios, large models may not be able to fully and accurately extract business rules, nor can they generate expected outputs under complex logic.

[0068] To solve this problem, the present invention introduces contextual learning technology. Contextual learning is a technology widely used in natural language processing tasks. Its core is to help the model better understand the background and expected output of the task by providing specific task examples for the large model, thereby improving execution efficiency and accuracy. In some cases, even a powerful pre-trained language model may have difficulty correctly interpreting individual task instructions (Prompt), resulting in results that deviate from expectations. Through contextual learning, the model's adaptability to specific business rule extraction tasks can be significantly improved.

[0069] The implementation of contextual learning is very intuitive, that is, adding examples based on the original prompt. These examples provide clear task guidance for the model by showing the mapping relationship between input and output, enhancing its ability to understand business logic. For example, the following is a few-shot example designed for the business rule extraction task:

[0070] Few Shot:

[0071] Input:

[0072] When customers purchase foreign exchange, our bank supports up to 39 currencies of popular countries or regions around the world, including RMB, USD, JPY, GBP, HKD, etc. After selecting the appropriate currency, customers need to select the corresponding banknote and remittance type according to the subsequent business types. Banknote and remittance types are divided into cash and remittance. Please select "cash" when withdrawing foreign currency, and select "remittance" when remitting foreign exchange. After completing the selection of banknote and remittance type, customers need to provide the purchase amount. If the purchase amount is less than or equal to the facilitated purchase amount of USD 50,000, customers can directly apply for the purchase with their valid ID card. If the purchase amount is greater than the facilitated purchase amount of USD 50,000, customers can apply for the purchase with their valid ID card and relevant supporting documents. After entering the purchase amount, customers need to select their customer type, i.e. domestic individual or overseas individual. If they are domestic individual customers, they also need to select the purpose of the purchase. Common uses of foreign exchange for domestic individuals include outbound travel, studying abroad and other current account businesses; if they are overseas individual customers, they need to select the source of the purchase. Common sources of foreign exchange for overseas individuals include original currency exchange, domestic legal income, etc.

[0073] Output:

[0074] <<Currency, including RMB, USD, JPY, GBP, HKD, etc. up to 39 types>, select the corresponding currency type>

[0075] <<Cash and exchange type, including cash and exchange>, enter the purchase amount>

[0076] <<Amount of foreign exchange purchase, less than or equal to, 50,000 US dollars>, directly apply with valid ID card>

[0077] <<Foreign exchange purchase amount, greater than, 50,000 US dollars>, apply for foreign exchange purchase business with valid identity documents and relevant certification materials>

[0078] <<Customer Type, equal to, domestic individual>, select the purpose of foreign exchange purchase>

[0079] <<Customer Type, equal to, Overseas Individual>, select the source of foreign exchange purchase>

[0080] <<Purpose of foreign exchange purchase, including, outbound tourism, studying abroad>, None>

[0081] <<Source of foreign exchange purchase, including original currency exchange and domestic legal income>, None>

[0082] In actual applications, business text is input into the big model and processed in combination with the Prompt designed in the previous article. By splicing the business text with the optimized Prompt, the model can accurately extract business rules and generate corresponding rule tuples. The input form is shown below.

[0083] Role

[0084] You are a professor of logical reasoning, able to infer the mutual dependence and logical relationship between different things. You are good at summarizing facts and analyzing situations. You are loyal to reality and creative.

[0085] definition

[0086] Definition 1: Business text can be seen as consisting of many business rules. It is a natural language description of a business scenario or process, and is usually used to help understand and standardize business operations and management requirements in a specific field.

[0087] Definition 2: Business rules are clear instructions that describe the operating procedures, conditions, and restrictions that must be followed in various business scenarios. They are all the rules contained in the business text. We can represent these business rules in the business text in the form of a tuple, in the format of <condition, action>, where:

[0088] 1. Condition: It consists of three attributes, namely <slot type, logical judgment, reference value>, which are used to describe the conditions for triggering business rules. Among them:

[0089] Slot type: refers to the classification used to capture and store user input information in the dialogue system. Each slot type corresponds to a specific information category. Requirements: The slot type must be a noun that exists in the business text. The slot cannot be a verb.

[0090] Logical judgment: describes the logical relationship between the slot type and the reference value. The value range is ["includes", "equal to", "less than", "greater than", "less than or equal to", "greater than or equal to"].

[0091] Reference value: refers to the specific data in the business text that corresponds to the slot type. Requirements: The reference value must be a phrase that exists in the business text and the reference value cannot be repeated. The types of reference values ​​are:

[0092] Enumeration type: used for specific enumeration values, such as currency types including USD, RMB, EUR, etc. Customer types include domestic individuals and overseas individuals.

[0093] Numeric type: used for numerical comparison. For example, the reference value of the amount is 50,000 US dollars, and the logical judgment of the amount is greater than or less than.

[0094] 2. Action: The specific business operation that needs to be performed after the condition is met. Requirements: The action must be a sentence that exists in the business text. Please do not generate sentences that are irrelevant to the business text.

[0095] fewshots

[0096] Input:

[0097] When customers purchase foreign exchange, our bank supports up to 39 currencies of popular countries or regions around the world, including RMB, USD, JPY, GBP, HKD, etc. After selecting the appropriate currency, customers need to select the corresponding banknote and remittance type according to the type of business to be handled later. Banknote and remittance types are divided into cash and remittance. Please select "cash" when withdrawing foreign currency, and select "remittance" when remitting foreign exchange. After completing the selection of banknote and remittance type, customers need to provide the purchase amount. If the purchase amount is less than or equal to the facilitated purchase amount of USD 50,000, customers can directly apply for it with their valid ID card. If the purchase amount is greater than the facilitated purchase amount of USD 50,000, customers can apply for the purchase of foreign exchange with their valid ID card and relevant supporting documents. After entering the purchase amount, customers need to select their customer type, i.e. domestic individual or overseas individual. If they are domestic individual customers, they also need to select the purpose of the purchase of foreign exchange. Common uses of foreign exchange for domestic individuals include outbound travel, studying abroad and other current account businesses; if they are overseas individual customers, they need to select the source of the purchase of foreign exchange. Common sources of foreign exchange for overseas individuals include original currency exchange, domestic legal income, etc. Output:

[0098] <<Currency, including RMB, USD, JPY, GBP, HKD, etc. up to 39 types>, select the corresponding currency type>

[0099] <<Cash and exchange type, including cash and exchange>, enter the purchase amount>

[0100] <<Amount of foreign exchange purchase, less than or equal to, 50,000 US dollars>, directly apply with valid ID card>

[0101] <<Foreign exchange purchase amount, greater than, 50,000 US dollars>, apply for foreign exchange purchase business with valid identity documents and relevant certification materials>

[0102] <<Customer Type, equal to, domestic individual>, select the purpose of foreign exchange purchase>

[0103] <<Customer Type, equal to, Overseas Individual>, select the source of foreign exchange purchase>

[0104] <<Purpose of foreign exchange purchase, including, outbound tourism, studying abroad>, None>

[0105] <<Source of foreign exchange purchase, including original currency exchange and domestic legal income>, None>

[0106] ##implement

[0107] You need to analyze, reason and execute strictly according to the definition.

[0108] Next, I will provide you with a piece of business text. You need to extract the business rule tuples contained in the business text and send it to me in the corresponding output format in the example.

[0109] Step 2: Identify business rule relationships

[0110] After the initial extraction of business rules is completed, the second step of the present invention is to analyze the dependency between conditions and actions and identify the parallel relationship between business rules by constructing an association judgment prompt for the extracted business rules. This step includes two main links: association extraction of conditions and actions, many-to-one relationship judgment, and construction of parallel relationships.

[0111] Step 2.1: Extracting the association between conditions and actions

[0112] In this step, the conditions and actions in the business rules are separated to form independent nodes, and the combination rules of all conditions and actions are generated through the condition and action combination generator Prompt.

[0113] Extract conditions with an action value of "None": Extract condition nodes that are not associated with specific actions from business rules.

[0114] Extract actions whose action value is not "None": Extract nodes that have clearly pointed to specific execution actions as action nodes.

[0115] Build a condition and action combination generator prompt: By building a combination generation prompt, guide the large model to parse the input rule set and generate combination rules for all conditions and actions.

[0116] Prompt example:

[0117] You are an intelligent combination generator, and you provide a set of rules of conditions and actions. Each element in the rule set is in the format of <condition, action>, where when the action is None, it means that the current condition is not bound to an action. When there is an action value in the rule set, you need to combine all conditions with the action value of None to generate a new rule set. When the action is not None, keep its current condition. The final output should contain all the combined rules.

[0118] Few-shot example:

[0119] Input: Rule set:

[0120] <Condition 1, None>

[0121] <Condition 2, None>

[0122] <Condition 3, Action A>

[0123] <Condition 4, Action B>

[0124] Output:

[0125] <Condition 1, Action A>

[0126] <Condition 2, Action A>

[0127] <Condition 1, Action B>

[0128] <Condition 2, Action B>

[0129] <Condition 3, Action A>

[0130] <Condition 4, Action B>

[0131] Input:

[0132] Ruleset: {user input ruleset}

[0133] Output:

[0134] The combined rule set:

[0135] Step 2.2: Many-to-one relationship judgment and parallel relationship construction

[0136] In this step, by constructing an association judgment prompt, the many-to-one association relationship between conditions and actions in the business rules is identified, and the parallel relationship is further extracted.

[0137] Extract the relationship between conditions and actions: Use the association judgment prompt to analyze the execution logic between conditions and actions in business rules. For multiple conditions associated with the same action, these conditions are defined as parallel relationships, that is, all conditions must be met at the same time to trigger the corresponding action.

[0138] Identify the many-to-one relationship between conditions and actions: Further determine whether an action can be associated with multiple conditions. By clarifying the many-to-one relationship, provide related data for subsequent process construction.

[0139] Build association judgment prompts: Build association judgment prompts to guide the large model to parse business rules, determine the specific association between conditions and actions, and generate business rules including parallel relationships. This process is used to build parallel structures in the flow chart to ensure the accuracy and completeness of the process logic.

[0140] Prompt for judging the relationship between two tuples:

[0141] You are an intelligent text analyzer, focusing on extracting associations from bigrams and texts. Based on the input bigram set and text content, you need to determine whether there is a logical association between the bigrams. If the association is established, output the associated bigram; if not, do not output it.

[0142] Specific requirements:

[0143] 1. Input:

[0144] Two-tuple set: multiple two-tuples in the format of <condition, action>.

[0145] Text description: A piece of business logic text, including possible logical relationships in the tuple.

[0146] 2. Task objective: Pair related tuples by analyzing the implicit logic in the text.

[0147] 3. Output format: The combined output of the established association relationship is:

[0148] <Condition 1, Condition 2, Action>

[0149] Few-shot example:

[0150] Input:

[0151] Two-tuple:

[0152] <<Admission letter, including admission letter from prestigious schools and ordinary colleges>, provide admission letter>

[0153] <<Study Abroad Country, Equivalent to, United States or United Kingdom>, Provide Admission Letter>

[0154] <<Admission letters, including admission letters from prestigious universities and ordinary colleges>, can apply for priority approval>

[0155] <<Study abroad in the United States or the United Kingdom>, you can apply for priority approval>

[0156] <<Admission letter, including admission letter from prestigious schools and ordinary colleges>, waiver of guarantee requirement>

[0157] <<Studying Country, Equivalent to, United States or United Kingdom>, No Guarantee Requirement>

[0158] <<Admission letter, including admission letter from prestigious schools and ordinary colleges>, provide proof of assets such as real estate>

[0159] <<Study abroad, equal to, the United States or the United Kingdom>, provide proof of assets such as real estate>

[0160] <<Study abroad in the United States, the United Kingdom, and Australia>, provide an admission letter>

[0161] <<Admission letter, equivalent to admission letter from a prestigious school>, can apply for priority approval>

[0162] <<Family annual income greater than or equal to 500,000 yuan>, no guarantee requirement>

[0163] <<Annual family income, less than, 500,000 yuan>, provide proof of real estate and other assets>

[0164] text:

[0165] When applying for a loan to study abroad, customers first need to choose a country to study in, including the United States, the United Kingdom, Australia, etc. After choosing a country to study in, customers need to provide an admission letter, which can be divided into admission letters from prestigious schools and admission letters from ordinary schools. If the country to study in is the United States or the United Kingdom, and you have obtained an admission letter from a prestigious school, you can apply for priority approval. Next, customers need to provide proof of family income. If the family's annual income is greater than or equal to 500,000 yuan, the guarantee requirement can be waived; if the family's annual income is less than 500,000 yuan, proof of assets such as real estate is required. Finally, customers need to choose the loan amount and term.

[0166] Output:

[0167] <<Study Abroad Country, Equivalent to, United States or United Kingdom>, <Admission Notice, Equivalent to, Admission Notice from a Prestigious University>, Can Apply for Priority Approval>

[0168] Input:

[0169] Two-tuple: {user input two-tuple}

[0170] Text: {user input text}

[0171] Output:

[0172] Output the combination of established associations:

[0173] Step 3: Business process building

[0174] In the business process construction stage, the present invention focuses on three basic structures: sequential structure, selection structure and parallel structure. As the core of process construction, the sequential structure reflects the characteristics of linear execution of various activities in a given order. In contrast, the selection structure plays a decision-making role in the business process, selectively executing different paths or activities according to the output of specific conditions or decision points. The parallel structure plays an important role in the business process, allowing multiple activities to proceed simultaneously without waiting for the completion of other activities. Reasonable design of parallel structure can improve the efficiency and flexibility of the process when processing multiple independent tasks. At this stage, we first integrate the output results of the first two modules to obtain results containing parallel and selection structures, where the selection structure can be directly extracted from the output of the first module. Then, using the dependency prompt (Prompt), the sequential relationship between business rules is determined for the integrated results according to the action value. Finally, a complete process model is constructed based on these relationships. The prompt design for business process construction is as follows:

[0175] Role

[0176] You are a business rule dependency analysis expert who is good at extracting and analyzing business rules from business text, identifying dependencies between slot types, and outputting the results in a standardized format.

[0177] definition

[0178] Definition 1: Business rules are clear instructions that guide the operating procedures, conditions, and restrictions that must be followed in a specific business scenario. In a dialogue system, business rules can be expressed as a tuple in the format of <condition, action>, where:

[0179] Condition: It consists of three attributes, namely <slot type, logical judgment, reference value>, which is used to describe the conditions for triggering business rules.

[0180] Slot type: A category used in the dialog system to capture and store user input information. Each slot type corresponds to a specific information category. Requirements: The slot type must be a noun that exists in the business text. The slot cannot be a verb.

[0181] Logical judgment: describes the logical relationship between the slot type and the reference value. The value range is ["includes", "equal to", "less than", "greater than", "less than or equal to", "greater than or equal to"].

[0182] Reference value: refers to the specific data in the business text that corresponds to the slot type. Requirements: The reference value must be a phrase that exists in the business text and the reference value cannot be repeated. The types of reference values ​​are:

[0183] Enumeration type: used for specific enumeration values, such as currency types including USD, RMB, EUR, etc. Customer types include domestic individuals and overseas individuals.

[0184] Numeric type: used for numerical comparison. For example, the reference value of the amount is 50,000 US dollars, and the logical judgment of the amount is greater than or less than.

[0185] Action: The specific business operation that needs to be performed after the condition is met. Requirements: The action must be a sentence that exists in the business text. Please do not generate sentences that are irrelevant to the business text.

[0186] Definition 2: Dependency Definition

[0187] Dependency relationships are used to describe the order or selection relationship between slot types, and are divided into the following structures:

[0188] Sequential structure: The selection of one slot type depends on the completion of the previous slot type.

[0189] Selection structure: Determine the slot type to be selected later based on the value of a slot type.

[0190] Parallel structure: Multiple slot types can be selected at the same time without dependencies between each other.

[0191] The dependency relationship between slot types is stored in a list format, in the following format: [id, slot, depend_id, depend_value]

[0192] id: The serial number of the slot type, starting from 0.

[0193] slot: slot type name.

[0194] depend_id: A list storing the ids of the dependent slot types.

[0195] depend_value: Only used in selection structures, indicates a specific value that depends on the slot type.

[0196] Steps:

[0197] 1. Parse business text: Identify the slot types involved in the business process and their order or selection relationship.

[0198] 2. Parse business rules: Understand the conditions and actions in each business rule and identify the dependencies between slots.

[0199] 3. Build a dependency list: Assign a unique id to each slot type, starting from 0 in order of appearance. Determine the preceding slot type (depend_id) that each slot type depends on based on business processes and business rules. For selection structures, specify depend_value to indicate dependencies under specific conditions.

[0200] 4. Output formatted results: Output the dependency relationship in the specified list format. The format is as follows: [id, slot, depend_id, depend_value]

[0201] Example:

[0202] Input:

[0203] When applying for a loan to study abroad, customers first need to choose a country to study in, including the United States, the United Kingdom, Australia, etc. After choosing a country to study in, customers need to provide an admission letter, which can be divided into admission letters from prestigious schools and admission letters from ordinary schools. If the country to study in is the United States or the United Kingdom, and you have obtained an admission letter from a prestigious school, you can apply for priority approval. Next, customers need to provide proof of family income. If the family's annual income is greater than or equal to 500,000 yuan, the guarantee requirement can be waived; if the family's annual income is less than 500,000 yuan, proof of assets such as real estate is required. Finally, customers need to choose the loan amount and term.

[0204] Business rules:

[0205] <<Study Abroad Country, Equivalent to, United States or United Kingdom>, <Admission Notice, Equivalent to, Admission Notice from a Prestigious University>, Can Apply for Priority Approval>

[0206] <<Study abroad in the United States, the United Kingdom, and Australia>, provide an admission letter>

[0207] <<Admission letter, including, admission letter from prestigious schools, admission letter from ordinary schools>, None>

[0208] <<Admission letter, equivalent to admission letter from a prestigious school>, can apply for priority approval>

[0209] <<Study Abroad Country, equals, the United States or the United Kingdom>, None>

[0210] <<Family annual income greater than or equal to 500,000 yuan>, no guarantee requirement>

[0211] <<Annual family income, less than, 500,000 yuan>, provide proof of real estate and other assets>

[0212] Output:

[0213] [[0,Study Abroad Country,[],],

[0214] [1,Admission letter,[],],

[0215] [2, annual family income, [0, 1], US or UK; admission letter from a prestigious university],]

[0216] After the implementation of the above three steps, the business rule mining and business process construction method based on prompt learning can be effectively realized, thus providing strong support for the optimization and automation of business processes.

Claims

1. A method for business rule mining and business process construction based on prompt learning, characterized in that: The following steps are involved: S1: First, construct a prompt for business rule mining, obtain the text of the business rule, and extract the business rule in the form of a tuple from the text of the business rule by combining the prompt for business rule mining with a large language model. The tuple format is <condition, action>, where the condition is used to describe the precondition for triggering the business rule, and the action defines the execution operation associated with the condition; S2: Input the bigram business rules and associated judgment prompts extracted in step S1 into the large language model, and use the large language model to judge and analyze the association relationship between the conditions and actions in the bigram. If there is an association relationship, the requirements of the parallel structure are met, and the bigram business rules that meet the requirements of the parallel structure are used as business rules that meet the parallel structure, and then proceed to step S3; S3: using dependency hints and context learning technology, analyzing the dependency between the actions in the binary business rules obtained in step S1 and the business rules satisfying the parallel structure obtained in step S2, and generating a business process model; S4: Apply the business process model to financial services equipment.

2. The method for business rule mining and business process construction based on prompt learning according to claim 1, characterized in that: In step S1, a prompt for business rule mining is first constructed to obtain the text of the business rule. The business rule mining prompt is combined with a large language model to extract the bigram business rule from the text of the business rule, specifically including: (1.1) Through conceptual definition, the core elements of business rules, including conditions and actions, are clarified, and then the prompts for business rule mining are constructed. The text of business rules is parsed through the optimized large language model to obtain business rules in the form of tuples.

3. The method for business rule mining and business process construction based on prompt learning according to claim 2, characterized in that: In step (1.1), the conditions include slot type, logical judgment and reference value.

4. The method for business rule mining and business process construction based on prompt learning according to claim 3 is characterized in that: In step (1.1), the slot type includes currency and customer type.

5. The method for business rule mining and business process construction based on prompt learning according to claim 3 is characterized in that: In step (1.1), the logic judgment is to describe the logical relationship between the slot type and the reference value.

6. The method for business rule mining and business process construction based on prompt learning according to claim 3, characterized in that: The reference value refers to the correspondence between the text used to describe the slot type and the business rule, and the reference value includes two types: digital type and enumeration type.

7. The method for business rule mining and business process construction based on prompt learning according to claim 1, characterized in that: In step S2, the business rules in the form of bigrams and the associated judgment prompts extracted in step S1 are input into the large language model, and the association relationship between the conditions and actions in the bigrams is judged and analyzed by the large language model, specifically including: (2.1) Separate the conditions and actions in the binary business rules to form independent nodes, and generate all possible condition and action combination rules through the association between conditions and actions; (2.2) Construct an association judgment prompt, input the association judgment prompt and all possible condition and action combination rules obtained in step (2.1) into the large language model, and identify the association relationship between the condition and the action.

8. The method for business rule mining and business process construction based on prompt learning according to claim 1, characterized in that: In step S3, the dependency relationship between the actions refers to the execution order of the actions.

Images