Business process compliance checking method, terminal and platform based on natural language processing

Abstract

The application discloses a business process compliance checking method, a terminal and a platform based on natural language processing. The method can extract semantic information from regulatory documents and perform compliance checking on a business process model from multiple angles. This method can evaluate the existence of non-compliant patterns in the process model during the design phase, helping enterprises detect violations at an early stage and prevent potential violations during subsequent process execution, thereby saving the enterprise's energy, time and financial resources.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of business process compliance inspection, in particular to a business process compliance inspection method based on natural language processing, a terminal and a platform. BACKGROUND

[0002] Business Process Management (BPM), namely business process management, is a comprehensive management mode to achieve the integration of various business links of an enterprise. Processes are everywhere. There are processes for R&D, production, planning, sales, service, personnel, and finance. In fact, various management systems applied in enterprises contain a large amount of business process management work.

[0003] Business Process Compliance (BPC) is an important part of business process management, which measures whether the organization's business processes comply with all relevant laws, regulations, guidelines and standards. If a company's business processes violate these regulations, the company may lose the trust of investors, even face fines and criminal charges. Therefore, it is essential for every organization to comply with rules from different sources to avoid huge penalty losses and improve business process transparency. However, in the practice of companies, checking and ensuring the consistency of the organization's business processes with regulatory documents is largely done manually, which greatly consumes human resources. At present, there are also some ways to automatically implement business process compliance inspection, for example, in the invention patent with application number CN202111083654.1, a cross-organizational business process model mining and compliance inspection method and system is disclosed, which can mine cross-organizational business process models in various scenarios, breaking through the problem that existing compliance inspection methods cannot effectively measure the mined cross-organizational business process models; and in the invention patent with application number CN202210920275.1, a business process compliance inspection method, computer equipment and readable storage medium are disclosed, based on the decision tree to judge whether the trajectory is compliant and prune it, so that it can more efficiently realize the trajectory compliance inspection. However, in these prior arts, the semantic information in the process model and the regulatory document cannot be automatically extracted, so there are still limitations in actual application. SUMMARY

[0004] In order to overcome the shortcomings of the prior art, the present application provides a business process compliance inspection method based on natural language processing, a terminal and a platform, which can effectively solve the above problems.

[0005] The technical solutions specifically adopted by the present application are as follows:

[0006] In a first aspect, the present invention provides a business process compliance check method based on natural language processing, which includes the following steps:

[0007] S1: Input process model library ∑= <m1,m2,…,m n > Extract semantic information of activity and event labels from each process model, where ∑ is composed of n process models m=(A m E m G m ,R m N m ,F m Composed of (,u,ρ), where A m Representative activities collection, E m G represents a set of events. m R represents a set of gateways. m N represents the set of executors. m =A m ∪E m ∪G m F represents the set of all nodes. m Represented by N m The set of ordered relations consisting of directed edges f of all nodes in the set, u represents the mapping relationship between activities and events to their operations and business objects, and ρ represents the mapping relationship between executors and activities and events.

[0008] S2: Obtain a pre-built business process compliance check rule base R, which consists of all rule records extracted from the business process supervision documents; each rule record in the rule base R is represented as r = (t r A r ,P r C r O r E e U r ,f r ), where t r This indicates the statement type classification result of the rule statement sent from which the rule record originates. Statement types include four categories: mandatory, prohibited, permitted, and definitional; A r p r C r O r and E r These represent the five semantic components extracted from the rule statement `sent`: operation, executor, condition, constraint, and exception; U r f represents the sequential relationship between an operation and a condition or an operation and a constraint. r It indicates the execution relationship between executors and operations, that is, it specifies which executors must perform which operations;

[0009] S3: Calculate the matching score between each process model to be inspected and all rule records. Set a matching score threshold δ. When the matching score between a rule record and a process model is greater than δ, the rule record matches the process model. The formula for calculating the matching score between any process model m∈∑ and a rule record r∈R is as follows:

[0010]

[0011] in Operation A in rule record r represents the operation. r Operations with process model m (act(A)) m ∪E m The set of operations formed by pairing elements together, || represents counting the number of elements in the set, (a r ,a m )∈D r,m Represents a m and a r It is the pair of matching operations with the highest similarity between the process model m to be checked and the rule record r, and a m Corresponding to the operation in process model m, a r Corresponding to the operation in rule record r; O r,m ∈P r ×bs_obj(A m ∪E m )∪R m The executor P in rule record r represents the executor. r With set bs_obj(A m ∪E m )∪R m The set of executor pairs in the executor list, bs_obj(A m ∪E m )∪R m For the business object bs_obj(A) of process model m m ∪E m ) and the set of executors R m The set composed of (o r ,o m )∈O r,m Represents o m and o r It is the pair of matching executors with the highest similarity between the process model m and the rule record r to be checked; sim(S1,S2) represents the similarity between two texts S1 and S2, and τ represents the minimum similarity threshold included in the summation term. If sim(a r ,a m If ) > τ holds, then sim(a) r ,a m The overall value of τ is sim(a)r ,a m Otherwise, sim(a) r ,a m The overall value of τ is set to 0;

[0012] S4. Calculate the violation score between the process model m to be inspected and its matching rule record. The violation score includes three categories: activity missing violation score, executor error violation score, and execution order error violation score.

[0013] The formula for calculating the activity missing violation score between process model m and rule record r is as follows:

[0014]

[0015] Where γ∈(0,1) represents the preset operation similarity threshold, {(a r ,a m )∈D r,m |sim(a r ,a m )<γ} represents D r,m The similarity between the two values ​​satisfies sim(a) r ,a m All matching operation pairs where ) < γ;

[0016] The formula for calculating the executor error violation score between process model m and rule record r is as follows:

[0017]

[0018] Parameter R r,m,γ and C r,m,γ The calculation formulas are as follows:

[0019]

[0020] Where θ∈(0,1) represents the preset executor similarity threshold; the symbol “∧” represents and; f r (g) represents the execution relationship between the executor and the operation in rule record r, f m (r) represents the execution relationship between executors and operations in process model m;

[0021] The formula for calculating the execution order error violation score between process model m and rule record r is as follows:

[0022]

[0023] Where γ∈(0,1) represents the preset operation violation similarity threshold;

[0024] (u r ,u′r )∈U r U r The order relationship in, and (u r ,u m ) and (u′ r ,u′ m Both (u) are the pair of matching operations with the highest similarity in the process model m and the rule record r. r ,u m )∈D r,m And(u′ r ,u′ m )∈D r,m ;

[0025] S5. Based on the activity missing violation score, executor error violation score, and execution order error violation score calculated in S4, and combined with the preset violation judgment rules, output the compliance check results of the process model to be checked.

[0026] As a preferred embodiment of the first aspect above, step S1, extracting the semantic information of activity and event labels in each process model, specifically includes the following sub-steps:

[0027] S11: For each process model m∈∑, the activity set A m and event set E m Union A m ∪E m The tags for activities and events are extracted and part-of-speech tagging is performed. The words in the tags are categorized by part of speech into verbs, adjectives, nouns, keywords, pronouns, prepositions, adverbs, conjunctions, auxiliary words, determiners, and numerals. Words tagged as verbs are extracted as the operation acts(A) of the process model m. m ∪E m );

[0028] S12: The union A of each process model m∈∑ m ∪E m Dependency parsing was performed on the tagged text of activities and events in S11. A dependency relation in the text connects two words, and the corresponding word is extracted from the text. m ∪E m Words with subject-predicate or verb-object relationships are used as business objects (bs_obj(A)) in the process model m. m ∪E m );

[0029] S13: The union A for each process model m m ∪E m Establish its relationship with the operation act(A) extracted from S11. m ∪E mThe business object bs_obj(A) extracted from S12 and S12 m ∪E m The mapping relationship u between A and B is stored in the process model m, where u is represented as A. m ∪E m →{act(A m ∪E m ),bs_obj(A m ∪E m )}.

[0030] As a preferred embodiment of the first aspect mentioned above, the dependency relationships are divided into subject-predicate relationships, verb-object relationships, indirect object relationships, prepositional object relationships, pivotal objects, attributive-head relationships, adverbial-head structures, verb-complement structures, parallel relationships, prepositional object relationships, left-additive relationships, right-additive relationships, independent structures, and core relationships.

[0031] As a preferred embodiment of the first aspect above, in S11, the part-of-speech tagging tool is selected from the natural language analysis tool Stanford NLP.

[0032] As a preferred embodiment of the first aspect above, in S12, the dependency parsing tool is selected as the natural language analysis tool Stanford NLP.

[0033] As a preferred embodiment of the first aspect above, in S3, the matching score threshold δ is set to 0.5.

[0034] As a preferred embodiment of the first aspect above, the operation similarity threshold γ is set to 0.8.

[0035] As a preferred embodiment of the first aspect above, the executor similarity threshold θ is set to 0.7.

[0036] As a preferred embodiment of the first aspect above, in S5, the violation determination rule is a first score threshold range, a second score threshold range, and a third score threshold range preset for violation scores of missing activity, incorrect executor, and incorrect execution order. When the score of a certain violation deviates from the corresponding score threshold range, it is considered that the violation has a non-compliance problem. The violation includes three types: missing activity, incorrect executor, and incorrect execution order.

[0037] In a second aspect, the present invention provides a smart terminal, which includes a memory and a processor;

[0038] The memory is used to store computer programs;

[0039] The processor is configured to, when executing the computer program, implement the business process compliance check method based on natural language processing as described in any of the solutions of the first aspect above.

[0040] As a preferred embodiment of the second aspect mentioned above, the smart terminal is either a local terminal or a cloud terminal.

[0041] Thirdly, the present invention provides a business platform, which includes a memory and a processor;

[0042] The memory is used to store computer programs;

[0043] The processor is configured to, when executing the computer program, perform a business process compliance check method based on natural language processing as described in any of the solutions in the first aspect above, for business processes that are to be run or are currently running on the platform.

[0044] Compared with traditional business process compliance inspection methods, this invention has the following advantages: 1. This invention automatically extracts semantic information from process models and regulatory documents, greatly reducing manual costs; 2. This invention uses deep learning and rule-based traditional natural language processing methods to extract semantics from the text of regulatory documents, improving accuracy and providing more detailed violation information. Attached Figure Description

[0045] Figure 1 This is a flowchart of the method of the present invention;

[0046] Figure 2 This is a diagram showing the results of missing activities and incorrect ordering in an embodiment of the present invention.

[0047] Figure 3 This is a diagram showing the result of an executor error in an embodiment of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0049] Conversely, this invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of the invention as defined in the claims. Furthermore, to provide a better understanding of the invention, certain specific details are described in detail below. However, those skilled in the art will fully understand the invention even without these detailed descriptions.

[0050] In a preferred embodiment of the present invention, a business process compliance check method based on natural language processing is provided, the compliance basis of which is derived from business process regulatory documents. These business process regulatory documents refer to the relevant laws, regulations, guidelines, and standards that business processes must comply with. For example, for insurance processes, examples of such documents include the Health Insurance Portability and Accountability Act (HIPAA), the Sarbanes-Oxley Act (SOX), and the General Data Protection Regulation (GDPR). If a company's business processes violate these regulations, the company may lose the trust of investors, even face fines and criminal charges. Therefore, compliance with rules from various sources is essential for every organization to avoid huge fines and improve the transparency of its business processes.

[0051] like Figure 1 As shown, in an embodiment of the present invention, the business process compliance check method based on natural language processing specifically includes the following steps:

[0052] S1: Input process model library ∑= <m1,m2,…,m n > Extract semantic information of activity and event labels from each process model, where ∑ is composed of n process models m=(A m E m G m ,R m N m ,F m Composed of (,u,ρ), where A m Representative activities collection, E m G represents a set of events. m Represents a set of gateways (specifically divided into exclusive gateways and parallel gateways), R m N represents the set of executors. m =A m ∪E m ∪G m F represents the set of all nodes. m Represented by N m The set of ordered relations consisting of directed edges f of all nodes in the set, u represents the mapping relationship between activities and events to their operations and business objects, ρ∶R m →A m ∪E m This represents the mapping relationship between the executor and the activities and events.

[0053] In an embodiment of the present invention, step S1 above, extracting the semantic information of activity and event tags in each process model, specifically includes the following sub-steps:

[0054] S11: For each process model m∈∑, the activity set A m and event set Em Union A m ∪E m The tags for activities and events are extracted and part-of-speech tagging is performed. The words in the tags are categorized by part of speech into verbs, adjectives, nouns, keywords, pronouns, prepositions, adverbs, conjunctions, auxiliary words, determiners, and numerals. Words tagged as verbs are extracted as the operation acts(A) of the process model m. m ∪E m );

[0055] S12: The union A of each process model m∈∑ m ∪E m Dependency parsing was performed on the tagged text of activities and events in S11. A dependency relation in the text connects two words, and the corresponding word is extracted from the text. m ∪E m Words with subject-predicate or verb-object relationships are used as business objects (bs_obj(A)) in the process model m. m ∪E m );

[0056] S13: The union A for each process model m m ∪E m Establish its relationship with the operation act(A) extracted from S11. m ∪E m The business object bs_obj(A) extracted from S12 and S12 m ∪E m The mapping relationship u between A and B is stored in the process model m, where u is represented as A. m ∪E m →{act(A m ∪E m ),bs_obj(A m ∪E m )}.

[0057] In embodiments of the present invention, the above-mentioned dependency relationships can be divided into subject-predicate relationship, verb-object relationship, indirect object relationship, prepositional object, pivotal object, attributive-head relationship, adverbial-head structure, verb-complement structure, parallel relationship, prepositional object relationship, left-additive relationship, right-additive relationship, independent structure, and core relationship.

[0058] In embodiments of the present invention, the part-of-speech tagging tool can be selected from the natural language processing tool Stanford NLP. The dependency parsing tool can also be selected from the natural language processing tool Stanford NLP.

[0059] S2: Obtain a pre-built business process compliance check rule base R, which consists of all rule records extracted from the business process supervision documents; each rule record in the rule base R is represented as r = (tr A r ,P r C r O r E r U r ,f r ), where t r This indicates the statement type classification result of the rule statement sent from which the rule record originates. Statement types include four categories: mandatory, prohibited, permitted, and definitional; A r P r C r O r and E r These represent the five semantic components extracted from the rule statement `sent`: operation, executor, condition, constraint, and exception; U r f represents the sequential relationship between an operation and a condition or an operation and a constraint. r It indicates the execution relationship between the executor and the operation, that is, it specifies which executors must perform which operations.

[0060] It should be noted that the aforementioned business process compliance check rule base R can be an existing rule base that has already been built. It can be built manually or automatically using machine learning and natural language processing technologies. The following is a brief introduction to the process of automatically building the aforementioned business process compliance check rule base R, including the following steps:

[0061] S21: Input a regulatory document library containing the target regulatory documents. right The text content c of each regulatory document E is preprocessed to remove noise from the text and each regulatory document is split into independent rule statements sent.

[0062] S22: Fine-tune the BERT model on a general corpus of legal language, then connect the fine-tuned BERT model with a classifier to form a rule-based sentence classification model; train the rule-based sentence classification model to identify the sentence type of each rule sentence, and obtain the classification result t for each rule sentence sentence. r The classification results labels for the statement types include mandatory, prohibited, permitted, and definition types.

[0063] The specific definitions of the above four statement types are as follows:

[0064] The term "mandatory" refers to a situation where something is legally required to be enforced; failure to meet this condition constitutes a violation.

[0065] The prohibited type indicates a situation where doing something is legally forbidden; if this condition is not met, a violation occurs.

[0066] The permissive type indicates that something is allowed to be done without violating the mandatory or prohibitive type; if this condition is not met, a violation may occur.

[0067] Definition-type clauses define legal terms that are conceptual.

[0068] S23: Through constituent syntactic analysis and dependency syntactic analysis, semantic constituent extraction is performed on the rule statement `sent` after the statement type has been identified, and operation A is extracted from it. r Executor P r Condition C r Constraint O r And exception E r There are a total of five semantic components, and the specific implementation is as follows:

[0069] S231: Perform text preprocessing on each rule statement sent, including word segmentation, part-of-speech tagging, and named entity recognition;

[0070] S232: Perform constituent parsing and dependency parsing on the preprocessed rule statement text using natural language processing tools to obtain constituent parsing trees T. c and T d Using query language in tree T c and T d Based on the extraction rules and keywords, five semantic components were extracted from the text, namely, Operation A. r Executor P r Condition C r Constraint O r And exception E r The specific definitions of the five semantic components are as follows:

[0071] Operations represent the specific content of the rules, that is, the mandatory, prohibited, and permitted content defined in the rules;

[0072] The executor refers to the role responsible for the operation and execution;

[0073] A condition indicates the state when a rule applies;

[0074] Constraints represent a special case specified by the rules;

[0075] An exception indicates a situation where the rule does not apply.

[0076] Furthermore, the specific steps for extracting the five semantic components from the text using the query language based on extraction rules and keywords are as follows:

[0077] S2321: Stanford Tregex is used as the query language, and the specific rule notation used is as follows:

[0078] A << B means that in the tree structure, A dominates B;

[0079] A < B means that in the tree structure, A directly dominates B;

[0080] A $ B means that in the tree structure, A and B are sister relationships;

[0081] S2322: Extract the executor P according to the following steps r :

[0082] First, design the extraction rules and extraction keywords required for the executor P r The executor extraction rules specifically include the following three types:

[0083] The first type: There is subject dependence and NP < (executor keyword)

[0084] The second type: There is object dependence and passive voice and PP < IN $ (NP < (executor keyword))

[0085] The third type: There is object dependence and active voice and NP < (executor keyword)

[0086] Among them, the executor keyword is the keyword in the rule statement used to indicate the potential position of the executor, and its dependency relationship is shown in the dependency analysis tree T d NP represents a noun phrase, PP represents a prepositional phrase, and IN represents a preposition;

[0087] Extract the noun phrase NP in the rule statement sent that satisfies any one of the executor extraction rules and use it as the executor P r ;

[0088] S2323: Extract the condition C according to the following steps r :

[0089] First, design the extraction rules and extraction keywords required for the condition C r The condition extraction rules specifically include the following two types:

[0090] SBAR << (condition keyword)

[0091] PP << (condition keyword)

[0092] Among them, the condition keyword is the keyword in the rule statement used to indicate the r potential position of the condition C, SBAR represents a clause, and PP represents a prepositional phrase;

[0093] Extract the clause SBAR or prepositional phrase PP in the rule statement sent that satisfies any one of the condition extraction rules and use it as the condition C r ;

[0094] S2324: Extract constraint O according to the following steps. r :

[0095] First, design and extract constraints O r The required extraction rules and keywords, and the specific extraction rules include the following two types:

[0096] NP < (constraint keyword)

[0097] PP<(IN<(constraint keyword))$NP

[0098] The constraint keywords are those used in the rule statement to indicate constraints O. r Keywords for potential positions: NP represents noun phrases, PP represents prepositional phrases, and IN represents prepositions;

[0099] Extract the noun phrase NP or prepositional phrase PP from the rule statement `sent` that satisfies any one of the constraint extraction rules, and use them as constraint O. r ;

[0100] S2325: Extract exception E according to the following steps r :

[0101] First, design the extraction of exception E. r The required extraction rules and keywords, and the exception extraction rules specifically include the following three types:

[0102] SBAR << (Exception Keyword)

[0103] PP << (Exception Keyword)

[0104] NP << (IN < exception keyword)

[0105] The exception keyword is used in the rule statement to indicate exception E. r Keywords for potential positions: SBAR represents clause, NP represents noun phrase, PP represents prepositional phrase, and IN represents preposition;

[0106] Extract the clauses SBAR, noun phrases NP, or prepositional phrases PP from the rule statement `sent` that satisfy any one of the constraint extraction rules, and treat them as exceptions E. r ;

[0107] S2326: Extract operation A according to the following steps. r :

[0108] First, design the extraction rules and keywords for modal verbs. The extraction rules for modal verbs specifically include the following three types:

[0109] (S <VP)<<MD

[0110] ((SINV< <S)<VP)<<MD

[0111] ((SBAR< <S)<VP)<<MD

[0112] Where S represents a simple declarative sentence, VP represents a phrasal verb, SINV represents a declarative sentence with subject inversion, MD represents a modal verb, and SBAR represents a clause;

[0113] Redesign Operation A r The extraction rules are as follows: the extraction rules are all except modal verbs (MD) and conditional verbs (C). r and constraint O r Verb phrases other than VP;

[0114] Finally, first extract the modal verbs (MD) in the rule statement `sent` that satisfy any modal verb extraction rule, and then extract the verb phrases (VP) in the rule statement `sent` that satisfy any operation extraction rule and use them as operation A. r .

[0115] S24: Classify the statement type obtained from the same rule statement sent into t. r Combined with the semantic components, a rule record r = (t) is generated. r A r ,P r C r O r E r U r ,f r ), where U r f represents the sequential relationship between an operation and a condition or an operation and a constraint. r It indicates the execution relationship between executors and operations, that is, it specifies which executors must perform which operations; the rule base R consists of all rule records and is used as the basis for compliance checks in business processes.

[0116] S3: Calculate the matching score between each process model to be inspected and all rule records. Set a matching score threshold δ. When the matching score between a rule record and a process model is greater than δ, the rule record matches the process model. The formula for calculating the matching score between any process model m∈∑ and a rule record r∈R is as follows:

[0117]

[0118] in Operation A in rule record r represents the operation. r Operations with process model m (act(A)) m ∪E mThe set of operations formed by pairing elements together, || represents counting the number of elements in the set, (a r ,a m )∈D r,m Represents a m and a r It is the pair of matching operations with the highest similarity between the process model m to be checked and the rule record r, and a m Corresponding to the operation in process model m, a r Corresponding to the operation in rule record r; O r,m ∈P r ×bs_obj(A m ∪E m )∪R m The executor P in rule record r represents the executor. r With set bs_obj(A m ∪E m )∪R m The set of executor pairs in the executor list, bs_obj(A m ∪E m )∪R m For the business object bs_obj(A) of process model m m ∪E m ) and the set of executors R m The set composed of (o r ,o m )∈O r,m Represents o m and o r It is the pair of matching executors with the highest similarity between the process model m and the rule record r to be checked; sim(S1,S2) represents the similarity between two texts S1 and S2, and τ represents the minimum similarity threshold included in the summation term. If sim(a r ,a m If ) > τ holds, then sim(a) r ,a m The overall value of τ is sim(a) r ,a m Otherwise, sim(a) r ,a m The overall value of τ is set to 0.

[0119] In an embodiment of the present invention, in S3 above, the matching score threshold δ can be optimized and adjusted according to actual conditions, preferably 0.5.

[0120] S4. Calculate the violation score between the process model m to be inspected and its matching rule record. The violation score includes three categories: activity missing violation score, executor error violation score, and execution order error violation score.

[0121] 4.1) The formula for calculating the activity missing violation score between process model m and rule record r is as follows:

[0122]

[0123] Where γ∈(0,1) represents the preset operation similarity threshold, {(a r ,a m )∈D r,m |sim(a r ,a m )<γ} represents D r,m The similarity between the two values ​​satisfies sim(a) r ,a m All matching operation pairs where ) < γ;

[0124] 4.2) The formula for calculating the executor error violation score between the process model m and the rule record r is as follows:

[0125]

[0126] Parameter R r,m,γ and C r,m,γ The calculation formulas are as follows:

[0127]

[0128] Where θ∈(0,1) represents the preset executor similarity threshold; the symbol “∧” represents and; f r (g) represents the execution relationship between the executor and the operation in rule record r, f m (r) represents the execution relationship between executors and operations in process model m;

[0129] 4.3) The formula for calculating the execution order error violation score between process model m and rule record r is as follows:

[0130]

[0131] Where γ∈(0,1) represents the preset operation violation similarity threshold;

[0132] (u r ,u′ r )∈U r U r The order relationship in, and (u r ,u m ) and (u′ r ,u′ m Both (u) are the pair of matching operations with the highest similarity in the process model m and the rule record r. r ,u m )∈Dr,m And(u′ r ,u′ m )∈D r,m .

[0133] In the embodiments of the present invention, the above-mentioned operation similarity threshold γ can be optimized and adjusted according to actual conditions, preferably 0.8, and similarly, the above-mentioned executor similarity threshold θ can be optimized and adjusted according to actual conditions, preferably 0.7.

[0134] S5. Based on the activity missing violation score, executor error violation score, and execution order error violation score calculated in S4, and combined with the preset violation judgment rules, output the compliance check results of the process model to be checked.

[0135] The aforementioned violation determination rules can be adjusted according to the actual business process violation determination criteria. In the embodiments of the present invention, the aforementioned violation determination rules are preset first, second, and third score threshold ranges for violation scores of missing activities, incorrect executor, and incorrect execution order. When the score of a certain violation deviates from the corresponding score threshold range, it is considered that the violation has a non-compliance issue. The violations include three types: missing activities, incorrect executor, and incorrect execution order. Therefore, in the final output compliance check result of the process model to be checked, there may be one or more non-compliance issues.

[0136] Similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a smart terminal corresponding to the business process compliance inspection method based on natural language processing provided in the above embodiments, which includes a memory and a processor;

[0137] The memory is used to store computer programs;

[0138] The processor is configured to implement the business process compliance check method based on natural language processing as described in the foregoing embodiments when executing the computer program.

[0139] Similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a business platform corresponding to the natural language processing-based business process compliance inspection method provided in the above embodiments, which includes a memory and a processor;

[0140] The memory is used to store computer programs;

[0141] The processor is configured to, when executing the computer program, perform the business process compliance check method based on natural language processing as described in the foregoing embodiments for business processes that are to be run or are currently running on the platform.

[0142] It is understood that the aforementioned storage medium and memory can be random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Furthermore, the storage medium can also be any medium capable of storing program code, such as a USB flash drive, external hard drive, magnetic disk, or optical disk.

[0143] It is understood that the processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0144] It is understandable that the aforementioned smart terminals can be local terminals or cloud terminals.

[0145] It should also be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here. In the embodiments provided in this application, the division of steps or modules in the device and method is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple modules or steps may be combined or integrated together, and a module or step may also be split.

[0146] The following example demonstrates the implementation and technical effects of the natural language processing-based business process compliance check method described in the foregoing embodiments.

[0147] Example

[0148] The steps in this embodiment are the same as steps S1 to S5 described above in the specific implementation method, and will not be repeated here. The following section demonstrates some of the implementation process and results:

[0149] This embodiment uses four process models as inputs, which capture how the main privacy constraints of GDPR are implemented in the process. The characteristics of the four models are shown in Table 1.

[0150] Table 1. Characteristics of the Process Model

[0151]

[0152] Furthermore, this embodiment uses Articles 1 through 50 of the GDPR as the regulatory documents for these four process models. Articles 1 through 4 contain introductory statements, Articles 5 through 50 contain regulatory statements for the four models, and Articles 51 and above apply to regulatory agencies, not corporate organizations. Based on these four process models and the regulatory texts, this embodiment manually constructs a set of process models containing violations. First, the process models are supplemented to ensure full compliance. For violations related to missing operations, if the activity exists in the model, it is deleted. For violations related to executor errors, if it is correct in the model, it is manually modified. For violations related to out-of-order execution, the positions of activities in the model are swapped. Each check contains only one violation. Table 2 shows the violation data for each model.

[0153] Table 2 shows the violation data of the model.

[0154]

[0155] This embodiment uses common metrics for recommender systems to quantify the matching accuracy in the method of this invention: the average precision (AP) of each model and the average precision (MAP) of the entire model library, which are defined as follows:

[0156]

[0157]

[0158] Regarding the accuracy of violation detection, this embodiment uses accuracy as a measure, which is defined as follows:

[0159]

[0160] Where TP represents true positives; FP represents false positives; and accuracy represents the proportion of true positives among those predicted as positives.

[0161] As shown in Table 3, the matching results show that the best results are achieved at τ = 0.8, both for individual models and for overall accuracy.

[0162] Table 3 shows the matching results at different τ values.

[0163]

[0164] Regarding the results of non-compliant detections, such as detections with missing activities or incorrect execution sequences, for example... Figure 2As shown, for most models, the accuracy is highest when γ = 0.8 for both types of violations. However, for model 4, the accuracy is highest when γ = 0.7. This is because model 4 is the most complex. When γ increases, i.e., becomes more stringent, it identifies more violations, thus misclassifying normal operations as violations and consequently reducing accuracy.

[0165] For violations such as errors by the executor, the consequences are as follows: Figure 3 As shown, in this embodiment, when γ = 0.8, changing the value of θ reveals that the first two of the four models... The highest accuracy was achieved at that time, while the other two were... The latter two models have the highest accuracy. This is because they are more complex. Furthermore, the similarity between the executor and the rule enforcer in the latter two models is reduced due to their inclusion relationship, which also leads to lower accuracy when... Accuracy decreases when the speed is increased.

[0166] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A business process compliance check method based on natural language processing, characterized in that, Includes the following steps: S1: Input Process Model Library Extract semantic information of activity and event tags from each process model, where Composed of n process models Composition, in which Representative activities collection, Represents a set of events. Represents a set of gateways. Represents the set of executors. Represents the set of all nodes. Representative by Directed edges of all nodes in The set of ordered relations, This represents the mapping relationship between activities and events and their operations and business objects. This represents the mapping relationship between the executor and the activities and events; S2: Obtain a pre-built business process compliance check rule base The rule base It consists of all rule records extracted from business process monitoring documents; rule base Each rule record in the middle is represented as ,in This indicates the rule statement from which the rule record originates. The statement type classification results show that statement types include four categories: mandatory, prohibited, permitted, and definitional. , , , and These respectively represent the rules and statements. The five semantic components extracted are operation, executor, condition, constraint, and exception; This indicates the sequential relationship between operations and conditions, or operations and constraints. It indicates the execution relationship between executors and operations, that is, it specifies which executors must perform which operations; S3: Calculate the matching score between the process model to be inspected and all rule records one by one, and set the matching score threshold. When the matching score between the rule record and the process model is greater than When the rule matches the process model, the rule record is valid; for any process model... With rules record The formula for calculating the match score between them is as follows: ； in Representative rule record Operations in With process model Operation The set of operation pairs, consisting of two pairs of operations. This indicates calculating the number of elements in the set. represent and It is the process model to be inspected. and rule records The pair of matching operations with the highest similarity, and Corresponding to process model The operation in the middle, Corresponding to rule record Operations within; Representative rule record executor in With sets The set of executors in the text consists of two pairs of executors. For process model business objects With the executor group The set that is formed represent and It is the process model to be inspected. and rule records The pair of executors with the highest similarity; Represents two texts Similarity between them This represents the minimum similarity threshold to be included in the summation term. If established The overall value is Otherwise The overall value is set to 0; S4. Calculate the process model to be inspected. The violation score between the matching rule records, where the violation score includes three categories: activity missing violation score, executor error violation score, and execution order error violation score. Process Model With rules record The formula for calculating the violation score for missing activities is as follows: ； in This represents a preset operation similarity threshold. express The similarity satisfies All matching pairs; Process Model With rules record The formula for calculating the executor's error violation score is as follows: ； parameter and The calculation formulas are as follows: ； ； in Represents a preset executor similarity threshold; the symbol " "Represents and; Representative rule record The execution relationship between the executor and the operation. Representative process model The execution relationship between the executor and the operation; Process Model With rules record The formula for calculating the violation score due to incorrect execution order is as follows: ； ； in This represents a preset threshold for similarity between operational violations; express The order relationship in the text, and All are process models and rule records The pair of matching operations with the highest similarity. ; S5. Based on the activity missing violation score, executor error violation score, and execution order error violation score calculated in S4, and combined with the preset violation judgment rules, output the compliance check results of the process model to be checked.

2. The business process compliance check method based on natural language processing according to claim 1, characterized in that, In step S1, extracting the semantic information of activity and event labels in each process model specifically includes the following sub-steps: S11: For each process model activity collection and event collection union The tags for activities and events are extracted and part-of-speech tagging is performed. The words in the tags are classified according to their parts of speech as verbs, adjectives, nouns, keywords, pronouns, prepositions, adverbs, conjunctions, auxiliary words, determiners, and numerals. The words marked as verbs are extracted as the process model. Operation ; S12: For each process model union Dependency parsing was performed on the tagged text of activities and events in S11. A dependency relation in the text connects two words, and the relevant terms in S11 were extracted. Words that have subject-verb or verb-object relationships are used as process models. business objects ; S13: For each process model union Establish its connection with the operations extracted in S11 and the business objects extracted from S12 Mapping relationship between And save the mapping relationship in the process model. middle, .

3. The business process compliance check method based on natural language processing according to claim 2, characterized in that, The dependency relationships are divided into subject-predicate relationships, verb-object relationships, indirect object relationships, prepositional object relationships, pivotal objects, attributive-head relationships, adverbial-head structures, verb-complement structures, coordinate relationships, prepositional object relationships, left-additive relationships, right-additive relationships, independent structures, and core relationships.

4. The business process compliance check method based on natural language processing according to claim 2, characterized in that, In S11, the part-of-speech tagging tool selected is the natural language analysis tool Stanford NLP.

5. The business process compliance check method based on natural language processing according to claim 2, characterized in that, In S12, the dependency parsing tool selected is the natural language processing tool Stanford NLP.

6. The business process compliance check method based on natural language processing according to claim 1, characterized in that, In S3, the matching score threshold Take 0.

5.

7. The business process compliance check method based on natural language processing according to claim 1, characterized in that, The operation similarity threshold Take 0.

8.

8. The business process compliance check method based on natural language processing according to claim 1, characterized in that, The executor similarity threshold Take 0.

7.

9. The business process compliance check method based on natural language processing according to claim 1, characterized in that, In S5, the violation judgment rules are a first score threshold range, a second score threshold range, and a third score threshold range, respectively, for violation scores of missing activity, incorrect executor, and incorrect execution order. When the score of a certain violation deviates from the corresponding score threshold range, it is considered that the violation has a non-compliance problem. The violation includes three types: missing activity, incorrect executor, and incorrect execution order.

10. A smart terminal, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to implement the business process compliance check method based on natural language processing as described in any one of claims 1 to 9 when executing the computer program.

11. The smart terminal according to claim 10, characterized in that, The smart terminal can be a local terminal or a cloud terminal.

Images