A personalized legal risk assessment method based on big data analysis

Abstract

The application discloses a kind of individualized legal risk assessment methods based on big data analysis, it is related to data processing technical field, including the following steps: obtaining legal basis data and matter fact data, determine legal elements and element requirement data;Extract fact record, get fact value, source attribute and evidence capacity, determine establishment action direction, form element evidence data;Input improved GFlowNet model, generate evidence allocation scheme set and allocation probability through bipolar evidence flow conservation mechanism;According to allocation probability, the evidence capacity of establishment support stream and establishment weakening stream is converged, element loading result is generated, element satisfaction, element gap and risk load are determined, and legal risk assessment result is obtained.The application can avoid the same fact record repeated allocation, improve the accuracy and reliability of legal risk assessment.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a personalized legal risk assessment method based on big data analysis. Background Technology

[0002] With the continuous expansion of business operations, personal transactions, and online business, the amount of contract data, transaction records, business vouchers, and legal basis data related to legal matters is growing rapidly. This has led to widespread attention being paid to legal risk assessment technology based on big data analytics. Existing legal risk assessment methods primarily generate risk assessment results based on the correspondence between factual data and legally valid conditions through fixed rule matching, keyword retrieval, and scoring models. However, these methods generally suffer from the following problems in practical applications: Different data sources may record the same factual content, and existing methods struggle to distinguish between duplicate and independent factual records, easily leading to double counting of evidentiary value. The same factual record may correspond to multiple legal elements, and existing methods lack limitations on evidence capacity allocation, easily resulting in the repeated use of the same factual content in both supporting and weakening directions. Factual data may generate multiple evidence allocation schemes, and fixed rule matching and single-path inference are insufficient to reflect different schemes and their corresponding possibilities. Existing scoring methods do not separately handle the requirement fulfillment in the risk-increasing direction and the requirement gap in the risk-reducing direction, making it difficult for legal risk assessment results to accurately reflect the individualized factual situation of the assessed object.

[0003] Therefore, how to provide a personalized legal risk assessment method based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a personalized legal risk assessment method based on big data analysis. This invention determines the legal elements, evidence capacity, and direction of effect of the assessment, and uses an improved GFlowNet model to generate a set of evidence allocation schemes, thereby solving the problems of duplicate allocation of fact records and single-path assessment, and improving the accuracy and reliability of legal risk assessment results.

[0005] A personalized legal risk assessment method based on big data analysis according to an embodiment of the present invention includes the following steps: S1. Obtain legal basis data for the legal matter to be evaluated and factual data for the object to be evaluated. Determine the legal elements based on the legal basis data and form the requirement data based on the factual content limited by the legal elements. S2. Extract fact records corresponding to the legal elements from the factual data of the matter to obtain fact values ​​and source attributes; S3. Determine the evidence capacity based on the source attributes and factual values, and determine the direction of the effect of the factual records on the establishment of the legal elements based on the legal basis data and factual values, thus forming the evidence data of the elements; S4. Input the essential evidence data and essential requirement data into the improved GFlowNet model to obtain the evidence allocation scheme set and allocation probability; introduce a bipolar evidence flow conservation mechanism in the flow balance training module of the improved GFlowNet model to construct the establishment support flow and the establishment weakening flow respectively. S5. For evidence allocation schemes that are centrally allocated to the support stream and weakening stream of the same legal constituent element, perform weighted aggregation according to the corresponding allocation probability to generate the element loading result. S6. Based on the requirement data and the requirement loading results, determine the requirement fulfillment quantity and the requirement gap quantity, and determine the direction of risk increase and risk decrease based on the legal basis data. The risk increase direction uses the requirement fulfillment quantity to form the risk load quantity, and the risk decrease direction uses the requirement gap quantity to form the risk load quantity. S7. Aggregate the risk load of the legal constituent elements to obtain the legal risk assessment result of the object to be assessed.

[0006] Optionally, S1 specifically includes: Obtain legal basis data for the legal matters to be evaluated and factual data of the matters to be evaluated; the factual data includes data content and data source; Extract the conditions for the establishment of candidate legal consequences from the legal basis data, divide the conditions for the establishment into essential factual items according to the factual content that cannot be replaced by each other, and combine the essential factual items according to the candidate legal consequences to obtain the legal elements. The number of factual items corresponding to the same legal constituent element is determined as the evidence requirement quantity. The legal constituent element, factual items, and evidence requirement quantity are combined to form the constituent element requirement data.

[0007] Optionally, S2 specifically includes: Match the data content in the factual data of the matter with the factual content defined by the factual items, and extract the corresponding data content. Data that belongs to the same data source and expresses the same factual content is merged to form a factual record; Determine the fact values ​​corresponding to the essential fact items from the fact records, and identify the data sources corresponding to the fact records as source attributes.

[0008] Optionally, S3 specifically includes: Group fact records according to fact values, merge fact records with the same source attribute in the same group, and retain fact records with different source attributes in the same group; The capacity of evidence is determined by limiting the coverage ratio of the factual content of the essential factual items based on the number of retained factual records and the factual values. Extract the supporting and weakening factual content corresponding to the essential factual items from the legal basis data, and map the factual values ​​to the supporting and weakening factual content respectively to determine the direction of support and weakening of the establishment. The factual records, legal elements, factual items, evidentiary capacity, and direction of validity are combined to form essential evidentiary data.

[0009] Optionally, S4 specifically includes: The improved GFlowNet model includes a state construction module, a bidirectional policy module, a reward calculation module, and a flow balance training module. The state construction module encodes the essential evidence data and essential requirement data to obtain the initial state of evidence allocation; The two-way strategy module selects fact records to be allocated from fact records with a remaining evidence capacity greater than 0, selects legal elements to be loaded from legal elements whose loaded evidence capacity is less than the evidence requirement, and determines the amount of evidence capacity to be allocated. According to the evidence capacity allocation, the evidence capacity of the factual records to be allocated is allocated to the legal elements to be loaded. The evidence capacity allocation action is recorded according to the direction of the establishment effect. The remaining evidence capacity and the loaded evidence capacity are updated to obtain the next evidence allocation status. The next evidence allocation state is generated repeatedly until there are no more factual records and legal elements that meet the allocation conditions. The termination of the evidence allocation state is then determined as the evidence allocation scheme. The reward calculation module calculates the termination reward based on the required amount of evidence, the capacity of already loaded evidence, and the remaining capacity of evidence. A bipolar evidence flow conservation mechanism is introduced into the flow balance training module. Based on the direction of the establishment effect, the establishment support flow and the establishment weakening flow are constructed respectively, and the bipolar policy module is trained according to the termination reward. Multiple evidence allocation schemes are generated by the trained bidirectional policy module to form an evidence allocation scheme set. The allocation probability corresponding to the evidence allocation scheme is obtained based on the positive state transition path probability leading to the same evidence allocation scheme.

[0010] Optionally, the bipolar evidence flow conservation mechanism is specifically as follows: Based on the direction of the establishment effect of the fact record, the evidence capacity allocation action corresponding to the fact record is divided into the establishment support flow and the establishment weakening flow; Set the remaining evidence capacity for the same fact record that is jointly referenced by the support flow and the weakening flow. Deduct the remaining evidence capacity according to the evidence capacity allocation amount corresponding to the evidence capacity allocation action, and accumulate the evidence capacity allocation amount to the corresponding support flow or weakening flow. The difference between the inflow and outflow of evidence in the evidence allocation state that support the flow is obtained by calculating the support flow balance deviation. The difference between the inflow and outflow evidence flows that constitute a weakening flow is calculated to obtain the weakening flow balance bias. Calculate the sum of the cumulative evidence capacity allocation in the support stream, the cumulative evidence capacity allocation in the weakening stream, and the remaining evidence capacity for the same fact record, and determine the difference between the sum and the evidence capacity of the fact record as the capacity conservation deviation; The flow balance training loss is determined based on the support flow balance bias, the weakening flow balance bias, and the capacity conservation bias, and the model parameters of the bidirectional policy module are updated based on the flow balance training loss.

[0011] Optionally, S5 specifically includes: From each evidence allocation scheme in the evidence allocation scheme set, extract the capacity of evidence supporting the establishment of the same legal element and the capacity of evidence weakening the establishment of the same legal element. The capacity for supporting evidence and the capacity for weakening evidence are respectively accumulated within the same evidence allocation scheme and allocated to the same legal elements. The accumulated supporting evidence capacity and weakening evidence capacity are multiplied by their corresponding allocation probabilities, and the multiplication results of the same legal elements in the evidence allocation scheme set are summarized to obtain the supporting load and weakening load. The legal elements, supporting load, and weakening load are combined to generate the element loading result.

[0012] Optionally, determining the requirement fulfillment amount and requirement gap amount based on requirement data and requirement loading results specifically involves: Extract the amount of evidence required for the legal elements from the requirement data, and extract the supporting and weakening loads for the legal elements from the requirement loading results. Calculate the difference between the supporting and weakening loads of the same legal element; When the difference is greater than 0, the smaller value between the difference and the evidence requirement is determined as the requirement satisfaction quantity; when the difference is not greater than 0, the requirement satisfaction quantity is determined as 0. The difference between the required amount of evidence and the required amount of fulfillment of the requirements is calculated to obtain the requirement gap.

[0013] Optionally, the determination of the risk increase direction and risk decrease direction based on legal basis data, wherein the risk increase direction uses the requirement satisfaction amount to form the risk load, and the risk decrease direction uses the requirement gap amount to form the risk load, specifically: Extract the liability and mitigation provisions corresponding to candidate legal consequences from legal basis data; The legal elements constituting the corresponding liability assumption are defined as the direction of risk increase, and the legal elements constituting the corresponding liability mitigation are defined as the direction of risk reduction. The amount of legal elements required to constitute the risk increase direction is determined as the risk load. The gap in the elements constituting the legal elements corresponding to the direction of risk reduction is determined as the risk load.

[0014] Optionally, S7 specifically includes: The elements constituting a legal crime are categorized according to the candidate legal consequences; By summing the risk load of the legal elements in the direction of risk increase and the risk load of the legal elements in the direction of risk decrease for the same candidate legal consequence, we obtain the total risk increase load and the total risk decrease load. Calculate the difference between the total risk increase load and the total risk decrease load. When the difference is greater than 0, the difference is determined as the comprehensive risk load corresponding to the candidate legal consequences. When the difference is not greater than 0, the comprehensive risk load is determined as 0. Candidate legal consequences are ranked according to their comprehensive risk load, and the objects to be assessed, candidate legal consequences, and comprehensive risk load are combined to obtain the legal risk assessment results.

[0015] The beneficial effects of this invention are: (1) This invention introduces a bipolar evidence flow conservation mechanism in the flow balance training module of the improved GFlowNet model, so that the establishment support flow and the establishment weakening flow jointly deduct the remaining evidence capacity of the same fact record, thereby avoiding the repeated allocation of the same fact record in different establishment action directions and multiple legal elements, and improving the rationality of evidence capacity allocation. (2) The present invention performs weighted aggregation on the evidence capacity of the supporting flow and the evidence capacity of the weakening flow according to the allocation probability, which can reflect the multiple evidence allocation results corresponding to the factual data of the matter and improve the accuracy of the loading result of the elements; (3) The present invention uses the requirement satisfaction amount corresponding to the risk increase direction and the requirement gap amount corresponding to the risk decrease direction to form the risk load amount, thereby avoiding the use of the same assessment method for different risk impact directions and improving the reliability of personalized legal risk assessment results. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a personalized legal risk assessment method based on big data analysis proposed in this invention; Figure 2 This is a schematic diagram of the bipolar evidence flow conservation mechanism proposed in this invention; Figure 3 This is a data flow diagram of a personalized legal risk assessment method based on big data analysis proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-3 A personalized legal risk assessment method based on big data analysis includes the following steps: S1. Obtain legal basis data for the legal matter to be evaluated and factual data for the object to be evaluated. Determine the legal elements based on the legal basis data and form the requirement data based on the factual content limited by the legal elements. S2. Extract fact records corresponding to the legal elements from the factual data of the matter to obtain fact values ​​and source attributes; S3. Determine the evidence capacity based on the source attributes and factual values, and determine the direction of the effect of the factual records on the establishment of the legal elements based on the legal basis data and factual values, thus forming the evidence data of the elements; S4. Input the essential evidence data and essential requirement data into the improved GFlowNet model to obtain the evidence allocation scheme set and allocation probability; introduce a bipolar evidence flow conservation mechanism in the flow balance training module of the improved GFlowNet model to construct the establishment support flow and the establishment weakening flow respectively. S5. For evidence allocation schemes that are centrally allocated to the support stream and weakening stream of the same legal constituent element, perform weighted aggregation according to the corresponding allocation probability to generate the element loading result. S6. Based on the requirement data and the requirement loading results, determine the requirement fulfillment quantity and the requirement gap quantity, and determine the direction of risk increase and risk decrease based on the legal basis data. The risk increase direction uses the requirement fulfillment quantity to form the risk load quantity, and the risk decrease direction uses the requirement gap quantity to form the risk load quantity. S7. Aggregate the risk load of the legal constituent elements to obtain the legal risk assessment result of the object to be assessed.

[0019] In this embodiment, S1 specifically refers to: Obtain the legal basis data corresponding to the legal matter to be evaluated and the factual data of the matter to be evaluated. The legal basis data records the conditions for the establishment of the candidate legal consequences according to the candidate legal consequences. The data source is recorded for each data item in the factual data. Based on the candidate legal consequences, the conditions for establishment are determined from the legal basis data, and the facts that need to be met independently and cannot be replaced by other facts are divided into essential fact items. Factual items that share the same condition for establishment are grouped into the same legal elements, and the corresponding legal elements are grouped according to the candidate legal consequences. The number of factual items corresponding to the same legal element is counted, and the statistical results are determined as the amount of evidence required for the legal element. Record the corresponding factual items and evidentiary requirements for each element according to the legal elements, and form the element requirement data.

[0020] In this embodiment, S2 specifically refers to: Obtain the data content and the corresponding data source from the factual data of the matters item by item; Convert the date format, numerical units, and name expressions in the data content into a unified form, and match the converted data content with the factual content specified by the essential factual items one by one; When the converted data content expresses the same fact object, fact behavior, and fact state as the fact content defined by the essential fact item, the correspondence is determined to be successful, and the corresponding successful data content is extracted. The successful data content is collected according to the data source and the essential fact items. Multiple data contents expressing the same fact content from the same data source are merged into one fact record, and data contents expressing the same fact content from different data sources are formed into separate fact records. When the same data content is successfully matched with multiple essential fact items, fact records are formed separately according to the essential fact items; Extract the record results that directly correspond to the factual content defined by the essential factual items from the factual records, and determine the record results as factual values; The data source corresponding to the data content that forms the fact record is determined as the source attribute.

[0021] In this embodiment, S3 specifically refers to: The fact records are grouped according to the essential fact items, and the fact values ​​of the fact records corresponding to the same essential fact item are compared. Fact records with the same fact value are grouped into the same group. Compare the source attributes of fact records within the same group. If the source attributes are the same, merge them into one fact record. If the source attributes are different, retain the fact records separately. The factual content limited by the essential factual items is divided into multiple content parts according to the actual content that appears in the factual object, factual behavior, factual state, and limiting conditions; Compare the fact values ​​with multiple content sections one by one, count the number of corresponding successful content sections, and divide the number of corresponding successful content sections by the total number of content sections to obtain the coverage ratio. The coverage ratio is determined as the evidentiary capacity of each retained fact record, and the sum of the evidentiary capacities of fact records within the same group is equal to the product of the coverage ratio and the number of retained fact records; Determine the supporting and weakening facts for each element of the legal basis from the data. The fact value is compared with the fact content that establishes the fact and the fact content that weakens the fact. When the fact value corresponds to the fact content that establishes the fact, the direction of support for the fact is recorded. When the fact value corresponds to the fact content that weakens the fact, the direction of weakening the fact is recorded. Based on the correspondence between the factual records and the essential factual items, the factual records that determine the direction of the establishment, the legal constituent elements, the essential factual items, the evidentiary capacity, and the direction of the establishment are recorded item by item to form essential evidentiary data.

[0022] In this embodiment, S4 specifically refers to: The improved GFlowNet model includes a state construction module, a bidirectional policy module, a reward calculation module, and a flow balance training module; The status construction module encodes the fact records, evidence capacity, and direction of establishment in the evidence data according to the correspondence between fact records and essential fact items, and encodes the legal constituent elements, essential fact items, and evidence requirement quantity in the requirement demand data. Set the remaining evidence capacity of each fact record to the corresponding evidence capacity, set the loaded evidence capacity of each legal element to 0, and combine the fact record, remaining evidence capacity, direction of establishment, legal element, element fact item, evidence requirement, and loaded evidence capacity to obtain the initial state of evidence allocation. The two-way strategy module determines the fact records with a remaining evidence capacity greater than 0 from the evidence allocation status, and determines the legal elements that can be received for the evidence capacity based on the corresponding fact items of the fact records; From the legal elements that can receive the amount of evidence, determine the legal elements whose loaded evidence capacity is less than the required amount of evidence, and form an allocatable combination of fact records to be allocated and legal elements to be loaded. Calculate the positive state transition probability for each allocatable combination, and select the evidence capacity allocation action based on the positive state transition probability; Calculate the remaining evidence capacity of the factual records to be allocated and the difference between the evidence requirement for the legal elements to be loaded and the already loaded evidence capacity, and determine the smaller of the two values ​​as the evidence capacity allocation amount; According to the evidence capacity allocation, the evidence capacity of the factual records to be allocated is allocated to the legal elements to be loaded. The evidence capacity allocation action is recorded according to the direction of the establishment. The evidence capacity allocation amount is deducted from the remaining evidence capacity of the factual records to be allocated, and the evidence capacity allocation amount is included in the loaded evidence capacity of the legal elements to be loaded, so as to obtain the next evidence allocation status. Based on the evidence capacity allocation action in the next evidence allocation state, calculate the reverse state transition probability of the next evidence allocation state returning to the previous evidence allocation state. Continue to determine the allocatable combinations and evidence capacity allocation actions according to the next evidence allocation status, until the factual records with a remaining evidence capacity greater than 0 cannot form an allocatable combination with the legal elements whose loaded evidence capacity is less than the evidence requirement, and the evidence allocation status is terminated. The evidence allocation scheme is formed by combining the evidence capacity allocation action corresponding to the fact record in the terminated evidence allocation state, the evidence capacity supporting the establishment of the legal elements, and the evidence capacity weakening the establishment of the legal elements. The reward calculation module calculates the difference between the required amount of evidence for each legal element and the capacity of the loaded evidence, and sums up the differences for all legal elements to obtain the unloaded evidence capacity; it then sums up the remaining evidence capacity of all fact records, and determines the termination reward by dividing 1 by the reciprocal of the sum of the unloaded evidence capacity and the sum of the remaining evidence capacity. The flow balance training module introduces a bipolar evidence flow conservation mechanism, which constructs a support flow and a weakening flow based on the direction of the evidence. The flow balance training loss is determined based on the positive state transition probability, the negative state transition probability, and the termination reward, and the model parameters of the bidirectional policy module are updated based on the flow balance training loss. The product of the number of fact records and the number of legal elements is determined as the number of schemes generated. The trained bidirectional strategy module generates evidence allocation schemes according to the number of schemes generated. Evidence allocation schemes with the same content are merged to form a set of evidence allocation schemes. The product of the positive state transition probabilities in each positive state transition path leading to the same evidence allocation scheme is determined as the path probability. The path probabilities leading to the same evidence allocation scheme are summed to obtain the positive state transition path probability corresponding to the evidence allocation scheme. The probability of a positive state transition path corresponding to an evidence allocation scheme is obtained by dividing the probability of a positive state transition path corresponding to all evidence allocation schemes in the evidence allocation scheme set by the probability of a positive state transition path corresponding to all evidence allocation schemes in the evidence allocation scheme set.

[0023] In this embodiment, the bipolar evidence flow conservation mechanism is specifically as follows: Based on the direction of the establishment effect of the fact record, the evidence capacity allocation action corresponding to the establishment support direction is divided into the establishment support flow, and the evidence capacity allocation action corresponding to the establishment weakening direction is divided into the establishment weakening flow. Only one remaining evidence capacity is set for the same fact record. The evidence capacity allocation action in both the support flow and the weakening flow is deducted from the same remaining evidence capacity. When the evidence capacity allocation action is assigned to the establishment support flow, the evidence capacity allocation amount is accumulated to the cumulative evidence capacity allocation amount recorded in the establishment support flow. When the evidence capacity allocation action is assigned to the weakening of the evidence stream, the evidence capacity allocation amount is accumulated to the cumulative evidence capacity allocation amount recorded in the weakening of the evidence stream. For non-terminating evidence allocation states, the product of the evidence flow corresponding to the previous evidence allocation state that entered the evidence allocation state through the evidence capacity allocation action of establishing support direction and the positive state transition probability is added to obtain the inflow evidence flow of establishing support flow. Multiply the evidence flow corresponding to the evidence allocation state by the sum of the positive state transition probabilities corresponding to all supporting evidence capacity allocation actions to obtain the outflow evidence flow of the supporting flow. The absolute value of the difference between the inflow of evidence supporting the establishment of a support flow and the outflow of evidence supporting the establishment of a support flow is defined as the support flow balance deviation. For non-terminating evidence allocation states, the product of the evidence flow corresponding to the previous evidence allocation state that entered the evidence allocation state through the evidence capacity allocation action of establishing the weakening direction and the positive state transition probability is added to obtain the inflow evidence flow of the weakening flow. Multiply the evidence flow corresponding to the evidence allocation state by the sum of the positive state transition probabilities corresponding to all valid weakening direction evidence capacity allocation actions to obtain the outflow evidence flow of the valid weakening flow. The absolute value of the difference between the inflow of evidence establishing a weakening flow and the outflow of evidence establishing a weakening flow is defined as the weakening flow balance deviation. For the termination of evidence allocation status, the cumulative evidence capacity allocation in the supporting stream and the evidence capacity allocation in the weakening stream are added together. When the sum of the two cumulative results is greater than 0, the termination reward is multiplied by the proportion of the cumulative evidence capacity allocation for the establishment of the support flow to the sum of the two cumulative results to obtain the termination outflow evidence flow for the establishment of the support flow. Multiply the termination reward by the proportion of the cumulative evidence capacity allocation for the weakening flow to the sum of the two cumulative results to obtain the termination outflow evidence flow for the weakening flow. When the sum of the two cumulative results equals 0, the outflow evidence flow for establishing the supporting flow and the outflow evidence flow for establishing the weakening flow are both set to 0. The absolute value of the difference between the inflow of supporting evidence and the outflow of supporting evidence in the terminated evidence allocation state is defined as the supporting flow balance deviation. The absolute value of the difference between the inflow of weakening evidence and the outflow of weakening evidence in the termination evidence allocation state is defined as the weakening flow balance deviation. Calculate the sum of the cumulative evidence capacity allocation in the support stream, the cumulative evidence capacity allocation in the weakening stream, and the remaining evidence capacity for the same fact record; The absolute value of the difference between the sum and the evidentiary capacity of the factual record is defined as the capacity conservation bias. Calculate the sum of squares of the support flow balance bias, the sum of squares of the weakening flow balance bias, and the sum of squares of the capacity conservation bias corresponding to all evidence allocation states. The sum of squares of the supporting flow balance bias, the sum of squares of the weakening flow balance bias, and the sum of squares of the capacity conservation bias are added together to obtain the flow balance training loss; the model parameters of the bidirectional policy module are then updated in reverse based on the flow balance training loss.

[0024] In this embodiment, the improved GFlowNet model inherits the state construction module, bidirectional policy module, reward calculation module, and flow balancing training module of the GFlowNet model. The state construction module encodes the essential evidence data and essential requirement data. The bidirectional policy module generates evidence capacity allocation actions based on the evidence allocation state. The reward calculation module calculates the termination reward based on the termination evidence allocation state. The flow balancing training module trains the bidirectional policy module based on the forward state transition probability, the reverse state transition probability, and the termination reward. The improved GFlowNet model introduces a bipolar evidence flow conservation mechanism based on the flow balancing training module. Evidence capacity allocation actions are counted into the establishment support flow and the establishment weakening flow according to the direction of their effect. Furthermore, for the same fact record, both the establishment support flow and the establishment weakening flow are configured. The remaining evidence capacity after joint deduction is used to calculate the flow balance deviation for the support flow and the weakening flow, respectively. At the same time, the capacity conservation deviation is determined based on the cumulative evidence capacity allocation of the same fact record in the support flow, the cumulative evidence capacity allocation in the weakening flow, and the remaining evidence capacity. The support flow balance deviation, the weakening flow balance deviation, and the capacity conservation deviation are all included in the flow balance training loss. The improved GFlowNet model can limit the repeated allocation of evidence capacity for the same fact record in the support flow and the weakening flow, keep the evidence capacity allocation process consistent with the evidence capacity of the fact record, and generate multiple evidence allocation schemes with corresponding allocation probabilities under the condition of meeting the evidence requirements of the legal elements, thereby improving the consistency of the element loading results and the reliability of the personalized legal risk assessment results.

[0025] In this embodiment, S5 specifically refers to: The evidence allocation schemes are traversed item by item according to the legal elements, and the factual records, evidence capacity allocation, and direction of establishment are determined for each evidence allocation scheme that is assigned to the same legal element. The amount of evidence capacity allocated to the direction of establishing support is included in the evidence capacity of the support flow, and the amount of evidence capacity allocated to the direction of establishing weakening is included in the evidence capacity of the weakening flow. By summing up the capacity of evidence supporting the establishment of a legal element and the capacity of evidence weakening the establishment of a legal element within the same evidence allocation scheme, we obtain the total capacity of evidence supporting the establishment of a legal element and the total capacity of evidence weakening the establishment of a legal element corresponding to the evidence allocation scheme. When there is no capacity of evidence supporting the establishment of the legal elements in the evidence allocation scheme, the total capacity of evidence supporting the establishment of the legal elements is recorded as 0. When there is no capacity of evidence weakening the establishment of the legal elements in the evidence allocation scheme, the total capacity of evidence weakening the establishment of the legal elements is recorded as 0. Multiply the total capacity of evidence supporting the establishment of the legal elements corresponding to each evidence allocation scheme by the corresponding allocation probability to obtain the weighted result of the establishment support flow. Multiply the total amount of evidence weakening the legal elements corresponding to each evidence allocation scheme by the corresponding allocation probability to obtain the weighted result of the weakening evidence. If the cumulative evidence allocation schemes support the weighted result of the same legal element, the cumulative result is determined as the supporting load; if the cumulative evidence allocation schemes weaken the weighted result of the same legal element, the cumulative result is determined as the weakened load. Record the supporting and weakening load quantities item by item according to the legal requirements, and generate the load result of the requirements.

[0026] In this embodiment, based on the requirement data and the requirement loading results, the requirement fulfillment amount and the requirement gap amount are determined, specifically as follows: The data required for each element and the loading results of each element are matched according to the legal constituent elements; Obtain the amount of evidence required for the legal elements from the requirement data, and obtain the supporting and weakening loads for the same legal element from the requirement loading results; Subtracting the weakening load from the supporting load yields the difference corresponding to the same legal element. When the difference is greater than 0, compare the difference with the required amount of evidence, and determine the smaller value as the requirement to be met, limiting the requirement to be met to no more than the required amount of evidence; when the difference is not greater than 0, the requirement to be met is determined to be 0. Subtract the required number of elements from the required number of evidence to obtain the required number of elements; record the required number of elements and the required number of elements in accordance with the legal elements.

[0027] In this implementation, the directions of risk increase and risk decrease are determined based on legal basis data. The risk increase direction uses the requirement fulfillment amount to form the risk load, while the risk decrease direction uses the requirement gap amount to form the risk load. Specifically: Based on the candidate legal consequences, extract the conditions for establishing liability from the legal basis data to form the liability assumption content, and extract the conditions for establishing liability mitigation to form the liability mitigation content; The elements of a legal constitution are matched one by one with the content of liability. When the content of the facts limited by the elements of a legal constitution is consistent with the content of liability, the elements of a legal constitution are identified as the direction of increased risk. The elements of a legal constitution are matched one by one with the content of liability mitigation. When the factual content limited by the elements of a legal constitution is consistent with the content of liability mitigation, the elements of a legal constitution are determined as the direction of risk reduction. The quantity of fulfilled requirements and the quantity of missing requirements are obtained item by item according to the legal constituent elements; For the legal elements corresponding to the direction of increased risk, the amount of element satisfaction is determined as the risk load. When the amount of element satisfaction is 0, the risk load is 0. For the legal elements corresponding to the direction of risk reduction, the element gap is determined as the risk load. When the element gap is 0, the risk load is 0. Record the legal elements, risk increase direction, risk decrease direction, and risk load for each candidate legal consequence.

[0028] In this embodiment, S7 specifically refers to: Determine the legal elements corresponding to each candidate legal consequence from the legal basis data, and collect the legal elements according to the candidate legal consequences; classify the legal elements corresponding to the same candidate legal consequence according to the direction of risk increase and risk decrease. The risk load corresponding to each legal element constituting the risk increase direction is obtained item by item, and the obtained risk loads are summed to obtain the total risk increase load. When there is no legal element constituting the risk increase direction, the total risk increase load is determined to be 0. The risk load corresponding to each legal element in the direction of risk reduction is obtained one by one, and the obtained risk loads are summed to obtain the total risk reduction load. If there is no legal element in the direction of risk reduction, the total risk reduction load is determined to be 0. Subtracting the total risk reduction from the total risk increase in the total load yields the difference corresponding to the candidate legal consequences. When the difference is greater than 0, the difference is determined as the comprehensive risk load; when the difference is not greater than 0, the comprehensive risk load is determined as 0; the candidate legal consequences are arranged in descending order of comprehensive risk load. The legal risk assessment results are obtained by matching and summarizing the objects to be assessed, the ranked candidate legal consequences, and the comprehensive risk load corresponding to the candidate legal consequences.

[0029] Example 1: To verify the feasibility of this invention in practice, it was applied to a contract performance legal risk assessment for an equipment manufacturing company. The company, as the equipment supplier, signed a production line delivery contract with a customer for a total amount of 28.6 million yuan. For late delivery, the company would bear a penalty of 0.03% of the undelivered equipment value per day. The customer claimed a 17-day delay and demanded a penalty of 1.86 million yuan. The company argued that the customer had repeatedly changed equipment interface parameters during production, and that the transportation of some key components was affected by extreme weather, necessitating a comprehensive assessment of the possibility of liability, mitigation, and exemption.

[0030] The legal basis data includes the conditions for establishing contract performance, liability for breach of contract, mitigation of liability, and force majeure. The factual data includes contract texts, technical agreements, design change emails, production records, inspection records, logistics records, weather certificates, payment vouchers, and communication records between the parties, totaling 86 data items from 9 different data sources. Data items expressing the same factual content and belonging to the same data source were merged, resulting in 59 factual records. Based on the legal basis data, the legal elements constituting the contract's validity, agreed delivery date, actual delayed delivery, attribution of the cause of delay, impact of customer design changes, and proof of force majeure notification were determined, and the required amount of evidence was determined based on the number of corresponding factual items.

[0031] The factual values ​​are mapped to the factual content defined by the essential factual items. Nine data items repeatedly forwarded within the same email chain in the design change email are merged into three factual records, and six repeatedly updated transportation statuses in the logistics system are merged into two factual records. Evidence capacity is determined based on source attributes and the coverage ratio of factual content, and the directions supporting and weakening the establishment are identified. After inputting the essential evidence data and essential requirement data into the improved GFlowNet model, a total of 240 evidence allocation schemes are generated. Schemes with identical content are merged to form 37 evidence allocation schemes, with a total allocation probability of 1. The establishment support flow and the establishment weakening flow jointly deduct the remaining evidence capacity of the same factual record to avoid the design change email being completely allocated to both the establishment of overdue liability and the mitigation of liability directions simultaneously. Table 1 shows the loading results of representative legal elements.

[0032] Table 1. Requirement Load and Risk Load Results for Typical Delayed Delivery Items

[0033] Table 1 shows that the supporting load for the two legal elements of a valid contract and breach of contract clause, and a clearly defined delivery date, is close to the evidentiary requirement, indicating that the basic facts of the contract are relatively complete. The attribution of the delay to the supplier has both supporting and weakening loads, reflecting that production records support the supplier's delay, while design change emails and customer confirmation records weaken the attribution of liability. The requirement of customer design changes affecting delivery is met at a rate of 2.09, indicating a strong factual basis for mitigating liability. The requirement of force majeure notification and proof is lacking at a rate of 1.85, mainly because weather evidence exists but a complete contract notification record is not formed. The bipolar evidence flow conservation mechanism restricts the reuse of the same factual record, ensuring consistency between the required load and the actual source of evidence.

[0034] To verify the evaluation effect, 180 contract performance items reviewed by the company's legal personnel were selected, of which 130 items were used for model training and 50 items were used for testing. The present invention was compared with the fixed rule matching method and the GFlowNet model without the introduction of the bipolar evidence flow conservation mechanism. The results are shown in Table 2.

[0035] Table 2 Test Results of Different Legal Risk Assessment Methods

[0036] Table 2 shows that both the fixed rule matching method and the unmodified GFlowNet model suffer from the problems of duplicate allocation of evidence capacity and bidirectional over-allocation. This invention reduces the number of duplicate allocations and over-allocations to zero by jointly deducting the remaining evidence capacity through the establishment of support and weakening flows. The first-to-last consistency rate of candidate legal consequences increased from 74.0% to 92.0%, the average absolute difference in risk load decreased to 0.34, and the high-risk event identification rate reached 95.2%. Although the bipolar evidence flow conservation mechanism adds a small amount of training and computational overhead, the processing time per item is still controlled at 4.9 minutes, significantly lower than the fixed rule matching method, indicating that this invention can balance the rationality of evidence allocation, the accuracy of legal risk assessment, and processing efficiency.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A personalized legal risk assessment method based on big data analysis, characterized in that, Includes the following steps: S1. Obtain legal basis data for the legal matter to be evaluated and factual data for the object to be evaluated. Determine the legal elements based on the legal basis data and form the requirement data based on the factual content limited by the legal elements. S2. Extract fact records corresponding to the legal elements from the factual data of the matter to obtain fact values ​​and source attributes; S3. Determine the evidence capacity based on the source attributes and factual values, and determine the direction of the effect of the factual records on the establishment of the legal elements based on the legal basis data and factual values, thus forming the evidence data of the elements; S4. Input the essential evidence data and essential requirement data into the improved GFlowNet model to obtain the evidence allocation scheme set and allocation probability. In the improved GFlowNet model's flow balance training module, a bipolar evidence flow conservation mechanism is introduced to construct the establishment support flow and the establishment weakening flow, respectively. S5. For evidence allocation schemes that are centrally allocated to the support stream and weakening stream of the same legal constituent element, perform weighted aggregation according to the corresponding allocation probability to generate the element loading result. S6. Based on the requirement data and the requirement loading results, determine the requirement fulfillment quantity and the requirement gap quantity, and determine the direction of risk increase and risk decrease based on the legal basis data. The risk increase direction uses the requirement fulfillment quantity to form the risk load quantity, and the risk decrease direction uses the requirement gap quantity to form the risk load quantity. S7. Aggregate the risk load of the legal constituent elements to obtain the legal risk assessment result of the object to be assessed.

2. The personalized legal risk assessment method based on big data analysis according to claim 1, characterized in that, Specifically, S1 is: Obtain legal basis data for the legal matters to be evaluated and factual data of the matters to be evaluated; the factual data includes data content and data source; Extract the conditions for the establishment of candidate legal consequences from the legal basis data, divide the conditions for the establishment into essential factual items according to the factual content that cannot be replaced by each other, and combine the essential factual items according to the candidate legal consequences to obtain the legal elements. The number of factual items corresponding to the same legal constituent element is determined as the evidence requirement quantity. The legal constituent element, factual items, and evidence requirement quantity are combined to form the constituent element requirement data.

3. The personalized legal risk assessment method based on big data analysis according to claim 1, characterized in that, Specifically, S2 is: Match the data content in the factual data of the matter with the factual content defined by the factual items, and extract the corresponding data content. Data that belongs to the same data source and expresses the same factual content is merged to form a factual record; Determine the fact values ​​corresponding to the essential fact items from the fact records, and identify the data sources corresponding to the fact records as source attributes.

4. The personalized legal risk assessment method based on big data analysis according to claim 1, characterized in that, Specifically, S3 is: Group fact records according to fact values, merge fact records with the same source attribute in the same group, and retain fact records with different source attributes in the same group; The capacity of evidence is determined by limiting the coverage ratio of the factual content of the essential factual items based on the number of retained factual records and the factual values. Extract the supporting and weakening factual content corresponding to the essential factual items from the legal basis data, and map the factual values ​​to the supporting and weakening factual content respectively to determine the direction of support and weakening of the establishment. The factual records, legal elements, factual items, evidentiary capacity, and direction of validity are combined to form essential evidentiary data.

5. The personalized legal risk assessment method based on big data analysis according to claim 1, characterized in that, Specifically, S4 is: The improved GFlowNet model includes a state construction module, a bidirectional policy module, a reward calculation module, and a flow balance training module. The state construction module encodes the essential evidence data and essential requirement data to obtain the initial state of evidence allocation; The two-way strategy module selects fact records to be allocated from fact records with a remaining evidence capacity greater than 0, selects legal elements to be loaded from legal elements whose loaded evidence capacity is less than the evidence requirement, and determines the evidence capacity allocation amount. According to the evidence capacity allocation, the evidence capacity of the factual records to be allocated is allocated to the legal elements to be loaded. The evidence capacity allocation action is recorded according to the direction of the establishment effect. The remaining evidence capacity and the loaded evidence capacity are updated to obtain the next evidence allocation status. The next evidence allocation state is generated repeatedly until there are no more factual records and legal elements that meet the allocation conditions. The termination of the evidence allocation state is then determined as the evidence allocation scheme. The reward calculation module calculates the termination reward based on the required amount of evidence, the capacity of already loaded evidence, and the remaining capacity of evidence. A bipolar evidence flow conservation mechanism is introduced into the flow balance training module. Based on the direction of the establishment effect, the establishment support flow and the establishment weakening flow are constructed respectively, and the bipolar policy module is trained according to the termination reward. Multiple evidence allocation schemes are generated by the trained bidirectional policy module to form an evidence allocation scheme set. The allocation probability corresponding to the evidence allocation scheme is obtained based on the positive state transition path probability leading to the same evidence allocation scheme.

6. The personalized legal risk assessment method based on big data analysis according to claim 5, characterized in that, The bipolar evidence flow conservation mechanism is as follows: Based on the direction of the establishment effect of the fact record, the evidence capacity allocation action corresponding to the fact record is divided into the establishment support flow and the establishment weakening flow; Set the remaining evidence capacity for the same fact record that is jointly referenced by the support stream and the weakening stream. Deduct the remaining evidence capacity according to the evidence capacity allocation amount corresponding to the evidence capacity allocation action, and accumulate the evidence capacity allocation amount to the corresponding support stream or weakening stream. The difference between the inflow and outflow of evidence in the evidence allocation state that support the flow is obtained by calculating the support flow balance deviation. The difference between the inflow and outflow evidence flows that constitute a weakening flow is calculated to obtain the weakening flow balance bias. Calculate the sum of the cumulative evidence capacity allocation in the support stream, the cumulative evidence capacity allocation in the weakening stream, and the remaining evidence capacity for the same fact record, and determine the difference between the sum and the evidence capacity of the fact record as the capacity conservation deviation; The flow balance training loss is determined based on the support flow balance bias, the weakening flow balance bias, and the capacity conservation bias, and the model parameters of the bidirectional policy module are updated based on the flow balance training loss.

7. The personalized legal risk assessment method based on big data analysis according to claim 1, characterized in that, Specifically, S5 is: From each evidence allocation scheme in the evidence allocation scheme set, extract the capacity of evidence supporting the establishment of the same legal element and the capacity of evidence weakening the establishment of the same legal element. The capacity for supporting evidence and the capacity for weakening evidence are respectively accumulated within the same evidence allocation scheme and allocated to the same legal elements. The accumulated supporting evidence capacity and weakening evidence capacity are multiplied by their corresponding allocation probabilities, and the multiplication results of the same legal elements in the evidence allocation scheme set are summarized to obtain the supporting load and weakening load. The legal elements, supporting load, and weakening load are combined to generate the element loading result.

8. The personalized legal risk assessment method based on big data analysis according to claim 1, characterized in that, The determination of the requirement fulfillment quantity and requirement gap quantity based on the requirement data and requirement loading results is as follows: Extract the amount of evidence required for the legal elements from the requirement data, and extract the supporting and weakening loads for the legal elements from the requirement loading results. Calculate the difference between the supporting and weakening loads of the same legal element; When the difference is greater than 0, the smaller value between the difference and the evidence requirement is determined as the requirement satisfaction quantity; when the difference is not greater than 0, the requirement satisfaction quantity is determined to be 0. The difference between the required amount of evidence and the required amount of fulfillment of the requirements is calculated to obtain the requirement gap.

9. The personalized legal risk assessment method based on big data analysis according to claim 1, characterized in that, The determination of risk increase and risk decrease directions based on legal basis data is as follows: Risk increase is represented by the requirement fulfillment amount forming the risk load, while risk decrease is represented by the requirement gap amount forming the risk load. Extract the liability and mitigation provisions corresponding to candidate legal consequences from legal basis data; The legal elements constituting the corresponding liability assumption are defined as the direction of risk increase, and the legal elements constituting the corresponding liability mitigation are defined as the direction of risk reduction. The amount of legal elements required to constitute the risk increase direction is determined as the risk load. The gap in the elements constituting the legal elements corresponding to the direction of risk reduction is determined as the risk load.

10. The personalized legal risk assessment method based on big data analysis according to claim 1, characterized in that, Specifically, S7 is: The elements constituting a legal crime are categorized according to the candidate legal consequences; By summing the risk load of the legal elements in the direction of risk increase and the risk load of the legal elements in the direction of risk decrease for the same candidate legal consequence, we obtain the total risk increase load and the total risk decrease load. Calculate the difference between the total risk increase load and the total risk decrease load. When the difference is greater than 0, the difference is determined as the comprehensive risk load corresponding to the candidate legal consequences. When the difference is not greater than 0, the comprehensive risk load is determined as 0. Candidate legal consequences are ranked according to their comprehensive risk load, and the objects to be assessed, candidate legal consequences, and comprehensive risk load are combined to obtain the legal risk assessment results.

Images