Abnormality degree identification method, device, medium and equipment based on large language model

By using an anomaly identification method based on a large language model, combined with subject association graphs and anomalous event texts, the relationships between subjects are analyzed, solving the problem of low identification accuracy in existing technologies and achieving higher accuracy in anomaly identification.

CN118673445BActive Publication Date: 2024-11-08HANGZHOU YSCREDIT CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411148183.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-21
Publication Date
2024-11-08
Estimated Expiration
2044-08-21

AI Technical Summary

Technical Problem

In existing technologies, methods for identifying the degree of anomaly of a subject fail to effectively consider the relationships between subjects, resulting in low accuracy and difficulty in meeting the needs of fields such as bank risk management.

Method used

An anomaly identification method based on a large language model is adopted. By acquiring the text of the anomalous event and subsequent events of the subject, and combining it with a pre-set subject association graph, the association type and the degree of anomalous transmission between subjects are analyzed to identify the anomalous degree of the target subject.

Benefits of technology

It improves the accuracy of identifying the degree of anomaly of target entities, and can more accurately measure the abnormal impact between entities, meeting the needs of fields such as bank risk management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118673445B_ABST
    Figure CN118673445B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of data processing, and in particular to a method, device, medium and equipment for identifying abnormality levels based on a large language model, comprising: inputting abnormal event text and subsequent event text into a preset large language model, obtaining an initial subject, a first abnormality level and a subject role type corresponding to the initial subject, obtaining a third abnormality level corresponding to each abnormal source subject according to the subject role type and the first abnormality level, and obtaining an association path between a target subject and an abnormal source subject in combination with a preset subject association graph, as well as an association type set, an association level set, a first abnormality conduction level set and a third abnormality level set corresponding to each association path, to measure the abnormality conducted from other initial subjects except the target subject on the association path to the target subject, and obtaining a target abnormality level corresponding to the target subject in combination with the third abnormality level of the target subject itself, thereby improving the accuracy of the target abnormality level.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a method, device, medium and equipment for identifying abnormality based on a large language model. Background Art

[0002] Identification of the abnormal degree of the subject can identify, qualitatively and quantitatively analyze the potential abnormal factors of the subject, help organizations and individuals understand the various risks they may face, and better avoid and control abnormalities. It can also help organizations and individuals sort and prioritize credit anomalies and rationally allocate limited resources. It is of great significance in many fields such as banking, corporate management, medical and health, and information security. For example, in the banking field, risk identification of customer subjects can determine whether customers applying for loans have a risk of default, thereby helping bank staff decide whether to lend to customers, reduce bad debt risks while reducing operating costs, and improve the business security of banks.

[0003] In the existing abnormality degree identification methods, the abnormal probability of the subject in each credit dimension is mainly measured, and the abnormal probability of multiple credit dimensions is combined with the corresponding importance weights of each credit dimension to obtain the overall credit abnormality level of the subject. However, the above method ignores the impact of the relationship between subjects on the subject abnormality, which makes it difficult to reasonably evaluate the credit abnormality level of the subject, resulting in low identification accuracy of the target subject. For example, in the field of banking business, customer risks not only come from themselves, but can also be transmitted through related parties such as equity chains, supply chains, and capital chains. It is difficult to accurately assess the risk level of a single customer through risk monitoring, which makes it difficult to meet the needs of bank risk management.

[0004] Therefore, how to improve the recognition accuracy of the abnormality degree of the target subject becomes an urgent problem to be solved. Summary of the invention

[0005] In view of the above technical problems, the technical solution adopted by the present invention is a method for identifying the degree of abnormality based on a large language model, and the method for identifying the degree of abnormality includes the following steps:

[0006] S1, obtain M abnormal event texts corresponding to N initial subjects, subsequent event texts corresponding to each abnormal event text, and a preset subject association map, wherein the preset subject association map includes N initial subjects as nodes, association types and association degrees between connected initial subjects, and a first abnormal conduction degree corresponding to each association type, wherein N and M are integers greater than 0, and the first abnormal conduction degree is used to characterize the amplitude value of abnormal conduction between corresponding connected initial subjects.

[0007] S2, for any abnormal event text, input the current abnormal event text into a preset large language model, and obtain the initial subject corresponding to the current abnormal event text and the subject role type corresponding to the current initial subject.

[0008] S3, inputting the current abnormal event text and the corresponding subsequent event text into a preset large language model to obtain a first abnormality degree corresponding to the current abnormal event text.

[0009] S4, determining the current initial subject as the abnormal source subject, and obtaining the second abnormal degree corresponding to the current abnormal source subject for the current abnormal event text according to the subject role type and the first abnormal degree corresponding to the current abnormal source subject.

[0010] S5, traverse all abnormal event texts, and obtain the third abnormal degree corresponding to all abnormal event texts of each abnormal source subject according to the second abnormal degree corresponding to each abnormal event text of the current abnormal source subject.

[0011] S6. Take any initial subject as the target subject to be identified, and according to the preset subject association map and the third abnormality degree corresponding to each abnormal source subject, obtain K association paths between the current target subject and each abnormal source subject, as well as the association type set, association degree set, first abnormal conduction degree set and third abnormality degree set corresponding to each association path, where K is an integer greater than 0.

[0012] S7, obtaining the target abnormality degree corresponding to the current target subject according to the association degree set, the first abnormal conduction degree set and the third abnormality degree set corresponding to each association path.

[0013] The present invention also provides an abnormality degree identification device based on a large language model, the abnormality degree identification device comprising:

[0014] The first data acquisition module is used to obtain M abnormal event texts corresponding to N initial subjects, subsequent event texts corresponding to each abnormal event text, and a preset subject association map, wherein the preset subject association map includes N initial subjects as nodes, association types and association degrees between connected initial subjects, and a first abnormal conduction degree corresponding to each association type, wherein N and M are integers greater than 0, and the first abnormal conduction degree is used to characterize the amplitude value of abnormal conduction between the corresponding connected initial subjects.

[0015] The large language model application module is used to input the current abnormal event text into a preset large language model for any abnormal event text, and obtain the initial subject corresponding to the current abnormal event text and the subject role type corresponding to the current initial subject.

[0016] The first abnormality degree acquisition module is used to input the current abnormal event text and the corresponding subsequent event text into a preset large language model to acquire the first abnormality degree corresponding to the current abnormal event text.

[0017] The second abnormality degree acquisition module is used to determine the current initial subject as the abnormal source subject, and acquire the second abnormality degree corresponding to the current abnormal source subject for the current abnormal event text according to the subject role type and the first abnormality degree corresponding to the current abnormal source subject.

[0018] The third abnormality degree acquisition module is used to traverse all abnormal event texts, and acquire the third abnormality degree corresponding to each abnormal source subject for all abnormal event texts according to the second abnormality degree corresponding to each abnormal event text of the current abnormal source subject.

[0019] The second data acquisition module is used to take any initial subject as the target subject to be identified, and obtain K association paths between the current target subject and each abnormal source subject according to a preset subject association map and the third abnormality degree corresponding to each abnormal source subject, as well as an association type set, an association degree set, a first abnormal conduction degree set and a third abnormality degree set corresponding to each association path, wherein K is an integer greater than 0.

[0020] The target abnormality degree acquisition module is used to acquire the target abnormality degree corresponding to the current target subject according to the association degree set corresponding to each association path, the first abnormal conduction degree set and the third abnormality degree set.

[0021] The present invention also provides a non-transitory computer-readable storage medium, in which at least one instruction or at least one program is stored, and the at least one instruction or at least one program is loaded and executed by a processor to implement the above-mentioned abnormality degree identification method based on a large language model.

[0022] The present invention also provides an electronic device, comprising a processor and the above-mentioned non-transitory computer-readable storage medium.

[0023] The present invention has at least the following beneficial effects: based on a preset large language model, the current abnormal event text is analyzed, the initial subject corresponding to the current abnormal event text and the subject role type corresponding to the current initial subject are obtained, the current abnormal event text and the corresponding subsequent event text are input into the preset large language model, and the current abnormal impact and subsequent abnormal impact of the abnormal event corresponding to the current abnormal event text are analyzed at the same time, and the first abnormal degree corresponding to the current abnormal event text is obtained, thereby improving the accuracy of characterizing the abnormal degree of the abnormal event corresponding to the current abnormal event text, determining the current initial subject as the abnormal source subject, and obtaining the second abnormal degree corresponding to the current abnormal source subject for the current abnormal event text according to the subject role type and the first abnormal degree corresponding to the current abnormal source subject, traversing all abnormal event texts, and analyzing the abnormal impact of the current abnormal event text according to the current abnormal source subject for each abnormal event text. The corresponding second abnormality degree is obtained, and the third abnormality degree corresponding to each abnormal source subject for all abnormal event texts is obtained. Any initial subject is taken as the target subject to be identified, and according to the preset subject association map and the third abnormality degree corresponding to each abnormal source subject, K association paths between the current target subject and each abnormal source subject are obtained, as well as the association type set, association degree set, first abnormal conduction degree set and third abnormality degree set corresponding to each association path. According to the association degree, first abnormal conduction degree, matching degree and the third abnormality degree corresponding to all initial subjects on the association path, the abnormalities transmitted to the target subject from other initial subjects except the target subject on the association path are measured, and combined with the third abnormality degree of the target subject itself, the target abnormality degree corresponding to the target subject is obtained, thereby improving the accuracy of the target abnormality degree. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 A flowchart of a method for identifying abnormality based on a large language model provided in Embodiment 1 of the present invention;

[0026] Figure 2 A schematic diagram of the structure of an abnormality degree identification device based on a large language model provided in Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It is understood that, where appropriate, the above-mentioned terms used to distinguish similar objects can be interchanged so that the present invention can also implement other embodiments other than the above-mentioned illustrated embodiments or described embodiments. In addition, the terms "including" and "having" and any variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] Embodiment 1

[0030] The first embodiment provides a method for identifying abnormality based on a large language model. The method for identifying abnormality based on a large language model includes the following steps: Figure 1 As shown:

[0031] S1, obtain M abnormal event texts corresponding to N initial subjects, subsequent event texts corresponding to each abnormal event text, and a preset subject association map, wherein the preset subject association map includes N initial subjects as nodes, association types and association degrees between connected initial subjects, and a first abnormal conduction degree corresponding to each association type, wherein N and M are integers greater than 0, and the first abnormal conduction degree is used to characterize the amplitude value of abnormal conduction between corresponding connected initial subjects.

[0032] Among them, the initial subject refers to the subject that needs to be identified for abnormality. By obtaining the abnormality level corresponding to the initial subject, it can help organizations and individuals understand the various risks they may face, so as to better avoid and control abnormalities. It can also help organizations and individuals sort and prioritize credit abnormalities and reasonably allocate limited resources. For example, in the banking business field, the initial subject can refer to the loan customer who needs to be risk assessed. In the medical and health field, the subject to be identified can refer to the patient or hospital that needs to be risk assessed. In the information security field, the subject to be identified can refer to the equipment or information for security assessment.

[0033] The abnormal event text refers to the description text of the abnormal event when the abnormal event occurs, and the subsequent event text can refer to the description text of the subsequent events that occur after the abnormal event occurs. By performing semantic analysis on the abnormal event text and the subsequent event text, the subject information, abnormal type information and abnormal degree information corresponding to the abnormal event can be extracted as the basis for identifying the abnormal degree of each initial subject. For example, the abnormal event text and the subsequent event text can refer to descriptive texts such as news texts and public account texts.

[0034] When constructing a preset subject association graph, N initial subjects are taken as nodes, edges are used to connect the initial subjects with association relationships, and the association type, association degree and the first abnormal conduction degree corresponding to the association type between the connected initial subjects are stored as edge attribute information, so as to construct a preset subject association graph based on the initial subjects and the association data between them.

[0035] Among them, the association type can represent the relationship information and direction information between the initial subjects. Based on the association relationship between the initial subjects, the abnormality of one initial subject will affect the abnormality of other initial subjects associated with it. The first abnormal conduction degree is used to represent the amplitude value of abnormal conduction between the corresponding connected initial subjects. Different association types between the initial subjects have different first abnormal conduction degrees when abnormal conduction occurs between the corresponding connected initial subjects, which can be used to analyze the mutual influence when abnormal conduction occurs between different initial subjects.

[0036] The specific association type and the corresponding first abnormal conduction degree can be set by the implementer according to actual conditions.

[0037] In a specific implementation, the types of association include relationships between subjects such as first group relationship, second group relationship, first equity relationship, second equity relationship, first position relationship, second position relationship, first guarantee relationship, second guarantee relationship, first supply relationship, second supply relationship, first transaction relationship, second transaction relationship, first public-private relationship, second public-private relationship, litigation relationship, and patent relationship. When there are association relationships of the above types of associations between initial subjects, abnormal conduction will occur between the associated initial subjects.

[0038] The first group relationship is from the group parent company to the group subsidiary company, and the first abnormal transmission degree corresponding to the first group relationship can be set to 0.8; the second group relationship is from the group subsidiary company to the group parent company, and the first abnormal transmission degree corresponding to the second group relationship can be set to 0.4; the first equity relationship is from the shareholder company to the invested company, and the first abnormal transmission degree corresponding to the first equity relationship can be set to 0.9; the second equity relationship is from the invested company to the shareholder company, and the first abnormal transmission degree corresponding to the second equity relationship can be set to 0.6; the first employment relationship is from the employee company to the enterprise company, and the first abnormal transmission degree corresponding to the first employment relationship can be set to 0.9; the second employment relationship is from the enterprise company to the employee company, and the first abnormal transmission degree corresponding to the second employment relationship can be set to 0.5; the first guarantee relationship is from the guaranteeing company to the guaranteed company, and the first abnormal transmission degree corresponding to the first guarantee relationship can be set to 0. The normal transmission degree can be set to 0.4; the second guarantee relationship points from the guaranteed enterprise entity to the guarantor enterprise entity, and the first abnormal transmission degree corresponding to the second guarantee relationship can be set to 1.0; the first supply relationship points from the supplier entity to the customer entity, and the first abnormal transmission degree corresponding to the first supply relationship can be set to 0.5; the second supply relationship points from the customer entity to the supplier entity, and the first abnormal transmission degree corresponding to the second supply relationship can be set to 0.8; the first transaction relationship points from the capital outflow enterprise entity to the capital inflow enterprise entity, and the first abnormal transmission degree corresponding to the first transaction relationship can be set to 0.8; the second transaction relationship points from the capital inflow enterprise entity to the capital outflow enterprise entity, and the first abnormal transmission degree corresponding to the second transaction relationship can be set to 0.5; there is no directionality in the litigation relationship and the patent relationship, and the first abnormal transmission degree corresponding to the litigation relationship can be set to 0.5, and the first abnormal transmission degree corresponding to the patent relationship can be set to 0.3.

[0039] In a specific implementation, the types of association include a first equity relationship and a second equity relationship, a first guarantee relationship and a second guarantee relationship, a first transaction relationship and a second transaction relationship, a litigation relationship and a patent relationship, and S1 further includes the following steps:

[0040] S11: If the type of association between the connected initial entities is a first equity relationship or a second equity relationship, the equity ratio between the connected initial entities is determined as the corresponding degree of association.

[0041] S12: If the type of association between the connected initial entities is a first guarantee relationship or a second guarantee relationship, the proportion of the guarantee amount between the connected initial entities is determined as the corresponding degree of association.

[0042] S13: If the type of association between the connected initial entities is a first transaction relationship and a second transaction relationship, the proportion of the transaction amount between the connected initial entities is determined as the corresponding degree of association.

[0043] S14: If the type of association between the connected initial entities is a litigation relationship or a patent relationship, a preset value is determined as the corresponding degree of association.

[0044] Among them, the greater the equity ratio, guarantee quota ratio, and transaction quota ratio between the connected initial entities, the closer the association between the corresponding initial entities, and the greater the impact of the abnormality of one initial entity on the abnormality of another initial entity. Therefore, the equity ratio, guarantee quota ratio, or transaction quota ratio between the connected initial entities are positively correlated with the corresponding degree of association.

[0045] The preset value may be set by the implementer according to actual conditions. For example, the preset value may be set to 1 to indicate that the degree of association between the two connected initial entities is extremely high.

[0046] S2, for any abnormal event text, input the current abnormal event text into a preset large language model, and obtain the initial subject corresponding to the current abnormal event text and the subject role type corresponding to the current initial subject.

[0047] Among them, large language models are usually natural language processing models with large-scale parameters and computing power. They can be trained with a large amount of data and parameters to analyze the structure, semantics, context and pragmatics of natural language and generate corresponding answers.

[0048] The preset large language model can be selected by the implementer according to the actual situation. Those skilled in the art know that any large language model in the prior art falls within the protection scope of the present invention, and will not be described in detail here.

[0049] Based on the preset large language model and the purpose of question and answer, the corresponding question template is obtained, the current abnormal event text is filled into the question template and then input into the preset large language model, and the initial subject corresponding to the current abnormal event text and the subject role type corresponding to the current initial subject output by the preset large language model are obtained to characterize the identity information of the current initial subject in the current abnormal event text. For example, the subject role type can refer to the executing subject role, that is, the executor of the abnormal event, or it can be the affected subject role, that is, the affected party of the abnormal event.

[0050] For example, the purpose of the question and answer in this embodiment is to obtain the initial subject corresponding to the current abnormal event text and the subject role type corresponding to the current initial subject. The corresponding question template can be "The known abnormal event text is:. Who is the subject corresponding to the above abnormal event? Is the subject role type of the corresponding subject the executing subject role or the affected subject role?".

[0051] In the above, the current abnormal event text is analyzed based on the preset large language model to obtain the initial subject corresponding to the current abnormal event text and the subject role type corresponding to the current initial subject, which provides a data basis for characterizing the abnormal degree of each initial subject.

[0052] S3, inputting the current abnormal event text and the corresponding subsequent event text into a preset large language model to obtain a first abnormality degree corresponding to the current abnormal event text.

[0053] The subsequent event text may refer to a description text of subsequent events that occur after an abnormal event occurs. By performing semantic analysis on the abnormal event text, the current abnormal impact caused by the abnormal event corresponding to the current abnormal event text may be analyzed.

[0054] In order to improve the accuracy of characterization of abnormal events, this embodiment performs semantic analysis on the current abnormal event text and the corresponding subsequent event text at the same time, and can analyze the subsequent abnormal impact of the abnormal event. The current abnormal impact and the subsequent abnormal impact are combined to obtain the first abnormal degree corresponding to the current abnormal event text, so as to characterize the magnitude of the abnormal degree of the abnormal event corresponding to the current abnormal event text.

[0055] In the above, the current abnormal event text and the corresponding subsequent event text are input into the preset large language model, and the current abnormal impact and subsequent abnormal impact of the abnormal event corresponding to the current abnormal event text are analyzed at the same time, and the first abnormal degree corresponding to the current abnormal event text is obtained, thereby improving the accuracy of the representation of the abnormal degree of the abnormal event corresponding to the current abnormal event text.

[0056] S4, determining the current initial subject as the abnormal source subject, and obtaining the second abnormal degree corresponding to the current abnormal source subject for the current abnormal event text according to the subject role type and the first abnormal degree corresponding to the current abnormal source subject.

[0057] In a specific implementation, the subject role types include an executing subject role and an affected subject role, and S4 further includes the following steps:

[0058] S41, obtaining a first preset influence degree corresponding to the executing subject role and a second preset influence degree corresponding to the affected subject role.

[0059] S42, according to the subject role type corresponding to the current abnormal source subject, obtaining the impact degree corresponding to the current abnormal source subject, wherein S42 also includes the following steps:

[0060] S421: If the subject role type corresponding to the current abnormal source subject is an execution subject role, the first preset impact degree is determined as the impact degree corresponding to the current abnormal source subject.

[0061] S422: If the subject role type corresponding to the current abnormal source subject is an affected subject role, the second preset impact degree is determined as the impact degree corresponding to the current abnormal source subject.

[0062] S43: Determine the product of the impact degree corresponding to the current abnormal source subject and the first abnormal degree as the second abnormal degree corresponding to the current abnormal source subject for the current abnormal event text.

[0063] Among them, the abnormality degree of the abnormal source subject itself and the abnormality degree transmitted to the associated initial subject are different when the abnormal source subject acts as the execution subject role and the abnormality degree transmitted to the associated initial subject role. Generally speaking, the corresponding abnormality degree and the corresponding abnormality degree transmitted are higher when the abnormal source subject acts as the execution subject role.

[0064] The specific values ​​of the first preset impact degree and the second preset impact degree can be set by the implementer according to the actual situation. Among them, the first preset impact degree is greater than the second preset impact degree.

[0065] As described above, based on the subject role type and the first abnormality degree corresponding to the current abnormal source subject, the second abnormality degree corresponding to the current abnormal source subject for the current abnormal event text is obtained, and the influence of the identity of the abnormal source subject in the abnormal event text on its own abnormality degree and the transmitted abnormality degree is taken into account, thereby improving the accuracy of the characterization of the abnormality degree.

[0066] S5, traverse all abnormal event texts, and obtain the third abnormal degree corresponding to all abnormal event texts of each abnormal source subject according to the second abnormal degree corresponding to each abnormal event text of the current abnormal source subject.

[0067] In a specific implementation, S5 further includes the following steps:

[0068] S51, for any abnormal source subject, obtaining all second abnormality degrees corresponding to the current abnormal source subject.

[0069] S52: Determine the sum of all second abnormality degrees corresponding to the current abnormal source subject as the third abnormality degree corresponding to the current abnormal source subject for all abnormal event texts.

[0070] In the above, the third abnormality degree corresponding to each abnormal source subject for all abnormal event texts is obtained to characterize the abnormality degree of each abnormal source subject itself, providing a data basis for identifying the target abnormality degree corresponding to each initial subject.

[0071] S6. Take any initial subject as the target subject to be identified, and according to the preset subject association map and the third abnormality degree corresponding to each abnormal source subject, obtain K association paths between the current target subject and each abnormal source subject, as well as the association type set, association degree set, first abnormal conduction degree set and third abnormality degree set corresponding to each association path, where K is an integer greater than 0.

[0072] In a specific implementation, S6 further includes the following steps:

[0073] S61, according to the preset subject association graph, obtain K association paths between the current target subject and each abnormal source subject.

[0074] S62, for any association path, in the order from the current abnormal source subject to the current target subject, all initial subjects in the current association path are determined as the first candidate subject, the second candidate subject, ..., the Tth candidate subject, wherein the current abnormal source subject is determined as the first candidate subject, and the current target subject is determined as the Tth candidate subject, and T refers to the total number of initial subjects corresponding to the current association path.

[0075] S63, according to the association types between the initial subjects connected in the preset subject association graph, obtain the association type set and the association degree set corresponding to the current association path, wherein the i-th association type in the association type set refers to the association type between the i-th candidate subject and the i+1-th candidate subject, and the i-th association degree in the association degree set refers to the association degree between the i-th candidate subject and the i+1-th candidate subject, i=1, 2, ..., T-1.

[0076] S64, acquiring a first abnormal conduction degree set corresponding to the current association path according to the association type set, wherein the i-th first abnormal conduction degree in the first abnormal conduction degree set refers to the first abnormal conduction degree corresponding to the i-th association type in the association type set.

[0077] S65, according to the third abnormality degree corresponding to each abnormal source subject, obtain the third abnormality degree set corresponding to the current association path, wherein the i-th third abnormality degree in the third abnormality degree set refers to the third abnormality degree corresponding to the i-th candidate subject, and the T-th third abnormality degree in the third abnormality degree set refers to the third abnormality degree corresponding to the current target subject.

[0078] Among them, the association path can represent the corresponding path when the abnormal source subject transmits its own abnormality to the target subject. The association type set refers to the set of association types between the initial subjects connected in the association path, which can be represented as the multi-layer association relationship between the abnormal source subject and the target subject. The association degree set refers to the set of association degrees between the initial subjects connected in the association path, which can be represented as the multi-layer association degree between the abnormal source subject and the target subject. The first abnormal conduction degree set refers to the set of first abnormal conduction degrees between the initial subjects connected in the association path, which can be represented as the multi-layer abnormal conduction degree between the abnormal source subject and the target subject. The third abnormal degree set refers to the third abnormal degree set corresponding to all the initial subjects on the association path.

[0079] As described above, based on the preset subject association map and the third abnormality degree corresponding to each abnormal source subject, K association paths between the current target subject and each abnormal source subject, as well as the association type set, association degree set, first abnormal conduction degree set and third abnormality degree set corresponding to each association path are obtained, which characterize the multi-layer association degree, multi-layer abnormal conduction degree and multi-layer abnormality degree between the abnormal source subject and the target subject, and serve as the data basis for measuring the abnormalities transmitted from other initial subjects to the target subject.

[0080] S7, obtaining the target abnormality degree corresponding to the current target subject according to the association degree set, the first abnormal conduction degree set and the third abnormality degree set corresponding to each association path.

[0081] Among them, by taking each initial subject as the target subject, the target abnormality degree corresponding to each initial subject can be obtained.

[0082] In a specific implementation, S7 further includes the following steps:

[0083] S71, determining the product of the i-th correlation degree, the i-th first abnormal conduction degree and the i-th third abnormal degree as the i-th first conduction abnormality.

[0084] S72, determining the sum of T-1 first conduction anomalies and the Tth third anomaly degree as the fourth anomaly degree of the current target subject for the current associated path;

[0085] S73, traversing all associated paths corresponding to the current target subject, and determining the sum of the fourth abnormality degrees of the K associated paths corresponding to the current target subject as the target abnormality degree corresponding to the current target subject.

[0086] As described above, based on the degree of association between the initial subjects connected in the association path, the first abnormality conduction degree and the third abnormality degree corresponding to all the initial subjects on the association path, the abnormalities transmitted to the target subject from other initial subjects except the target subject on the association path are measured, and combined with the third abnormality degree of the target subject itself, the target abnormality degree corresponding to the target subject is obtained, thereby improving the accuracy of the target abnormality degree.

[0087] In a specific implementation, the abnormality degree identification method further includes the following steps:

[0088] S10, input each abnormal event text and the corresponding subsequent event text into a preset large language model to obtain the matching degree between each abnormal event text and each associated type.

[0089] S20, for any association path, matching the abnormal source subject corresponding to each abnormal event text with the first candidate subject corresponding to the current association path, and obtaining the abnormal event text corresponding to the current association path.

[0090] S30, according to the matching degree between the abnormal event text corresponding to the current association path and each association type, and the association type set corresponding to the current association path, obtain a matching degree set corresponding to each association path, wherein the i-th matching degree in the matching degree set refers to the matching degree between the abnormal event text corresponding to the current association path and the i-th association type in the association type set.

[0091] S40, determining the product of the i-th correlation degree, the i-th first abnormal conduction degree, the i-th matching degree and the i-th third abnormal degree as the i-th second conduction abnormality.

[0092] S50, determining the sum of T-1 second conduction anomalies and the Tth third anomaly degree as the fifth anomaly degree of the current target subject for the current associated path;

[0093] S60, traversing all associated paths corresponding to the current target subject, and determining the sum of the fifth abnormality degrees of the K associated paths corresponding to the current target subject as the target abnormality degree corresponding to the current target subject.

[0094] Among them, the matching degree between the abnormal event text and the association type can be used to characterize the degree of influence of the abnormal event corresponding to the abnormal event text when the abnormal conduction occurs in the initial subject. For example, if the abnormal event belongs to an abnormal situation between corporate transactions, when the association type between the initial subjects is a transaction-related association type such as the first group relationship, the second group relationship, the first guarantee relationship, or the second guarantee relationship, the corresponding abnormal conduction degree is greater and the corresponding matching degree is greater. When the association type between the initial subjects is an association type unrelated to the transaction such as the first employment relationship, the second employment relationship, and the patent relationship, the corresponding abnormal conduction degree is smaller and the corresponding matching degree is smaller.

[0095] The matching degree set refers to the set of matching degrees between the initial subjects connected in the association path, which can be characterized as the multi-layer matching degrees between the abnormal source subject and the target subject.

[0096] As described above, based on the degree of association between the initial subjects connected in the association path, the degree of first abnormality transmission, the degree of matching, and the third abnormality degree corresponding to all the initial subjects on the association path, the abnormalities transmitted to the target subject from other initial subjects except the target subject on the association path are measured, and combined with the third abnormality degree of the target subject itself, the target abnormality degree corresponding to the target subject is obtained, thereby improving the accuracy of the target abnormality degree.

[0097] In the above, the current abnormal event text is analyzed based on the preset large language model, the initial subject corresponding to the current abnormal event text and the subject role type corresponding to the current initial subject are obtained, the current abnormal event text and the corresponding subsequent event text are input into the preset large language model, and the current abnormal impact and subsequent abnormal impact of the abnormal event corresponding to the current abnormal event text are analyzed at the same time, and the first abnormal degree corresponding to the current abnormal event text is obtained, which improves the accuracy of the representation of the abnormal degree of the abnormal event corresponding to the current abnormal event text, and determines the current initial subject as the abnormal source subject. According to the subject role type and the first abnormal degree corresponding to the current abnormal source subject, the second abnormal degree corresponding to the current abnormal event text of the current abnormal source subject is obtained, and all abnormal event texts are traversed, and the second abnormal degree corresponding to each abnormal event text of the current abnormal source subject is obtained according to the subject role type and the first abnormal degree. The third abnormality degree corresponding to each abnormal source subject for all abnormal event texts is obtained, and any initial subject is taken as the target subject to be identified. According to the preset subject association map and the third abnormality degree corresponding to each abnormal source subject, K association paths between the current target subject and each abnormal source subject, as well as the association type set, association degree set, first abnormal conduction degree set and third abnormality degree set corresponding to each association path are obtained. According to the association degree, first abnormal conduction degree, matching degree and the third abnormality degree corresponding to all initial subjects on the association path, the abnormalities transmitted to the target subject from other initial subjects except the target subject on the association path are measured, and the target abnormality degree corresponding to the target subject is obtained in combination with the third abnormality degree of the target subject itself, thereby improving the accuracy of the target abnormality degree.

[0098] Embodiment 2

[0099] The second embodiment provides an abnormality degree recognition device based on a large language model. The abnormality degree recognition device based on a large language model includes: Figure 2 As shown:

[0100] The first data acquisition module 21 is used to obtain M abnormal event texts corresponding to N initial subjects, subsequent event texts corresponding to each abnormal event text, and a preset subject association map, wherein the preset subject association map includes N initial subjects as nodes, association types and association degrees between connected initial subjects, and a first abnormal conduction degree corresponding to each association type, wherein N and M are integers greater than 0, and the first abnormal conduction degree is used to characterize the amplitude value of abnormal conduction between the corresponding connected initial subjects.

[0101] The large language model application module 22 is used to input the current abnormal event text into a preset large language model for any abnormal event text, and obtain the initial subject corresponding to the current abnormal event text and the subject role type corresponding to the current initial subject.

[0102] The first abnormality degree acquisition module 23 is used to input the current abnormal event text and the corresponding subsequent event text into a preset large language model to acquire the first abnormality degree corresponding to the current abnormal event text.

[0103] The second abnormality degree acquisition module 24 is used to determine the current initial subject as the abnormal source subject, and acquire the second abnormality degree corresponding to the current abnormal source subject for the current abnormal event text according to the subject role type and the first abnormality degree corresponding to the current abnormal source subject.

[0104] The third abnormality degree acquisition module 25 is used to traverse all abnormal event texts, and acquire the third abnormality degree corresponding to all abnormal event texts of each abnormal source subject according to the second abnormality degree corresponding to each abnormal event text of the current abnormal source subject.

[0105] The second data acquisition module 26 is used to take any initial subject as the target subject to be identified, and obtain K association paths between the current target subject and each abnormal source subject according to a preset subject association map and the third abnormality degree corresponding to each abnormal source subject, as well as an association type set, an association degree set, a first abnormal conduction degree set and a third abnormality degree set corresponding to each association path, wherein K is an integer greater than 0.

[0106] The target abnormality degree acquisition module 27 is used to acquire the target abnormality degree corresponding to the current target subject according to the association degree set corresponding to each association path, the first abnormal conduction degree set and the third abnormality degree set.

[0107] In a specific implementation, the association types include a first equity relationship and a second equity relationship, a first guarantee relationship and a second guarantee relationship, a first transaction relationship and a second transaction relationship, a litigation relationship and a patent relationship, and the first data acquisition module 21 further includes:

[0108] The first association degree acquisition submodule is used to determine the equity ratio between the connected initial entities as the corresponding association degree if the association type between the connected initial entities is the first equity relationship or the second equity relationship.

[0109] The second association degree acquisition submodule is used to determine the guarantee amount ratio between the connected initial entities as the corresponding association degree if the association type between the connected initial entities is the first guarantee relationship or the second guarantee relationship.

[0110] The third association degree acquisition submodule is used to determine the transaction amount ratio between the connected initial entities as the corresponding association degree if the association type between the connected initial entities is the first transaction relationship and the second transaction relationship.

[0111] The fourth association degree acquisition submodule is used to determine a preset value as the corresponding association degree if the association type between the connected initial entities is a litigation relationship or a patent relationship.

[0112] In a specific implementation, the subject role types include an executing subject role and an affected subject role, and the second abnormality degree acquisition module 24 further includes:

[0113] The second preset influence degree acquisition submodule is used to acquire the first preset influence degree corresponding to the execution subject role and the second preset influence degree corresponding to the affected subject role.

[0114] The influence degree acquisition submodule is used to obtain the influence degree corresponding to the current abnormal source subject according to the subject role type corresponding to the current abnormal source subject, wherein the influence degree acquisition submodule also includes:

[0115] The first impact degree acquisition unit is used to determine the first preset impact degree as the impact degree corresponding to the current abnormal source subject if the subject role type corresponding to the current abnormal source subject is an execution subject role.

[0116] The second influence degree acquisition unit is used to determine the second preset influence degree as the influence degree corresponding to the current abnormal source subject if the subject role type corresponding to the current abnormal source subject is the affected subject role.

[0117] The second abnormality degree acquisition submodule is used to determine the product of the impact degree corresponding to the current abnormal source subject and the first abnormality degree as the second abnormality degree corresponding to the current abnormal source subject for the current abnormal event text.

[0118] In a specific implementation, the third abnormality degree acquisition module 25 further includes:

[0119] The abnormality degree comprehensive acquisition module is used to acquire all second abnormality degrees corresponding to the current abnormality source subject for any abnormality source subject.

[0120] The third abnormality degree acquisition submodule is used to determine the sum of all second abnormality degrees corresponding to the current abnormal source subject as the third abnormality degree corresponding to the current abnormal source subject for all abnormal event texts.

[0121] In a specific implementation, the second data acquisition module 26 further includes:

[0122] The association path acquisition submodule is used to obtain K association paths between the current target subject and each abnormal source subject according to the preset subject association map.

[0123] The candidate subject acquisition submodule is used to determine, for any association path, all initial subjects in the current association path as the first candidate subject, the second candidate subject, ... the Tth candidate subject in the order from the current abnormal source subject to the current target subject, wherein the current abnormal source subject is determined as the first candidate subject, the current target subject is determined as the Tth candidate subject, and T refers to the total number of initial subjects corresponding to the current association path.

[0124] The first data acquisition submodule is used to obtain the association type set and the association degree set corresponding to the current association path according to the association type between the initial subjects connected in the preset subject association map, wherein the i-th association type in the association type set refers to the association type between the i-th candidate subject and the i+1-th candidate subject, and the i-th association degree in the association degree set refers to the association degree between the i-th candidate subject and the i+1-th candidate subject, i=1, 2,..., T-1.

[0125] The second data acquisition submodule is used to acquire a first abnormal conduction degree set corresponding to the current association path according to the association type set, wherein the i-th first abnormal conduction degree in the first abnormal conduction degree set refers to the first abnormal conduction degree corresponding to the i-th association type in the association type set.

[0126] The third data acquisition submodule is used to obtain the third abnormality degree set corresponding to the current association path according to the third abnormality degree corresponding to each abnormal source subject, wherein the i-th third abnormality degree in the third abnormality degree set refers to the third abnormality degree corresponding to the i-th candidate subject, and the T-th third abnormality degree in the third abnormality degree set refers to the third abnormality degree corresponding to the current target subject.

[0127] In a specific implementation, the target abnormality degree acquisition module 27 further includes:

[0128] The first conduction abnormality obtaining submodule is used to determine the product of the i-th correlation degree, the i-th first abnormal conduction degree and the i-th third abnormality degree as the i-th first conduction abnormality.

[0129] A fourth abnormality degree acquisition submodule, used to determine the sum of T-1 first conduction abnormalities and the Tth third abnormality degree as the fourth abnormality degree of the current target subject for the current associated path;

[0130] The target abnormality degree acquisition submodule is used to traverse all associated paths corresponding to the current target subject, and determine the sum of the fourth abnormality degrees of the K associated paths corresponding to the current target subject as the target abnormality degree corresponding to the current target subject.

[0131] In a specific embodiment, the abnormality degree identification device based on the large language model further includes:

[0132] The matching degree acquisition module is used to input each abnormal event text and the corresponding subsequent event text into a preset large language model to obtain the matching degree between each abnormal event text and each associated type.

[0133] The subject matching module is used to match the abnormal source subject corresponding to each abnormal event text with the first candidate subject corresponding to the current association path for any association path, and obtain the abnormal event text corresponding to the current association path.

[0134] The third data acquisition module is used to obtain a matching degree set corresponding to each association path according to the matching degree between the abnormal event text corresponding to the current association path and each association type, and the association type set corresponding to the current association path, wherein the i-th matching degree in the matching degree set refers to the matching degree between the abnormal event text corresponding to the current association path and the i-th association type in the association type set.

[0135] The second conduction degree acquisition module is used to determine the product of the i-th correlation degree, the i-th first abnormal conduction degree, the i-th matching degree and the i-th third abnormal degree as the i-th second conduction abnormality.

[0136] A fifth abnormality degree acquisition module, used to determine the sum of T-1 second conduction abnormalities and the Tth third abnormality degree as the fifth abnormality degree of the current target subject for the current associated path;

[0137] The target abnormality degree identification module is used to traverse all associated paths corresponding to the current target subject, and determine the sum of the fifth abnormality degrees of the K associated paths corresponding to the current target subject as the target abnormality degree corresponding to the current target subject.

[0138] It should be noted that the information interaction, execution process and other contents between the above modules are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0139] Embodiment 3

[0140] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the non-transitory computer-readable storage medium, and the at least one instruction or at least one program is loaded and executed by a processor to implement the steps:

[0141] S1, obtain M abnormal event texts corresponding to N initial subjects, subsequent event texts corresponding to each abnormal event text, and a preset subject association map, wherein the preset subject association map includes N initial subjects as nodes, association types and association degrees between connected initial subjects, and a first abnormal conduction degree corresponding to each association type, wherein N and M are integers greater than 0, and the first abnormal conduction degree is used to characterize the amplitude value of abnormal conduction between corresponding connected initial subjects.

[0142] S2, for any abnormal event text, input the current abnormal event text into a preset large language model, and obtain the initial subject corresponding to the current abnormal event text and the subject role type corresponding to the current initial subject.

[0143] S3, inputting the current abnormal event text and the corresponding subsequent event text into a preset large language model to obtain a first abnormality degree corresponding to the current abnormal event text.

[0144] S4, determining the current initial subject as the abnormal source subject, and obtaining the second abnormal degree corresponding to the current abnormal source subject for the current abnormal event text according to the subject role type and the first abnormal degree corresponding to the current abnormal source subject.

[0145] S5, traverse all abnormal event texts, and obtain the third abnormal degree corresponding to all abnormal event texts of each abnormal source subject according to the second abnormal degree corresponding to each abnormal event text of the current abnormal source subject.

[0146] S6. Take any initial subject as the target subject to be identified, and according to the preset subject association map and the third abnormality degree corresponding to each abnormal source subject, obtain K association paths between the current target subject and each abnormal source subject, as well as the association type set, association degree set, first abnormal conduction degree set and third abnormality degree set corresponding to each association path, where K is an integer greater than 0.

[0147] S7, obtaining the target abnormality degree corresponding to the current target subject according to the association degree set, the first abnormal conduction degree set and the third abnormality degree set corresponding to each association path.

[0148] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchronization Link) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0149] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0150] Embodiment 4

[0151] A fourth embodiment of the present invention provides an electronic device, which includes a processor and the non-transitory computer-readable storage medium in the third embodiment of the present invention.

[0152] The above are only preferred embodiments of the present invention, and are not intended to limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any technician familiar with the profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for identifying abnormality based on a large language model, characterized in that: The abnormality degree identification method comprises the following steps: S1, obtaining M abnormal event texts corresponding to N initial subjects, subsequent event texts corresponding to each abnormal event text, and a preset subject association map, wherein the preset subject association map includes N initial subjects as nodes, association types and association degrees between connected initial subjects, and a first abnormal conduction degree corresponding to each association type, wherein N and M are integers greater than 0, and the first abnormal conduction degree is used to characterize the amplitude value of abnormal conduction between corresponding connected initial subjects; S2, for any abnormal event text, input the current abnormal event text into a preset large language model, and obtain the initial subject corresponding to the current abnormal event text and the subject role type corresponding to the current initial subject; S3, inputting the current abnormal event text and the corresponding subsequent event text into a preset large language model to obtain a first abnormality degree corresponding to the current abnormal event text; S4, determining the current initial subject as the abnormal source subject, and obtaining the second abnormal degree corresponding to the current abnormal source subject for the current abnormal event text according to the subject role type and the first abnormal degree corresponding to the current abnormal source subject; S5, traversing all abnormal event texts, and obtaining the third abnormal degree corresponding to all abnormal event texts of each abnormal source subject according to the second abnormal degree corresponding to each abnormal event text of the current abnormal source subject; S6, taking any initial subject as the target subject to be identified, and obtaining K association paths between the current target subject and each abnormal source subject, and an association type set, an association degree set, a first abnormal conduction degree set, and a third abnormal degree set corresponding to each association path, according to the preset subject association map and the third abnormal degree corresponding to each abnormal source subject, wherein K is an integer greater than 0, and S6 further includes the following steps: S61, according to the preset subject association graph, obtaining K association paths between the current target subject and each abnormal source subject; S62, for any association path, in the order from the current abnormal source subject to the current target subject, all initial subjects in the current association path are determined as the first candidate subject, the second candidate subject, ..., the Tth candidate subject, wherein the current abnormal source subject is determined as the first candidate subject, and the current target subject is determined as the Tth candidate subject, where T refers to the total number of initial subjects corresponding to the current association path; S63, according to the association types between the initial subjects connected in the preset subject association graph, obtaining an association type set and an association degree set corresponding to the current association path, wherein the i-th association type in the association type set refers to the association type between the i-th candidate subject and the i+1-th candidate subject, and the i-th association degree in the association degree set refers to the association degree between the i-th candidate subject and the i+1-th candidate subject, i=1, 2, ..., T-1; S64, acquiring a first abnormal conduction degree set corresponding to the current association path according to the association type set, wherein the i-th first abnormal conduction degree in the first abnormal conduction degree set refers to the first abnormal conduction degree corresponding to the i-th association type in the association type set; S65, according to the third abnormality degree corresponding to each abnormal source subject, obtaining a third abnormality degree set corresponding to the current association path, wherein the i-th third abnormality degree in the third abnormality degree set refers to the third abnormality degree corresponding to the i-th candidate subject, and the T-th third abnormality degree in the third abnormality degree set refers to the third abnormality degree corresponding to the current target subject; S10, inputting each abnormal event text and the corresponding subsequent event text into a preset large language model to obtain the matching degree between each abnormal event text and each associated type; S20, for any association path, matching the abnormal source subject corresponding to each abnormal event text with the first candidate subject corresponding to the current association path, and obtaining the abnormal event text corresponding to the current association path; S30, according to the matching degree between the abnormal event text corresponding to the current association path and each association type, and the association type set corresponding to the current association path, obtaining a matching degree set corresponding to each association path, wherein the i-th matching degree in the matching degree set refers to the matching degree between the abnormal event text corresponding to the current association path and the i-th association type in the association type set; S40, determining the product of the i-th correlation degree, the i-th first abnormal conduction degree, the i-th matching degree and the i-th third abnormal degree as the i-th second conduction abnormality; S50, determining the sum of T-1 second conduction anomalies and the Tth third anomaly degree as the fifth anomaly degree of the current target subject for the current associated path; S60, traversing all associated paths corresponding to the current target subject, and determining the sum of the fifth abnormality degrees of the K associated paths corresponding to the current target subject as the target abnormality degree corresponding to the current target subject.

2. The method for identifying abnormality based on a large language model according to claim 1, characterized in that: The types of association include a first equity relationship and a second equity relationship, a first guarantee relationship and a second guarantee relationship, a first transaction relationship and a second transaction relationship, a litigation relationship and a patent relationship. S1 also includes the following steps: S11, if the type of association between the connected initial entities is the first equity relationship or the second equity relationship, the equity ratio between the connected initial entities is determined as the corresponding degree of association; S12: If the type of association between the connected initial entities is a first guarantee relationship or a second guarantee relationship, the guarantee amount ratio between the connected initial entities is determined as the corresponding degree of association; S13, if the type of association between the connected initial entities is the first transaction relationship and the second transaction relationship, determining the proportion of the transaction amount between the connected initial entities as the corresponding degree of association; S14: If the type of association between the connected initial entities is a litigation relationship or a patent relationship, a preset value is determined as the corresponding degree of association.

3. The method for identifying abnormality based on a large language model according to claim 1, characterized in that: The subject role types include an executing subject role and an affected subject role. S4 further includes the following steps: S41, obtaining a first preset influence degree corresponding to the executing subject role and a second preset influence degree corresponding to the affected subject role; S42, according to the subject role type corresponding to the current abnormal source subject, obtaining the impact degree corresponding to the current abnormal source subject, wherein S42 also includes the following steps: S421, if the subject role type corresponding to the current abnormal source subject is an execution subject role, determining the first preset impact degree as the impact degree corresponding to the current abnormal source subject; S422: If the subject role type corresponding to the current abnormal source subject is an affected subject role, the second preset impact degree is determined as the impact degree corresponding to the current abnormal source subject; S43: Determine the product of the impact degree corresponding to the current abnormal source subject and the first abnormal degree as the second abnormal degree corresponding to the current abnormal source subject for the current abnormal event text.

4. The abnormality degree identification method based on a large language model according to claim 3 is characterized in that: S5 also includes the following steps: S51, for any abnormal source subject, obtaining all second abnormality degrees corresponding to the current abnormal source subject; S52: Determine the sum of all second abnormality degrees corresponding to the current abnormal source subject as the third abnormality degree corresponding to the current abnormal source subject for all abnormal event texts.

5. An abnormality degree recognition device based on a large language model, characterized in that: The abnormality degree identification device comprises: A first data acquisition module is used to acquire M abnormal event texts corresponding to N initial subjects, subsequent event texts corresponding to each abnormal event text, and a preset subject association map, wherein the preset subject association map includes N initial subjects as nodes, association types and association degrees between connected initial subjects, and a first abnormal conduction degree corresponding to each association type, wherein N and M are integers greater than 0, and the first abnormal conduction degree is used to characterize the amplitude value of abnormal conduction between the corresponding connected initial subjects; The large language model application module is used to input the current abnormal event text into a preset large language model for any abnormal event text, and obtain the initial subject corresponding to the current abnormal event text and the subject role type corresponding to the current initial subject; A first abnormality degree acquisition module, used to input the current abnormal event text and the corresponding subsequent event text into a preset large language model, and acquire a first abnormality degree corresponding to the current abnormal event text; A second abnormality degree acquisition module is used to determine the current initial subject as the abnormal source subject, and acquire the second abnormality degree corresponding to the current abnormal source subject for the current abnormal event text according to the subject role type and the first abnormality degree corresponding to the current abnormal source subject; A third abnormality degree acquisition module is used to traverse all abnormal event texts, and acquire the third abnormality degree corresponding to all abnormal event texts of each abnormal source subject according to the second abnormality degree corresponding to each abnormal event text of the current abnormal source subject; The second data acquisition module is used to take any initial subject as a target subject to be identified, and acquire K association paths between the current target subject and each abnormal source subject according to the preset subject association map and the third abnormality degree corresponding to each abnormal source subject, as well as an association type set, an association degree set, a first abnormal conduction degree set and a third abnormality degree set corresponding to each association path, wherein K is an integer greater than 0. The second data acquisition module includes: The association path acquisition submodule is used to acquire K association paths between the current target subject and each abnormal source subject according to the preset subject association map; The candidate subject acquisition submodule is used to determine, for any association path, all initial subjects in the current association path as the first candidate subject, the second candidate subject, ..., the Tth candidate subject in the order from the current abnormal source subject to the current target subject, wherein the current abnormal source subject is determined as the first candidate subject, the current target subject is determined as the Tth candidate subject, and T refers to the total number of initial subjects corresponding to the current association path; A first data acquisition submodule is used to acquire an association type set and an association degree set corresponding to a current association path according to the association type between the initial subjects connected in the preset subject association graph, wherein the i-th association type in the association type set refers to the association type between the i-th candidate subject and the i+1-th candidate subject, and the i-th association degree in the association degree set refers to the association degree between the i-th candidate subject and the i+1-th candidate subject, i=1, 2, ..., T-1; A second data acquisition submodule is configured to acquire a first abnormal conduction degree set corresponding to a current associated path according to the association type set, wherein an i-th first abnormal conduction degree in the first abnormal conduction degree set refers to a first abnormal conduction degree corresponding to an i-th association type in the association type set; A third data acquisition submodule is used to acquire a third abnormality degree set corresponding to the current association path according to the third abnormality degree corresponding to each abnormal source subject, wherein the i-th third abnormality degree in the third abnormality degree set refers to the third abnormality degree corresponding to the i-th candidate subject, and the T-th third abnormality degree in the third abnormality degree set refers to the third abnormality degree corresponding to the current target subject; A matching degree acquisition module is used to input each abnormal event text and the corresponding subsequent event text into a preset large language model to obtain the matching degree between each abnormal event text and each associated type; The subject matching module is used to match the abnormal source subject corresponding to each abnormal event text with the first candidate subject corresponding to the current association path for any association path, and obtain the abnormal event text corresponding to the current association path; A third data acquisition module is used to acquire a matching degree set corresponding to each association path according to the matching degree between the abnormal event text corresponding to the current association path and each association type, and the association type set corresponding to the current association path, wherein the i-th matching degree in the matching degree set refers to the matching degree between the abnormal event text corresponding to the current association path and the i-th association type in the association type set; A second conduction degree acquisition module, used to determine the product of the i-th correlation degree, the i-th first abnormal conduction degree, the i-th matching degree and the i-th third abnormal degree as the i-th second conduction abnormality; A fifth abnormality degree acquisition module, used to determine the sum of T-1 second conduction abnormalities and the Tth third abnormality degree as the fifth abnormality degree of the current target subject for the current associated path; The target abnormality degree identification module is used to traverse all associated paths corresponding to the current target subject, and determine the sum of the fifth abnormality degrees of the K associated paths corresponding to the current target subject as the target abnormality degree corresponding to the current target subject.

6. A non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the non-transitory computer-readable storage medium, characterized in that: The at least one instruction or at least one program is loaded and executed by the processor to implement the abnormality degree identification method based on a large language model as described in any one of claims 1-4.

7. An electronic device, characterized in that: The invention comprises a processor and the non-transitory computer-readable storage medium as claimed in claim 6.