A dynamic risk early warning method and system

Abstract

The application provides a dynamic risk early warning method and system, the method comprises the following steps: collecting multi-mode heterogeneous data and preprocessing the multi-mode heterogeneous data; determining a risk event vector corresponding to the same event based on the preprocessed multi-mode heterogeneous data; embedding and constraining the first type of language data in the risk event vector using a knowledge graph to generate target embedding features aligned with the risk features of the second type of language data; determining the attention weights of the multi-dimensional risk features in the risk event vector and the embedding features; determining the comprehensive risk score of the risk event by combining the multi-dimensional risk features and the corresponding attention weights; inputting the comprehensive risk score into a pre-constructed sequence prediction model to obtain the comprehensive risk score trend of the corresponding data in the future T time windows; determining the risk escalation probability by combining the comprehensive risk score trend; and performing corresponding early warning based on the comprehensive risk score trend and the risk escalation probability.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of computer technology, and in particular to a dynamic risk warning method and system. Background Technology

[0002] In international energy, finance, and manufacturing industries, risk sources are often dispersed across multimodal and multilingual publicly available information. Existing risk monitoring systems primarily rely on single text, general machine translation, and static indicator scoring, which cannot effectively address the following challenges: Intermodal semantic gap: It is difficult to achieve deep semantic alignment between suggestive information in images, contextual information in videos, and text content; Distortion in less commonly taught languages: General translation models lack domain knowledge of specialized terms in fields such as energy and geopolitics (e.g., sanctions, compliance, nationalization), leading to semantic shifts after translation and weakening or misinterpreting risk signals; Static indicator assessment: Risk score weights are fixed, and the assessment focus cannot be dynamically adjusted according to the real-time evolution speed of risk events and information consistency. Summary of the Invention

[0003] To address the aforementioned technical problems, embodiments of the present invention provide a dynamic risk warning method, comprising: Collect multi-modal heterogeneous data and preprocess the multi-modal heterogeneous data; Based on the preprocessed multimodal heterogeneous data, determine the risk event vector corresponding to the same event; The first type of language data in the risk event vector is embedded using a knowledge graph to generate target embedding features that are aligned with the risk features of the second type of language data; Determine the attention weights of the multidimensional risk features in the embedded features of the risk event vector; The comprehensive risk score of the risk event is determined by combining the multidimensional risk characteristics and the corresponding attention weights. The comprehensive risk score is input into a pre-built sequence prediction model to obtain the comprehensive risk score trend of the data corresponding to the next T time windows; The probability of risk escalation is determined by combining the overall risk score trend; Early warnings will be issued based on the overall risk score trend and the probability of risk escalation.

[0004] In one embodiment, determining the risk event vector corresponding to the same event based on the preprocessed multimodal heterogeneous data includes: The different modal data in the multimodal heterogeneous data are encoded respectively, and the encoding form matches the data type of the corresponding modality; By applying a self-supervised modality contrastive learning loss function, the encoded data of different modalities corresponding to the same event are aligned in a shared semantic space to generate the risk event vector.

[0005] In one embodiment, the step of using a knowledge graph to embed the first type of language data in the risk event vector into target embedding features aligned with the risk features of the second type of language data includes: The first type of language data in the risk event vector is identified, and the first type of language data is minority language data. The first type of language data is processed by calling a pre-built M-MLM model, and the embedding constraints are performed using the knowledge graph to obtain target embedding features that are aligned with the risk features of the second type of language data. The M-MLM model is obtained by secondary training on corpora in the energy and finance fields, and the domain of the knowledge graph is the same as the domain to which the current risk event belongs.

[0006] In one embodiment, the method further includes: Multidimensional risk features are extracted from the risk event vector and embedded features; The multidimensional risk characteristics include one or more of the following: sentiment indicator characteristics, event nature characteristics, information consistency characteristics, entity correlation characteristics, and temporal suddenness characteristics.

[0007] In one embodiment, the emotion index features include the proportion of negative emotions, the amplitude of emotion fluctuations, and the intensity of extreme emotions; The characteristics of the event include the formality of the event's timing, the severity of the event, and the scope of the event's impact; The information consistency feature is a data source credibility indicator feature, including cross-platform consistency, verifiability of source information, and degree of unprocessed information. The entity correlation characteristics include the correlation between the risk entity and the designated enterprise, the proximity of the entity and the designated enterprise in terms of location and progress, and the importance of the role of the entity. The information consistency feature belongs to the time decay and suddenness index features, including the freshness of the event, the information growth rate of the event, and the suddenness coefficient of the event.

[0008] In one embodiment, determining the attention weights of the multidimensional risk features embedded in the risk event vector includes: The pre-built neural network model is invoked to process the risk event vector and embedded features, and the attention weight of each feature in the multidimensional risk features is dynamically determined. The neural network model includes a DARA module, and the attention weight is determined based on the DARA module.

[0009] In one embodiment, determining the comprehensive risk score of the risk event by combining the multidimensional risk features and the corresponding attention weights includes: Determine the preset score corresponding to each risk characteristic; The comprehensive risk score of the risk event is calculated by combining the preset score, multidimensional risk features, and corresponding attention weights, using the following formula. :

[0010] For the first i Attention weights for each feature For the first i Preset scores for each feature.

[0011] In one embodiment, the method further includes: The comprehensive risk score is adjusted for timeliness using the following dynamic decay function to obtain the corrected result. The final comprehensive risk score :

[0012] The attenuation coefficient is positively correlated with the propagation speed and importance of the event involved.

[0013] In one embodiment, the step of issuing a corresponding early warning based on the comprehensive risk score trend and the probability of risk escalation includes: The target early warning level is determined based on the comprehensive risk score and the probability of risk escalation. Whether to trigger an alarm event is determined based on the target warning level; In response to the triggering of an alarm event, the alarm push content is determined based on the comprehensive risk score and risk escalation probability, and corresponding reports are generated, and the alarm event is archived.

[0014] Another embodiment of the present invention also provides a dynamic risk early warning system, comprising: The acquisition module is used to acquire multi-modal heterogeneous data and preprocess the multi-modal heterogeneous data; The first determining module is used to determine the risk event vector corresponding to the same event based on the preprocessed multimodal heterogeneous data; The generation module is used to embed the first type of language data in the risk event vector using a knowledge graph to generate target embedding features that are aligned with the risk features of the second type of language data. The second determining module is used to determine the attention weights of the multidimensional risk features in the embedded features of the risk event vector; The third determining module is used to determine the comprehensive risk score of the risk event by combining the multidimensional risk features and the corresponding attention weights. The input module is used to input the comprehensive risk score into a pre-built sequence prediction model to obtain the comprehensive risk score trend of the data corresponding to the next T time windows; The fourth determining module is used to determine the probability of risk escalation by combining the trend of the comprehensive risk score; The early warning module is used to issue corresponding early warnings based on the comprehensive risk score trend and the probability of risk escalation.

[0015] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0016] The technical solution of this application will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the dynamic risk warning method in an embodiment of the present invention.

[0019] Figure 2 This is a flowchart illustrating a dynamic risk warning method according to another embodiment of the present invention.

[0020] Figure 3 This is a flowchart illustrating the dynamic risk warning method in another embodiment of the present invention.

[0021] Figure 4 This is a structural block diagram of the dynamic risk warning system in an embodiment of the present invention. Detailed Implementation

[0022] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings, but these are not intended to limit the scope of the invention.

[0023] It should be understood that various modifications can be made to the embodiments disclosed herein. Therefore, the following description should not be considered as limiting, but merely as an example of embodiments. Other modifications within the scope of this disclosure will be apparent to those skilled in the art.

[0024] The accompanying drawings, which are included in and form part of this specification, illustrate embodiments of the present disclosure and, together with the general description of the disclosure given above and the detailed description of the embodiments given below, serve to explain the principles of the disclosure.

[0025] These and other features of the invention will become apparent from the following description of preferred forms of embodiments given as non-limiting examples, with reference to the accompanying drawings.

[0026] It should also be understood that although the invention has been described with reference to some specific examples, those skilled in the art can certainly implement many other equivalent forms of the invention, which have the features described in the claims and are therefore all within the scope of protection defined herein.

[0027] The above and other aspects, features and advantages of this disclosure will become more apparent when taken in conjunction with the accompanying drawings and in view of the following detailed description.

[0028] Specific embodiments of the present disclosure are described thereafter with reference to the accompanying drawings; however, it should be understood that the disclosed embodiments are merely examples of the present disclosure and can be implemented in various ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that could obscure the present disclosure. Therefore, the specific structural and functional details disclosed herein are not intended to be limiting, but merely to serve as the basis and representative basis for the claims to teach those skilled in the art to use the present disclosure in a variety of substantially any suitable detailed structures.

[0029] This specification may use the phrases “in one embodiment,” “in another embodiment,” “in yet another embodiment,” or “in still another embodiment,” all of which may refer to one or more of the same or different embodiments according to this disclosure.

[0030] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0031] like Figure 1 As shown, this embodiment of the invention provides a dynamic risk warning method, including: S1: Collect multi-modal heterogeneous data and preprocess the multi-modal heterogeneous data; S2: Determine the risk event vector corresponding to the same event based on the preprocessed multimodal heterogeneous data; S3: Use a knowledge graph to embed the first type of language data in the risk event vector into a target embedding feature that is aligned with the risk features of the second type of language data; S4: Determine the attention weights of the multidimensional risk features in the embedded features of the risk event vector; S5: Combine the multidimensional risk characteristics and corresponding attention weights to determine the comprehensive risk score of the risk event; S6: Input the comprehensive risk score into the pre-built sequence prediction model to obtain the comprehensive risk score trend of the data corresponding to the next T time windows; S7: Determine the probability of risk escalation by combining the aforementioned comprehensive risk score trend; S8: Issue a corresponding early warning based on the comprehensive risk score trend and the probability of risk escalation.

[0032] This embodiment collects multimodal heterogeneous data from publicly available sources such as the global internet, news media, and regulatory announcements. The collected multimodal heterogeneous data is then cleaned, format-converted, and standardized to prepare for subsequent modality coding. This process constitutes the data preprocessing. After preprocessing, the system sequentially encodes, identifies, aligns, confirms feature weights, calculates a comprehensive risk score, calculates the comprehensive risk score trend, and calculates the probability of risk escalation, ultimately making a risk warning judgment based on this.

[0033] Based on the above-mentioned scheme, it can be seen that the scheme can construct a unified semantic space, eliminate cross-modal information redundancy and semantic ambiguity; achieve zero-distortion embedding and semantic alignment of specialized terms in minority languages; establish a dynamic attention risk assessment (DARA) mechanism to achieve adaptive weight adjustment; and provide forward-looking prediction capabilities for risk evolution trends, enabling early warning and effective risk avoidance.

[0034] In one embodiment, such as Figure 2 As shown, determining the risk event vector corresponding to the same event based on the preprocessed multimodal heterogeneous data includes: S201: Encode the different modal data in the multimodal heterogeneous data respectively, and match the encoding form with the data type of the corresponding modality; S202: Apply the self-supervised modality contrastive learning loss function to align the encoded data of different modalities corresponding to the same event in the shared semantic space to generate the risk event vector.

[0035] For example, different modalities are encoded separately, using corresponding encoding methods for different data types. For instance, text data is encoded using Transformer, visual data using ViT, and audio data using ASR / Audio-ViT, etc., with no specific limitation. After encoding, the system applies a self-supervised cross-modal contrastive learning loss function. Aligning different modal features from the same event in a shared semantic space generates a unified, high-density risk event vector. .

[0036] Furthermore, the step of using a knowledge graph to embed the first type of language data in the risk event vector into target embedding features aligned with the risk features of the second type of language data includes: S301: Identify the first type of language data in the risk event vector, wherein the first type of language data is minority language data; S302: Call the pre-built M-MLM model to process the first type of language data, and use the knowledge graph to perform embedding constraints to obtain target embedding features that are aligned with the risk features of the second type of language data. The M-MLM model is obtained by secondary training on corpora in the energy and financial fields, and the domain of the knowledge graph is the same as the domain to which the current risk event belongs.

[0037] When applying, it will automatically identify The included minority language data may include, but is not limited to, Spanish and Portuguese data. The content from these minority languages ​​is then input into an M-MLM model that has undergone secondary pre-training (Domain-Adaptive Pre-training) of corpora from the energy, finance, and other fields; specifically, it is input into the model's embedding layer. This allows the model to perform feature embedding. During the embedding process, this embodiment also utilizes a domain knowledge graph to constrain the embedding process, generating zero-distortion cross-language risk embeddings and risk features aligned with the target language, i.e., the second type of language features. The second type of language is a non-minority language, such as Chinese / English.

[0038] After embedding risk features, the system further identifies risk embedding features and risk features in the risk event vector, and performs risk assessment accordingly. Specifically, the method also includes: S9: Extract multidimensional risk features from the risk event vector and embedded features; S10: The multidimensional risk characteristics include one or more of the following: sentiment index characteristics, event nature characteristics, information consistency characteristics, entity correlation characteristics, and temporal suddenness characteristics.

[0039] Specifically, the emotional indicator features in this embodiment include the proportion of negative emotions, the amplitude of emotional fluctuations, and the intensity of extreme emotions. The proportion of negative emotions includes, for example, worry, panic, and condemnation. The amplitude of emotional fluctuations includes, for example, emotional changes over time. The intensity of extreme emotions includes, for example, high-risk words such as "bankruptcy," "illegal," and "crisis." Public opinion risks usually have a higher weight under these indicators. That is, in the case of public opinion risk monitoring, the emotional indicator features have a higher weight.

[0040] The characteristics of the event include the formality of the event's timing, the severity of the event, and the scope of the event's impact. The formality of the event can be, for example, data. The severity of the event can be, for example, economic penalties, business interruptions, or supply chain failures. The scope of the event's impact can be, for example, internal, industry-wide, or national-level.

[0041] The information consistency feature is a data source credibility indicator feature, including cross-platform consistency, verifiability of source information, and the degree of unprocessed information. Cross-platform consistency includes whether multiple language and national media outlets report the same event; verifiability of source information includes, for example, whether it contains the name of an authoritative organization; and the degree of unprocessed information includes, for example, whether it contains a large amount of inferential description. This indicator is primarily used to measure "whether the content is reliable," rather than the credibility of the website itself.

[0042] The entity relevance features include the relevance between the risky entity and the designated company, the proximity of the entity to the designated company, and the importance of the entity's role. For example, based on NLP and minority language parsing, an entity chain can be established, where the relevance between the risky entity and the target company includes the relevance between the risky entity and suppliers, parent company, partners, and the risky entity; the proximity of the entity's location includes the co-occurrence distance with a company name in the text; and the importance of the entity's role includes, for example, core suppliers > peripheral partners, etc.

[0043] The information consistency characteristic belongs to the time decay and suddenness indicator characteristics, including the freshness of the event, the information growth rate of the event, and the suddenness coefficient of the event. These indicators are used for dynamic early warning. The event freshness includes events that have occurred recently, with higher weighting; the consultation growth rate includes situations where consultations occur in a concentrated manner within a short period; and the suddenness coefficient characterizes whether the event is a risky event that is occurring for the first time.

[0044] Based on the actual data type, after the system uses the above indicators to identify multidimensional features, it will further calculate the attention weights of each dimension of features. Specifically, determining the attention weights of the multidimensional risk features embedded in the risk event vector includes: S401: Call a pre-built neural network model to process the risk event vector and embedded features, and dynamically determine the attention weight of each feature in the multidimensional risk features. The neural network model includes a DARA module, and the attention weight is determined based on the DARA module.

[0045] After determining the attention weights for different features across each dimension, the system combines the multidimensional risk features and their corresponding attention weights to determine the comprehensive risk score for the risk event, including: S501: Determine the preset score corresponding to each risk characteristic; S502: Combining the preset score, multidimensional risk features, and corresponding attention weights, the comprehensive risk score of the risk event is calculated using the following formula. :

[0046] For the first i Attention weights for each feature For the first i Preset scores for each feature.

[0047] For example, a score table will be pre-built for each indicator's different risk characteristics. Based on these different risk characteristics, the system can determine the corresponding preset / initial score from this score table. Next, combining the initial score with each risk characteristic and its corresponding attention weight, the comprehensive risk score for the corresponding risk event is calculated using the above formula.

[0048] As an optional embodiment, the method further includes: S503: Apply the following dynamic decay function to perform a timeliness correction on the comprehensive risk score, resulting in... To the revised comprehensive risk score :

[0049] The attenuation coefficient is positively correlated with the propagation speed and importance of the event involved.

[0050] The steps described in this embodiment are to use a dynamic decay function to correct the timeliness of the comprehensive risk score calculated by the above formula, and finally obtain an accurate comprehensive risk score that conforms to the current actual risk situation.

[0051] Furthermore, such as Figure 3 As shown, after obtaining the comprehensive risk score, the system will issue corresponding warnings based on the trend of the comprehensive risk score and the probability of risk escalation, including: S801: Determine the target early warning level based on the comprehensive risk score and the probability of risk escalation; S802: Determine whether to trigger an alarm event based on the target warning level; S803: In response to the triggering of an alarm event, determine the alarm push content based on the comprehensive risk score and risk escalation probability, generate the corresponding report, and complete the archiving of the alarm event.

[0052] For example, input past time window sequences into a Transformer sequence prediction model. Predict the risk score trend for the next T time windows and calculate the probability of risk level escalation. Based on this, classify warning levels and trigger alarm push notifications, report generation, and event archiving.

[0053] Based on the content disclosed in the above embodiments, it is clear that the solution of this application can be truly implemented: Multimodal deep semantic fusion (self-supervised alignment): By introducing VLP / ViT architecture and cross-modal contrastive learning, the semantic alignment problem of multimodal information is fundamentally solved, and the accuracy of risk information extraction is greatly improved.

[0054] High-precision semantic embedding for minor languages ​​(domain adaptive): Employing domain-specific pre-training and knowledge graph constraints, this ensures zero distortion of energy-related terminology in minor language processing and effectively detects hidden risks.

[0055] Dynamic Adaptive Risk Indicator Weights (DARA Mechanism): For the first time, a self-attention mechanism is introduced to dynamically allocate risk indicator weights in real time, enabling the system to maintain optimal sensitivity and recognition efficiency across different risk types and evolution stages.

[0056] It possesses the ability to predict risk evolution trends in advance: By combining risk warning with sequence prediction models, it can predict the likelihood and timing of risk outbreaks, giving users valuable lead time for decision-making.

[0057] like Figure 4 As shown, another embodiment of the present invention also provides a dynamic risk early warning system, including: The acquisition module is used to acquire multi-modal heterogeneous data and preprocess the multi-modal heterogeneous data; The first determining module is used to determine the risk event vector corresponding to the same event based on the preprocessed multimodal heterogeneous data; The generation module is used to embed the first type of language data in the risk event vector using a knowledge graph to generate target embedding features that are aligned with the risk features of the second type of language data. The second determining module is used to determine the attention weights of the multidimensional risk features in the embedded features of the risk event vector; The third determining module is used to determine the comprehensive risk score of the risk event by combining the multidimensional risk features and the corresponding attention weights. The input module is used to input the comprehensive risk score into a pre-built sequence prediction model to obtain the comprehensive risk score trend of the data corresponding to the next T time windows; The fourth determining module is used to determine the probability of risk escalation by combining the trend of the comprehensive risk score; The early warning module is used to issue corresponding early warnings based on the comprehensive risk score trend and the probability of risk escalation.

[0058] In one embodiment, determining the risk event vector corresponding to the same event based on the preprocessed multimodal heterogeneous data includes: The different modal data in the multimodal heterogeneous data are encoded respectively, and the encoding form matches the data type of the corresponding modality; By applying a self-supervised modality contrastive learning loss function, the encoded data of different modalities corresponding to the same event are aligned in a shared semantic space to generate the risk event vector.

[0059] In one embodiment, the step of using a knowledge graph to embed the first type of language data in the risk event vector into target embedding features aligned with the risk features of the second type of language data includes: The first type of language data in the risk event vector is identified, and the first type of language data is minority language data. The first type of language data is processed by calling a pre-built M-MLM model, and the embedding constraints are performed using the knowledge graph to obtain target embedding features that are aligned with the risk features of the second type of language data. The M-MLM model is obtained by secondary training on corpora in the energy and finance fields, and the domain of the knowledge graph is the same as the domain to which the current risk event belongs.

[0060] In one embodiment, the device further includes: The extraction module is used to extract multidimensional risk features from the risk event vector and embedded features; The multidimensional risk characteristics include one or more of the following: sentiment indicator characteristics, event nature characteristics, information consistency characteristics, entity correlation characteristics, and temporal suddenness characteristics.

[0061] In one embodiment, the emotion index features include the proportion of negative emotions, the amplitude of emotion fluctuations, and the intensity of extreme emotions; The characteristics of the event include the formality of the event's timing, the severity of the event, and the scope of the event's impact; The information consistency feature is a data source credibility indicator feature, including cross-platform consistency, verifiability of source information, and degree of unprocessed information. The entity correlation characteristics include the correlation between the risk entity and the designated enterprise, the proximity of the entity and the designated enterprise in terms of location and progress, and the importance of the role of the entity. The information consistency feature belongs to the time decay and suddenness index features, including the freshness of the event, the information growth rate of the event, and the suddenness coefficient of the event.

[0062] In one embodiment, determining the attention weights of the multidimensional risk features embedded in the risk event vector includes: The pre-built neural network model is invoked to process the risk event vector and embedded features, and the attention weight of each feature in the multidimensional risk features is dynamically determined. The neural network model includes a DARA module, and the attention weight is determined based on the DARA module.

[0063] In one embodiment, determining the comprehensive risk score of the risk event by combining the multidimensional risk features and the corresponding attention weights includes: Determine the preset score corresponding to each risk characteristic; The comprehensive risk score of the risk event is calculated by combining the preset score, multidimensional risk features, and corresponding attention weights, using the following formula. :

[0064] For the first i Attention weights for each feature For the first i Preset scores for each feature.

[0065] In one embodiment, the device further includes: The correction module is used to apply the following dynamic decay function to perform a timeliness correction on the comprehensive risk score, resulting in a corrected score. The final comprehensive risk score :

[0066] The attenuation coefficient is positively correlated with the propagation speed and importance of the event involved.

[0067] In one embodiment, the step of issuing a corresponding early warning based on the comprehensive risk score trend and the probability of risk escalation includes: The target early warning level is determined based on the comprehensive risk score and the probability of risk escalation. Whether to trigger an alarm event is determined based on the target warning level; In response to the triggering of an alarm event, the alarm push content is determined based on the comprehensive risk score and risk escalation probability, and corresponding reports are generated, and the alarm event is archived.

[0068] Another embodiment of the present invention also provides an electronic device, comprising: One or more processors; Memory, configured to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the dynamic risk warning method as described in any one of the above descriptions.

[0069] Furthermore, one embodiment of the present invention also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the dynamic risk warning method described above. It should be understood that the various solutions in this embodiment have the corresponding technical effects in the above-described method embodiments, and will not be repeated here.

[0070] Furthermore, embodiments of the present invention also provide a computer program product, which is tangibly stored on a computer-readable medium and includes computer-readable instructions that, when executed, cause at least one processor to perform a dynamic risk warning method such as those described in the embodiments above.

[0071] It should be noted that the computer storage medium of the present invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access storage medium (RAM), a read-only storage medium (ROM), an erasable programmable read-only storage medium (EPROM or flash memory), an optical fiber, a portable compact disk read-only storage medium (CD-ROM), an optical storage medium, a magnetic storage medium, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. In the present invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program configured for use by or in connection with an instruction execution system, system, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, antenna, optical fiber, RF, etc., or any suitable combination thereof.

[0072] Furthermore, those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0073] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0074] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0075] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

Claims

1. A dynamic risk early warning method, characterized in that, include: Collect multi-modal heterogeneous data and preprocess the multi-modal heterogeneous data; Based on the preprocessed multimodal heterogeneous data, determine the risk event vector corresponding to the same event; The first type of language data in the risk event vector is embedded using a knowledge graph to generate target embedding features that are aligned with the risk features of the second type of language data; Determine the attention weights of the multidimensional risk features in the embedded features of the risk event vector; The comprehensive risk score of the risk event is determined by combining the multidimensional risk characteristics and the corresponding attention weights. The comprehensive risk score is input into a pre-built sequence prediction model to obtain the comprehensive risk score trend of the data corresponding to the next T time windows; The probability of risk escalation is determined by combining the overall risk score trend; Early warnings will be issued based on the overall risk score trend and the probability of risk escalation.

2. The dynamic risk early warning method according to claim 1, characterized in that, The process of determining risk event vectors corresponding to the same event based on the preprocessed multimodal heterogeneous data includes: The different modal data in the multimodal heterogeneous data are encoded respectively, and the encoding form matches the data type of the corresponding modality; By applying a self-supervised modality contrastive learning loss function, the encoded data of different modalities corresponding to the same event are aligned in a shared semantic space to generate the risk event vector.

3. The dynamic risk early warning method according to claim 1, characterized in that, The process of using a knowledge graph to embed the first type of language data in the risk event vector into target embedding features aligned with the risk features of the second type of language data includes: The first type of language data in the risk event vector is identified, and the first type of language data is minority language data. The first type of language data is processed by calling a pre-built M-MLM model, and the embedding constraints are performed using the knowledge graph to obtain target embedding features that are aligned with the risk features of the second type of language data. The M-MLM model is obtained by secondary training on corpora in the energy and finance fields, and the domain of the knowledge graph is the same as the domain to which the current risk event belongs.

4. The dynamic risk early warning method according to claim 3, characterized in that, The method further includes: Multidimensional risk features are extracted from the risk event vector and embedded features; The multidimensional risk characteristics include one or more of the following: sentiment indicator characteristics, event nature characteristics, information consistency characteristics, entity correlation characteristics, and temporal suddenness characteristics.

5. The dynamic risk early warning method according to claim 4, characterized in that, The emotional indicators include the proportion of negative emotions, the magnitude of emotional fluctuations, and the intensity of extreme emotions. The characteristics of the event include the formality of the event's timing, the severity of the event, and the scope of the event's impact; The information consistency feature is a data source credibility indicator feature, including cross-platform consistency, verifiability of source information, and degree of unprocessed information. The entity correlation characteristics include the correlation between the risk entity and the designated enterprise, the proximity of the entity and the designated enterprise in terms of location and progress, and the importance of the role of the entity. The information consistency feature belongs to the time decay and suddenness index features, including the freshness of the event, the information growth rate of the event, and the suddenness coefficient of the event.

6. The dynamic risk early warning method according to claim 1, characterized in that, The determination of the attention weights for the risk event vector and the multidimensional risk features embedded in the features includes: The pre-built neural network model is invoked to process the risk event vector and embedded features, and the attention weight of each feature in the multidimensional risk features is dynamically determined. The neural network model includes a DARA module, and the attention weight is determined based on the DARA module.

7. The dynamic risk early warning method according to claim 1, characterized in that, The process of determining the comprehensive risk score of the risk event by combining the multidimensional risk features and corresponding attention weights includes: Determine the preset score corresponding to each risk characteristic; The comprehensive risk score of the risk event is calculated by combining the preset score, multidimensional risk features, and corresponding attention weights, using the following formula. : For the first i Attention weights for each feature For the first i Preset scores for each feature.

8. The dynamic risk early warning method according to claim 7, characterized in that, The method further includes: The comprehensive risk score is adjusted for timeliness using the following dynamic decay function to obtain the corrected result. The final comprehensive risk score : The attenuation coefficient is positively correlated with the propagation speed and importance of the event involved.

9. The dynamic risk early warning method according to claim 1, characterized in that, The corresponding early warning based on the comprehensive risk score trend and the probability of risk escalation includes: The target early warning level is determined based on the comprehensive risk score and the probability of risk escalation. Whether to trigger an alarm event is determined based on the target warning level; In response to the triggering of an alarm event, the alarm push content is determined based on the comprehensive risk score and risk escalation probability, and corresponding reports are generated, and the alarm event is archived.

10. A dynamic risk early warning system, characterized in that, include: The acquisition module is used to acquire multi-modal heterogeneous data and preprocess the multi-modal heterogeneous data; The first determining module is used to determine the risk event vector corresponding to the same event based on the preprocessed multimodal heterogeneous data; The generation module is used to embed the first type of language data in the risk event vector using a knowledge graph to generate target embedding features that are aligned with the risk features of the second type of language data. The second determining module is used to determine the attention weights of the multidimensional risk features in the embedded features of the risk event vector; The third determining module is used to determine the comprehensive risk score of the risk event by combining the multidimensional risk features and the corresponding attention weights. The input module is used to input the comprehensive risk score into a pre-built sequence prediction model to obtain the comprehensive risk score trend of the data corresponding to the next T time windows; The fourth determining module is used to determine the probability of risk escalation by combining the trend of the comprehensive risk score; The early warning module is used to issue corresponding early warnings based on the comprehensive risk score trend and the probability of risk escalation.

Images