A Bond Sentiment Monitoring Method and System Based on Risk Perception and Dynamic Prioritization
By establishing a multi-source data collection and semantic recognition model, calculating dynamic risk heat values, constructing a priority queue, deeply analyzing high-risk events, generating a repayment pressure index, and optimizing information source collection, the problems of scattered and inconsistent bond public opinion information have been solved, achieving efficient and intelligent risk monitoring.
Patent Information
- Application Number
- CN202611308237.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-27
- Publication Date
- 2026-11-17
AI Technical Summary
Bond-related public opinion information comes from scattered sources and comes in various formats, making it difficult for traditional methods to efficiently extract deep risk signals. How can we achieve timely response and dynamic prioritization of high-risk events with limited resources?
Establish a multi-source data collection system, standardize the format and clean the fields, identify risk types through a lightweight semantic recognition model, calculate dynamic risk heat values, construct a dynamic priority queue, perform in-depth analysis of high-priority events, generate a repayment pressure index, and optimize information source collection strategies.
It has improved the efficiency and accuracy of risk information processing, shifted from reactive post-event response to proactive pre-event prediction, reduced the waste of resources from invalid data collection, and enhanced the intelligence level of the bond public opinion monitoring system.
Smart Images

Figure CN123118263A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of bond public opinion monitoring technology, specifically to a bond public opinion monitoring method and system based on dynamic prioritization of risk perception. Background Technology
[0002] The sources of bond-related public opinion information have increased dramatically and become highly dispersed. The massive amounts of heterogeneous information generated from different channels, such as exchange announcements, regulatory disclosures, judicial information, financial media reports, and investor interaction platforms, vary significantly in terms of data format, release time, expression, and credibility. This poses an objective technical challenge to the comprehensive perception and systematic monitoring of bond risks. Bond risk information is often hidden in complex textual descriptions, and traditional information processing methods are insufficient to efficiently extract deep risk signals from large amounts of unstructured data. Semantic understanding and correlation analysis are needed to achieve accurate identification and structured transformation of risk information.
[0003] Chinese invention patent application CN115936835A discloses a method and device for intelligent monitoring of bond risk. The method includes: acquiring indicator data affecting corporate default; inputting the indicator data into a trained default analysis model for analysis and calculation to obtain real-time risk assessment results; wherein the default analysis model is obtained by comparing the training and evaluation results of multiple preset algorithm models to select an analysis model, and determining whether the analysis result of the selected analysis model reaches the preset risk assessment result, so as to train the selected analysis model according to the determination result; and determining the risk value of corporate financial risk based on the real-time risk assessment results.
[0004] Different risk events have varying degrees of impact and development trends on bond repayment capacity. How to achieve timely response and priority handling of high-risk events under limited computing resources is a practical engineering problem that needs to be solved in the design of a bond market sentiment monitoring system. Simultaneously, the risk status of bond issuers is constantly changing, and information collection strategies need to be dynamically adapted to the evolution of risks to balance monitoring coverage and resource consumption. Based on these technical requirements, it is necessary to establish a bond market sentiment monitoring method and system that integrates multi-source heterogeneous information processing, dynamic priority scheduling, and adaptive collection control, providing more efficient technical support for risk identification in the bond market. Summary of the Invention
[0005] The purpose of this invention is to address the problems existing in the background technology by proposing a bond public opinion monitoring method and system based on dynamic priority of risk perception.
[0006] The technical solution of this invention: a bond public opinion monitoring method based on dynamic priority of risk perception, comprising the following specific implementation steps: S1. Establish a multi-source data collection system to standardize the format, clean the fields, calibrate the time, and map the subject of public opinion data from exchange announcements, regulatory disclosures, judicial information, media reports, and investor interaction channels, and aggregate public opinion from different sources to the corresponding bond issuers to form risk event units; S2. Extract the degree of risk impact, information credibility, risk time change characteristics, and risk association and diffusion characteristics from the risk event unit, calculate the dynamic risk heat value, and construct a dynamic priority queue based on the dynamic risk heat value and its changing trend to prioritize the risk events. S3. Conduct in-depth analysis of high-priority risk events in the dynamic priority queue, extract risk factors such as debt repayment, judicial constraints, credit changes and operational liquidity, calculate the bond repayment pressure index, generate a three-level early warning result based on the repayment pressure index, and continuously update the early warning status in combination with new public opinion, risk disposal status and risk change trends. S4. Convert the early warning results into risk feedback intensity, evaluate the value of the information source by combining the risk contribution capability of the information source, the current risk activity status and feedback information, and dynamically adjust the information source collection cycle according to the evaluation results to complete the feedback optimization of the risk identification results to the public opinion collection process.
[0007] Preferably, step S1 specifically includes: The system parses webpage text, announcement documents, and structured interface data, converts public opinion data from different sources and in different formats into a unified data object, cleans up invalid characters, duplicate content, and non-business-related information in the text, retains information fields related to bond risk analysis, and performs time calibration on data from different sources to form a basic information object containing the release time, information source identifier, text content, and information of the releasing entity. Establish entity mapping relationships, calculate the degree of entity association based on enterprise name, bond identifier and entity relationship information, and when the degree of entity association meets the preset matching conditions, the corresponding public opinion information is attributed to the same bond entity; Based on a lightweight semantic recognition model, standardized public opinion texts are analyzed to identify types of risks, including abnormal bond repayment, debt default, major litigation, regulatory penalties, credit rating adjustments, and abnormal corporate operations. Risk labels containing risk categories and risk levels are generated by combining the intensity of risk triggers. Based on the relationships between the main entities, the continuity of time, and the evolution of risks, multiple public opinions are analyzed for correlation. Related public opinions formed by the same bond issuer within a preset time range are aggregated to construct a dynamic risk event unit that includes entity information, event evolution information, a set of related public opinions, and a set of risk tags. The event status is continuously updated based on new public opinions.
[0008] Preferably, in step S2, the specific steps for calculating the dynamic risk thermal value include: Using risk event units as the analysis objects, we extract characteristics of risk impact degree, information credibility degree, risk time change characteristics, and risk association and diffusion characteristics. Among them, the risk impact degree characteristics are determined based on the relationship between risk category and bond risk impact; the information credibility characteristics are determined based on the information source type, formal disclosure status, historical information accuracy, and multi-source verification status; the risk time change characteristics are determined based on the number of new risk information, risk label changes, and risk level change trends within a preset time range; and the risk association diffusion characteristics are determined based on the degree of impact of risk events on related entities and the spread of related risks. Based on the weights corresponding to the characteristics of each risk status, a dynamic risk heat value reflecting the current urgency of the risk event is generated through comprehensive calculation. The dynamic risk heat value corresponding to the risk event is continuously updated based on newly added public opinion information.
[0009] Preferably, in step S2, the specific steps for constructing a dynamic priority queue based on the dynamic risk heat value and its changing trend include: Based on the dynamic risk heat value corresponding to each risk event, the risk events are sorted and combined with the relationship between the dynamic risk heat value and the preset processing threshold, and the risk events are divided into high-risk real-time processing channel, key attention processing channel and ordinary processing channel. Calculate the trend of risk thermal changes, and determine whether the risk is increasing or decreasing based on the risk thermal changes between the current monitoring period and the previous monitoring period; The order of risk events in the dynamic priority queue is adjusted based on the dynamic risk heat value and the trend of risk heat change. When the risk heat value continues to increase, the processing priority is increased, and when the risk heat value continues to decrease, the processing priority is decreased, so as to realize the dynamic adjustment of the processing order of risk events.
[0010] Preferably, in step S3, the specific steps for performing in-depth analysis of high-priority risk events in the dynamic priority queue and extracting risk factors related to debt repayment, legal constraints, credit changes, and operational liquidity include: The high-priority risk event units output by the dynamic priority queue are used as the analysis objects. The event text, historical related events, risk tags and subject information are jointly analyzed to identify key entities such as issuers, bond names, related companies, creditor institutions, regulatory agencies and judicial institutions, and generate structured risk information based on the behavioral relationships between entities. Extract debt repayment pressure factors, which include bond principal repayment status, interest payment status, debt extension status, and overdue amount information; Extract judicial constraint factors, which include litigation amount, enforcement amount, asset freezing status and court enforcement status; Extract factors that cause credit changes, including the direction of rating adjustments, the circumstances of rating changes, the reasons for rating adjustments, and matters of concern to rating agencies. Extract operating liquidity factors, which include changes in main business, cash flow, changes in financing channels, and abnormal operating information.
[0011] Preferably, in step S3, the specific steps for calculating the bond repayment pressure index and generating a three-level early warning result based on the repayment pressure index include: Based on factors such as debt repayment pressure, legal constraints, credit changes, and operational liquidity, corresponding risk impact factors are generated, and the basic repayment pressure is calculated based on the degree of impact of each risk impact factor. Based on the simultaneous occurrence of risk factors, the timing of their occurrence, and the degree of correlation between risk factors, the synergistic impact of risks is determined, the basic repayment pressure is adjusted, and a bond repayment pressure index is generated. Based on the relationship between the bond repayment pressure index and the preset early warning threshold, a three-level early warning result is generated. When the risk pressure reaches the first-level early warning condition, a first-level early warning is generated and continuous tracking is implemented. When the risk pressure reaches the second-level early warning condition, a second-level early warning is generated and the monitoring frequency is increased. When the risk pressure reaches the third-level early warning condition, a third-level early warning is generated and a high-risk warning is output. The warning level will be continuously updated based on the changing trends of risk and pressure.
[0012] Preferably, the specific steps for continuously updating the alert status include: When new public opinion information is received, it is determined whether the new public opinion belongs to an existing risk event based on the consistency of the risk subject, the correlation of the risk type, and the correlation of the event time. When a new public opinion event belongs to an existing risk event, the new public opinion event will be added to the corresponding risk event unit, and the risk factors will be extracted again, the repayment pressure index will be calculated, and the warning level will be updated. The status of risk events is updated based on the progress of corporate debt resolution, changes in judicial status, and credit recovery. The level of risk concern is reduced when the risk resolution is completed.
[0013] Preferably, in step S4, the specific steps for evaluating the value of the information source by combining its risk contribution capability, current risk activity status, and feedback information include: The warning level is converted into risk feedback intensity, and the risk contribution capability of the information source is evaluated based on the correlation between the risk event and the information source. The contribution of information sources is determined based on the initial discovery of risk events, the provision of risk fields, and the early discovery of risks. The risk activity status of information sources is determined based on the number of risk events associated with information sources, changes in risk levels, and changes in risk pressure within the current period. Dynamic data collection needs are determined by combining the risk contribution capabilities and risk activity status of information sources, and information source collection strategies are adjusted according to these dynamic needs.
[0014] Preferably, in step S4, the specific steps for dynamically adjusting the information source acquisition cycle include: The original information collection cycle is adjusted according to dynamic collection needs. When the dynamic collection needs of information sources increase, the collection cycle is shortened, and when the dynamic collection needs of information sources decrease, the collection cycle is extended. Set the data collection period limit to ensure that the adjusted data collection period is within the preset data collection range; Obtain the results of manual review and feedback information on user browsing behavior. Determine the validity of the information provided by the information source based on the results of manual review, determine the degree of attention to risk information based on user browsing behavior, and update the risk contribution capability of the information source based on the feedback results, so that the updated risk contribution capability can be used in subsequent dynamic collection demand calculations.
[0015] The technical solution of this invention: A bond public opinion monitoring system based on risk perception and dynamic prioritization, comprising: Memory; processor; A computer program stored in the memory and capable of running on the processor; When the processor executes the computer program, it implements a bond sentiment monitoring method based on risk perception dynamic priority as described above.
[0016] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: This invention designs a bond public opinion monitoring method and system based on dynamic prioritization of risk perception. By constructing a unified collection and structured aggregation system for multi-source heterogeneous public opinion, it effectively solves the data fusion difficulties caused by the dispersion and inconsistent formats of bond information, significantly improving the organization efficiency and correlation analysis capabilities of risk information. Based on a dynamic thermodynamic computing mechanism for the fusion of multi-dimensional risk features, it realizes the dynamic tilt of computing resources towards high-risk events, greatly improving the timeliness of processing key risk information. Through deep semantic parsing, unstructured public opinion texts are transformed into quantifiable repayment pressure indices and three-level advanced warnings are generated, transforming risk monitoring from a reactive, post-event response to an active, pre-event prediction, enhancing the foresight and accuracy of risk identification. Crucially, the risk feedback-driven dynamic optimization mechanism of information sources enables the collection strategy to adaptively adjust with changes in risk status, forming a complete intelligent closed loop from risk perception to collection control, effectively reducing resource waste caused by invalid data collection, and comprehensively improving the intelligence level and operational efficiency of the bond public opinion monitoring system. Attached Figure Description
[0017] Figure 1 This is a flowchart of a bond public opinion monitoring method based on risk perception and dynamic priority proposed in this invention. Detailed Implementation
[0018] Example 1, as Figure 1 As shown, the present invention proposes a bond public opinion monitoring method based on dynamic priority of risk perception, which includes the following specific implementation steps: S1. By establishing a data collection system covering multiple channels such as exchange announcements, regulatory disclosures, judicial information, media reports, and investor interactions, the system performs unified format conversion, field cleaning, and time calibration on public opinion from different sources. It also performs subject mapping by combining bond name, code, issuer, and related enterprise information, and aggregates scattered public opinion to the corresponding bond issuer, thereby achieving unified organization and associated management of multi-source information. S2. Construct a dynamic priority scheduling mechanism for risk perception. By extracting multi-dimensional features such as the degree of risk impact, the credibility of information, the trend of time change, and the correlation of risk diffusion, a risk status description system is formed. Furthermore, dynamic risk heat values are calculated, and a dynamic priority queue is established based on the magnitude and trend of risk heat values to achieve priority processing of high-risk bond public opinion events and optimized allocation of computing resources. S3. Based on the high-priority risk event queue output in step S2, conduct in-depth risk analysis on key bond public opinion events. By extracting influencing factors such as debt repayment, judicial constraints, credit changes, and operating liquidity, construct a risk factor correlation analysis mechanism, calculate the bond repayment pressure index, and generate a three-level advanced early warning result based on the pressure index. At the same time, combine the subsequent new public opinion, corporate handling, and risk change trends to continuously update the status of the early warning event, realizing the transformation from risk information discovery to bond repayment risk prediction. S4. Based on the risk warning results and repayment pressure analysis results output in step S3, a risk feedback-driven dynamic optimization mechanism for information sources is constructed. By converting the warning level into the risk feedback intensity, and combining the contribution of information sources to the formation of risk events, the current risk activity status, and manual review feedback information, the value of different information sources is evaluated, and the information collection requirements and collection cycle are dynamically adjusted. This enables enhanced collection of high-risk and high-value information sources and reduced collection of low-value information sources, forming a closed-loop monitoring system where risk identification results drive the optimization of public opinion collection.
[0019] In a specific embodiment, step S1 establishes a multi-source information perception and risk event reconstruction mechanism around the bond issuer, converting discrete public opinion information from different data sources into structured risk event units with subject-related relationships, temporal evolution relationships, and risk attributes. The specific implementation process is as follows: S11. By establishing a data collection system covering multiple channels such as exchange announcements, regulatory disclosures, judicial information, media reports, and investor interactions, a unified format conversion, field cleaning, and time calibration are performed on public opinion from different sources. Furthermore, entity mapping is performed by combining bond names, codes, issuers, and related company information, aggregating scattered public opinion to the corresponding bond issuers. This achieves unified organization and associated management of multi-source information, specifically: Based on the different stages of information generation in the formation of bond risks, a multi-source bond public opinion perception system is established to simultaneously collect public information from different sources. The sources of information include at least: Information publicly disclosed by bond issuers includes issuance announcements, periodic reports, interim announcements, and explanations of major events; Public information disclosed by trading venues, including exchange announcements, abnormal bond trading alerts, and relevant regulatory disclosures; Regulatory and judicial information disclosure, including information on administrative penalties, regulatory inquiries, public disclosure of judicial proceedings, and enforcement information; Market communication information includes financial media reports, investor interaction information, and public discussion information; Because the aforementioned information sources differ significantly in data format, publication time, and expression, the collected raw information first undergoes unified preprocessing. The specific processing steps include: Parse web page text, announcement documents, and structured interface data, and convert data of different formats into a unified data object; Clean up invalid characters, duplicate content, and non-business-related information in the text content, and retain information fields related to bond risk assessment; Time calibration is performed on information from different sources to ensure that the order in which the same risk event appears in different channels remains consistent. After the above processing, each piece of public opinion information forms a basic information description object: ; In the formula, This represents the i-th original public opinion data object; This indicates the publication time information, used to record the time point when this public opinion event occurred; This indicates the source of the information and is used to identify the data channel to which the information belongs. This represents text content information for subsequent risk semantic analysis; This indicates the entity that published the information, used to identify the source of the information; It should be noted that the above parameters are all derived from the actual data collection process. The publication time is obtained from the data source publication time field, the information source is determined by the collection interface or webpage source address, the text content is obtained by page parsing, and the publishing subject is determined by the text entity recognition result and the source information. Furthermore, to avoid the problem of name differences for the same bond issuer in different data sources, this embodiment establishes a bond issuer association mapping mechanism; for example, the same issuer may have different expressions such as company abbreviation, group name, historical name and bond abbreviation, so the different names are uniformly associated with the same entity identifier through the entity mapping table; The subject matching degree is calculated as follows: ; In the formula, This indicates the degree of matching between the i-th piece of public opinion information and the corresponding bond issuer; This represents the name matching feature, used to characterize the degree of consistency between the text name and the company name; This indicates bond identifier matching features, used to characterize the matching of financial identifiers such as bond codes and bond abbreviations; This represents entity relationship matching features, used to characterize the degree of semantic association between enterprises, bonds, and related institutions in the text; , and These represent the corresponding feature weight coefficients; Among them, the name matching feature comes from the enterprise name standard library, the bond identifier matching feature comes from the bond basic information database, and the entity relationship matching feature comes from the text entity recognition result. When the subject matching degree exceeds a preset threshold, the public opinion information is attributed to the corresponding bond subject. S12. To address the issue of highly implicit bond risk information, semantic features are extracted from standardized public opinion texts. A lightweight semantic recognition model is used to analyze the text's contextual relationships, identifying risk types such as payment anomalies, debt defaults, litigation disputes, regulatory penalties, rating changes, and operational anomalies. Furthermore, risk label information, including risk category and risk level, is generated based on the risk trigger intensity. Specifically: After completing the standardization processing of multi-source information, since bond risk information is usually hidden in complex text descriptions, such as "tight liquidity", "partial debt extension", "received regulatory attention letter" and other expressions, it does not necessarily directly contain keywords such as "default" and "risk". Therefore, this embodiment addresses the problem that traditional keyword matching methods are prone to missing implicit risks by establishing a risk identification mechanism based on semantic understanding. Specifically, the standardized public opinion text is input into a lightweight semantic recognition model, and the risk event types contained therein are identified by analyzing the contextual relationship of the text. The types of risks mentioned include: abnormal bond repayment risk; debt default risk; major litigation risk; regulatory penalty risk; credit rating adjustment risk; and abnormal business operation risk. In the model processing, the text content is converted into semantic feature representation, and probability is calculated based on the risk category. The risk category prediction process is represented as follows: ; Obtain the probability values corresponding to different risk categories; In the formula, This represents the prediction result for different risk categories corresponding to the i-th public opinion event, i.e., the probability of the risk category; denoted as the feature vector formed after semantic encoding of the text, obtained by encoding the text context using a lightweight BERT model; W represents the risk category mapping parameter matrix, derived from the training process of historical bond risk samples; b represents the bias parameter, derived from the training process of historical bond risk samples. Furthermore, the results are corrected based on the risk trigger strength. The risk label strength is calculated as follows: ; In the formula, Indicates the intensity of the i-th public opinion risk label; The intensity of risk triggering is determined by a comprehensive assessment of the completeness of the risk behavior description, the clarity of the risk subject, and the certainty of the risk event. For example, "A company is expected to be unable to repay its bond principal on time" contains a clear entity, a clear action, and a clear impact, and therefore corresponds to a higher trigger strength; "The market is concerned about a company's financial situation" has a lower trigger strength because the risk entity and the degree of impact are unclear. S13. Based on the generated risk tag information, further analyze the subject association, temporal continuity, and risk type association among different public opinions. Aggregate multiple related public opinions generated by the same bond issuer within a certain time range to construct a dynamic risk event unit containing subject information, event process, related public opinions, and risk tags. This realizes the transformation from single public opinion analysis to risk event process analysis, specifically: Multiple pieces of public opinion information with the same subject relationship are analyzed for correlation, and multiple discrete pieces of information are combined into risk event units; specifically, the event correlation is established according to the following three dimensions: Entity Relationship: Determine whether different public opinions involve the same issuer, related companies, or the same bond; Temporal continuity: Determine whether multiple events occur consecutively within a reasonable time window; Risk evolution relationship: Determine whether there is a developmental link between different risk types; For example, "regulatory inquiries" and "rating downgrades" may both reflect changes in credit risk; "litigation freezes" and "payment delays" may both reflect increased repayment pressure. The correlation between events is calculated as follows: ; In the formula, This indicates the degree of correlation between the i-th and j-th public opinion items; The degree of subject relevance is derived from the subject mapping result in step S11; Indicates the degree of temporal correlation, calculated based on the time interval between the occurrence of two events; This indicates the degree of correlation between risk types, determined based on the historical evolutionary relationship between risk types; , and These represent the weights of the related factors; When the correlation reaches a set condition, multiple public opinion events are merged into a risk event unit, which is represented as follows: ; In the formula, This represents the i-th risk event unit; The risk entity information is derived from the entity mapping results; This indicates the temporal evolution of events, derived from the order of event publication times. This represents a collection of related public opinion, derived from the results of event correlation calculations. This represents the set of risk labels, derived from the results generated in step S12. S14. In response to the continuously changing nature of bond risks, existing risk event units are continuously updated based on new announcements, regulatory confirmations, and market feedback. The event status is adjusted by assessing the consistency of the subject, temporal relevance, and risk correlation between new information and existing events. Furthermore, the credibility of events is revised based on the reliability of information sources and subsequent confirmations, thereby improving the accuracy and timeliness of risk event data. Specifically: When the system detects new relevant public opinion information, it does not directly generate new isolated events, but first determines whether the information belongs to the development process of an existing risk event; The criteria for judgment include: whether the subjects are consistent; whether the risk types are related; and whether the time interval is within the event's impact period. If the association conditions are met, the new information will be added to the existing risk event unit and the risk status will be updated. Meanwhile, to improve the reliability of risk event units, an event credibility evaluation mechanism is introduced: ; In the formula, Indicates the credibility of the current risk event; This indicates the initial confidence level, derived from the model's recognition results; The strength of the subsequent confirmation information is indicated by information such as subsequent regulatory announcements and judicial results. This indicates the credibility of the information source, derived from the level of the issuing organization. , and These represent the corresponding influence weights; The status of the risk event unit is dynamically adjusted based on the updated credibility. For example, when the media reports on a company's abnormal financial situation, but no official information confirms it afterward, the credibility of the risk remains low; when the exchange announces confirmation of a bond's delayed repayment, the credibility of the risk increases rapidly. Ultimately, this will create a continuously updated database of risk events.
[0020] In a specific embodiment, step S2 constructs a risk heat map-driven dynamic information priority scheduling mechanism. Combining the characteristics of bond risk propagation, it integrates and analyzes the current impact of risk events, the credibility of information sources, and the risk diffusion trend to form a dynamically changing risk heat map value. This risk heat map value is then used to adjust information processing priorities in real time. The specific implementation process is as follows: S21. Based on the structured risk event units output in step S1, further analyze the risk subject, risk label, event time, and correlation, extract key features such as the degree of risk impact, information credibility, risk change rate, and risk diffusion relationship, and construct a multi-dimensional feature set to describe the bond risk status, transforming static risk events into risk status objects with dynamic attributes. Specifically: In step S1, risk event units containing risk subject information, event time information, risk label information, and event correlation information have been obtained. However, since different risk events affect the bond market in different ways, relying solely on the risk labels generated in step S1 cannot accurately determine the current priority of event handling. For example, two events that both belong to "litigation risk" may have different impacts. One may only involve general commercial disputes and have a small impact on bond repayment ability, while the other may involve the freezing of major assets or the execution of core operating entities, which has a direct impact on bond repayment ability. Further feature analysis of risk event units is needed to quantify and describe the key factors affecting the priority of event handling. Specifically, each risk event unit is taken as the analysis object, and risk status characteristics are extracted from four aspects: the degree of risk impact, the credibility of information, the speed of risk development, and the degree of risk propagation correlation. Risk Impact Severity Characteristic: This characteristic describes the extent to which a current risk event may affect bond solvency, creditworthiness, and market stability. Specifically, based on the risk label set generated in step S1, the impact of different risk types is analyzed. For example, when a risk event contains information such as "principal cannot be redeemed", "bond default", or "repurchase failure", it directly involves the rights and interests of creditors and therefore corresponds to a higher degree of risk impact. When a risk event only contains information such as "fluctuations in corporate operations" or "increased market attention", its impact on bond security is relatively limited and therefore corresponds to a lower degree of risk impact. Furthermore, risk impact characteristics are formed based on the correspondence between risk labels and the risk impact of bonds; Information credibility characteristics: Due to the diversity of sources of bond public opinion, the credibility of information from different sources varies; for example, exchange-disclosed announcements are official information and their authenticity and validity are relatively high; information disclosed by regulatory agencies can directly reflect the regulatory status of enterprises; although media reports and investor discussions can reflect changes in market attention in advance, they still need further verification. Therefore, based on the source identification information in step S1, the reliability of information from different sources is described, specifically including: the level of the information publishing entity; whether the information is officially disclosed; the accuracy of historical information; and whether there is multi-source cross-verification. The above information forms the credibility characteristics of the information source, enabling the subsequent risk ranking process to distinguish between "highly credible risk events" and "risk information to be confirmed"; Risk time-varying characteristics: Bond risk has obvious time evolution characteristics. Although some risk events have low initial impact, they may deteriorate rapidly over time; for example: abnormal corporate operations → regulatory attention → rating adjustment → abnormal bond prices → repayment risk; Therefore, further analysis of the development speed of risk events is needed. Specifically, based on the time information in the risk event unit, the number of new risk information, changes in risk labels, and trends in risk level changes for the same entity within a certain time range should be statistically analyzed. When multiple related risk events occur consecutively for the same entity in a short period of time, it indicates that risks are accumulating and the priority of subsequent processing should be increased. Risk correlation and diffusion characteristics: Considering that bond risks usually have the characteristics of correlation and diffusion, we further analyze the correlation between current risk events and historical events; for example, after a major lawsuit is filed against the issuer, it may further lead to financing restrictions; after a rating agency pays attention, it may further trigger abnormal market transactions. Therefore, the degree to which the current risk event may trigger subsequent risks is described by the event correlation in step S1; This ultimately forms a set of risk status characteristics, including: risk impact degree characteristics. (Based on the risk label and risk intensity output from step S1), information credibility characteristics (Information source attributes derived from step S1), characteristics of risk time changes (Based on the timing and continuous changes of risk events), characteristics of risk association and diffusion. (Derived from the event association relationship built in step S1); S22. To address the differences in the impact and development status of various risk events, a dynamic risk heatmap mechanism is established by integrating and analyzing the degree of risk impact, the reliability of information sources, the time evolution characteristics of risks, and the risk diffusion capacity. This mechanism generates heatmap values that reflect the current urgency of risks, allowing risk ranking to be dynamically adjusted based on actual changes in public opinion, rather than relying on fixed levels. To further establish a risk heatmap calculation mechanism, this paper aims to measure the importance of different risk events at the current time point. Existing public opinion monitoring methods typically employ fixed rules, such as assigning fixed levels to keywords like "default" and "litigation" and then ranking them according to these levels. However, in the actual bond market environment, the importance of risk depends not only on the type of risk but also on the reliability of information sources, the speed of event development, and the trend of risk diffusion. Therefore, a dynamic risk heatmap calculation method is proposed, enabling the priority of risk events to change according to actual circumstances. Specifically, risk events are comprehensively evaluated based on a set of risk status characteristics. ; In the formula, This represents the dynamic risk heat value of the i-th risk event; , , and These represent the weights of the corresponding factors. Each weight is configured according to the bond risk monitoring objectives and is used to adjust the degree of influence of different factors in the risk heat map calculation. For example, if a media outlet publishes news about abnormal business operations, its risk intensity value may be at a medium level due to the low credibility of the source and the lack of other risk associations; however, if the exchange subsequently issues a regulatory inquiry announcement accompanied by a rating change, its risk intensity value will automatically increase due to the increased credibility and risk diffusion. S23. Based on the dynamic risk heat value corresponding to each risk event, the public opinion events to be processed are sorted in real time, and multiple task channels are divided into high-risk real-time processing, key attention processing, and ordinary processing. At the same time, the status and heat value of risk events are continuously updated in combination with newly added public opinion information, so as to realize the dynamic adjustment of the order of risk event processing, and to prioritize the allocation of limited analysis resources to high-value risk events. Specifically: After obtaining the risk heat value corresponding to each risk event, a dynamic priority queue for computing resource scheduling is further established; specifically, all risk events are sorted according to their risk heat values, and different processing channels are divided according to the range of risk heat values. Among them: the first processing channel is a high-risk real-time processing channel, which is used to process events with high risk heat values; For example: bond default announcements; information on the enforcement of major lawsuits; information on rapid rating downgrades; regulatory penalty decisions; Events entering this channel will prioritize the use of in-depth analysis resources and the generation of risk warnings. The second processing channel is the key attention processing channel, which is used to handle events with risk heat values in the middle range; for example: abnormal business operations, significantly increased market attention, and concentrated media coverage; These types of events will be continuously monitored, and we will wait for further changes in the risk status. The third processing channel is the ordinary processing channel, which is used to process information with low risk thermal values. For example: general market discussions, repetitive news, and information without a clear risk indication; This type of information is processed according to the regular cycle; Furthermore, to prevent risk events from remaining in their original queue positions due to subsequent information changes, a dynamic adjustment mechanism is established. When new public opinion information enters the system: first, it is determined whether the information is related to an existing risk event; if so, the corresponding risk event unit is updated; the risk heat value is recalculated; and the queue position is adjusted according to the new risk heat value. For example: if a company initially receives negative media reports, it is placed in the normal processing queue; if a regulatory inquiry announcement is subsequently issued, the risk heat value increases, and it is automatically placed in the key attention queue; if a bond repayment delay announcement is further issued, it will immediately be placed in the high-risk real-time processing queue. S24. Further analyze the trend of risk event heat values over time, determine whether the risk is in an enhancing or mitigating state by calculating the change in risk heat values, and perform secondary optimization of processing priorities based on the current risk heat value. This enables the system to not only focus on the current risk magnitude but also identify rapidly deteriorating potential risk events, achieving dynamic resource scheduling oriented towards the risk evolution process. Specifically: Further incorporating the trend of risk thermal changes, the priority is optimized a second time, specifically calculating the amount of risk thermal change: ; In the formula, This represents the thermal change of the i-th risk event; This indicates the current time risk heat value; This indicates the risk heat value for the previous time period; like This indicates that the risk is increasing and a higher priority for handling it is needed; if If the risk impact is reduced, the processing priority can be appropriately lowered. Furthermore, the final processing order is adjusted based on the risk heat value and its changing trend: for high-heat events that are continuously increasing, the processing level is increased; for high-heat events that are trending towards stability, continuous monitoring is maintained; and for low-heat events that are continuously decreasing, the processing frequency is reduced.
[0021] In a specific embodiment, step S3 performs structured analysis on factors such as debt status, credit changes, operational anomalies, and judicial impact in high-priority events, converting text-level risk information into calculable repayment pressure indicators, thereby achieving a shift from public opinion monitoring to risk impact prediction. The specific implementation process is as follows: S31. Based on the high-priority risk events selected in step S2, conduct in-depth analysis of the event text, subject information, related public opinion, and risk tags to identify key factors involved, such as debt repayment, judicial impact, credit changes, and operational liquidity. Then, convert the text descriptions into a structured set of risk factors that reflect the impact of bond repayment, specifically: Deep semantic analysis is conducted on high-priority risk events to extract and structure information related to bond repayment ability from the event text. Specifically, the high-priority risk event unit output from step S2 is used as the input object, and the event description text, historical related events, risk tags and subject information contained therein are jointly analyzed. Semantic decomposition of risk event texts is performed to identify key entities involved, including: issuers, bond names, related companies, creditor institutions, regulatory agencies, and judicial institutions. Analyze the behavioral relationships between entities, focusing on identifying risky behaviors that may affect bond repayment ability; For example, in the sentence "A certain issuer's bank account was frozen by the court due to a contract dispute," ordinary text processing methods can only identify keywords such as "frozen" and "litigation." This example further analyzes: risk subject and bond issuer. Risky behavior: judicial freeze; Impact: Reduced disposable funds; Potential impact: Decreased short-term solvency; By using the above methods, the textual description is transformed into structured information with a risk impact direction. Furthermore, the following risk factors are identified regarding bond repayment pressure: Debt repayment pressure factors: This factor describes the current debt repayment status of the issuer, including: principal repayment status, interest payment status, debt extension status, and the amount of overdue debt; For example, when information such as "failure to pay interest on time" or "bond extension" appears in the text, it indicates that the issuer is under direct pressure to fulfill its obligations. Judicial constraints: This factor describes the impact of judicial actions on corporate assets and cash flow, including: litigation amount, enforcement amount, asset freeze status, and court enforcement status; For example, when the core assets of the issuing entity are frozen, even if the company continues to operate, it may lead to a decline in its short-term capital allocation ability. Credit change factors: This factor describes changes in the market's and professional institutions' assessment of the issuer's creditworthiness, including: the direction of rating adjustments, the magnitude of rating changes, the reasons for rating adjustments, and matters of concern to rating agencies; For example, a rating change from "stable" to "negative" indicates that credit risk may increase in the future; Operating liquidity factors: This factor describes changes in a company's solvency due to its own operations, including: changes in main business, cash flow situation, changes in financing channels, and major operational anomalies; After extracting the above factors, a set of risk impact factors corresponding to high-priority events is formed: {debt repayment pressure factor, judicial constraint impact factor, credit change impact factor, operating liquidity impact factor}. It should be noted that the debt repayment pressure factor is derived from information such as overdue debts and delayed payments in the text; the judicial constraint impact factor is derived from judicial announcements and enforcement information; the credit change impact factor is derived from information disclosed by rating agencies; and the operating liquidity impact factor is derived from operating announcements and corporate financial information. Through the above processing, the risk events output in step S2 are further transformed from information objects that need attention into a set of risk factors that can affect the bond's repayment ability. S32. To address the issue of bond risk accumulation caused by the combined effects of multiple risk factors, a fusion analysis is performed on the extracted risk factors. This analysis comprehensively considers the impact of individual risks and the synergistic amplification relationship between risk factors to establish a repayment pressure calculation mechanism. This generates a repayment pressure index that reflects the potential repayment pressure of the issuer, thus achieving a shift from risk event identification to risk impact assessment. Specifically: Analyzing the combined impact of different risk factors reveals that in the actual process of bond risk formation, a single risk factor does not necessarily lead directly to repayment risk, while the simultaneous occurrence of multiple risk factors may produce a significant risk superposition effect. For example, if a company has sufficient cash flow when it is involved in general litigation, it may not affect bond repayment. However, if at the same time, there are high litigation amounts, credit rating downgrades, reduced financing capabilities, and deteriorating cash flow, the combined effect of multiple factors will significantly increase repayment pressure. Therefore, to establish a risk factor fusion calculation mechanism, the basic repayment pressure is first calculated based on the importance of each risk factor: ; In the formula, This represents the underlying repayment pressure corresponding to the i-th risk event; This represents the k-th type of risk factor in the i-th event; This represents the weight of the k-th risk factor; m represents the number of risk factor categories. Further consider the combined effects of different risk factors; for example, a rating downgrade alone may only indicate a change in credit expectations, but when a rating downgrade occurs simultaneously with debt extension, it indicates that credit risk has transformed into repayment risk. Therefore, a risk synergistic correction factor is introduced: ; In the formula, This indicates the final bond repayment pressure index; Indicates the risk synergistic correction factor; It should be noted that the risk synergistic correction factor is determined based on the following relationships: the number of risk factors occurring simultaneously; the time interval between risk factors; and the degree of historical correlation between risk factors. For example, if the same entity experiences consecutive regulatory penalties, rating downgrades, and litigation freezes within a single time period, the risk synergy correction factor will increase. S33. Based on the calculated bond repayment pressure index, risk events are classified into different levels, with potential risks, obvious risks, and high-risk repayment risks corresponding to different early warning levels. A three-level advanced early warning result is generated based on changes in risk factors, as follows: After obtaining the bond repayment pressure index, it is normalized to further convert the continuously changing risk pressure into actionable risk warning levels. A three-level early warning mechanism is established based on the magnitude of the repayment pressure index and the risk growth, specifically including: Level 1 Warning: When the repayment pressure index is lower than the set Level 1 warning threshold, but potential risk factors have already appeared, a Level 1 warning is generated. This is usually manifested as: abnormal corporate operating indicators, increased market attention, and rating agencies issuing information of concern. At this time, the system mainly performs continuous tracking. Level 2 Warning: When the repayment pressure index is between the set Level 1 warning threshold and Level 2 warning threshold (including Level 1 warning threshold and Level 2 warning threshold), and multiple risk factors are correlated, a Level 2 warning is generated. This is usually manifested as: major litigation, significant credit rating decline, and decreased financing capacity. At this time, the system increases the monitoring frequency and focuses on the development of risks. Level 3 warning: When the repayment pressure index is higher than the Level 2 warning threshold and risk factors have directly affected the repayment ability, a Level 3 warning is generated. It is usually manifested as: bond repayment delay, actual default, or core assets being frozen. At this time, a high-risk warning is immediately issued. To avoid delays in risk response based solely on current pressure values, this embodiment further incorporates the risk pressure growth trend for correction, specifically calculating the pressure change trend: ; In the formula, This indicates the trend of pressure change for the i-th risk event; This indicates the current time period risk and pressure value; This indicates the risk and pressure value for the previous time period; when When the value is continuously positive and exceeds the preset growth threshold, it indicates that the risk is rapidly deteriorating, and the current warning level will be raised; when If the decline continues, maintain the original level or reduce the level of risk attention; S34. In response to the continuously changing nature of bond risks, dynamic tracking of generated early warning events is implemented. By linking newly added public opinion information, risk handling results, and changes in the status of the entities involved, risk factors, repayment pressure indices, and early warning levels are updated. This achieves full-process management of risk events from their generation and development to mitigation or escalation. Specifically: Establish a dynamic tracking mechanism for risk events and continuously maintain the early warning results that have been generated. Specifically, when new relevant public opinion information is received, first determine whether the information belongs to the development process of an existing risk event. The criteria for judgment include: whether the risk subjects are consistent, whether the risk types are related, and whether the event time is within the risk impact period; If the correlation conditions are met, the new information will be added to the original risk event, and the risk factor extraction and repayment pressure calculation will be re-executed. For example, if a company initially faces litigation risk, the system generates a Level 1 warning. If a subsequent announcement of a rating downgrade occurs, the risk impact factor is updated, the repayment pressure index increases, and the warning level is raised. If further delays in repayment occur, the warning level is directly adjusted to Level 3. At the same time, the results of risk management will be used as the basis for updating the event status; For example, if a company completes debt restructuring, resolves litigation, or restores its credit rating, its risk status will be reduced.
[0022] In a specific embodiment, step S4 establishes a risk feedback-driven dynamic evaluation mechanism for information sources based on the risk warning results output in step S3. This mechanism reverses the identified risk results and applies them to the front-end public opinion collection process. By analyzing the contribution of different information sources to the formation of risk events, it dynamically adjusts the information source collection priority and collection cycle, enabling risk monitoring resources to dynamically migrate as the risk status changes. The specific implementation process is as follows: S41. Using the risk warning results generated in step S3, convert different levels of warnings into risk feedback intensities. Combine this with the correlation between risk events and information sources to analyze the role of each information source in the risk discovery process. Calculate the risk contribution capacity of each information source based on its ability to discover risk clues in advance, its ability to provide key fields, and its contribution to risk confirmation. Specifically: This result is used to trace the information sources involved in the formation of risk events in reverse, and to determine the actual contribution of different information sources in the risk discovery process. Specifically, for each risk event that has generated an early warning result, the set of information sources corresponding to the risk event is obtained, and the information content, release time, completeness of risk fields, and impact on the formation of the final early warning level are recorded for each information source. For example: An issuer faces the risk of delayed bond repayment; in this case: the exchange's announcement first discloses the fact of delayed repayment, which directly leads to an increase in the risk level; media reports are subsequently reprinted, only expanding the scope of market attention; At this point, exchange announcements contribute more to risk identification than media reports. Therefore, the warning level output in step S3 is converted into a continuous risk feedback intensity to avoid the problem that discrete levels cannot participate in the calculation. Specifically, it is expressed as follows: ; In the formula, This represents the risk feedback intensity corresponding to the i-th risk event; This represents the warning level value corresponding to the i-th risk event; This represents the highest warning level value set; in this embodiment, the value is 3. Among them, the first-level warning corresponds to Level II warning corresponds to Level 3 warning correspondence ; Furthermore, based on the correlation between risk events and information sources, the risk contribution capability of information sources is calculated: ; In the formula, This represents the risk contribution capability of the j-th information source; This represents the degree of correlation between the j-th information source and the i-th risk event. This represents the number of risk events analyzed from the j-th information source; This indicates the total number of risk events included in the evaluation; It should be noted that the contribution correlation Determined based on a comprehensive consideration of the following factors: First: Judge based on whether the information source provides risk clues for the first time. If an information source discovers a risk change for the first time, and the risk is subsequently verified by other channels, then its relevance is increased. Second: Judge based on the importance of the risk fields provided by the information source; For example, information that directly affects the determination of repayment, such as bond redemption date, default amount, and regulatory penalty decision, has a higher relevance than general descriptive information; Third: Make a judgment based on the interval between the time the information source was released and the time the risk was confirmed; If a source of information exposes a risk trend in advance, its contribution will be greater; S42. Based on the risk contribution capability of information sources, and further combined with the current number of bond risk events, the trend of risk level changes, and the changes in repayment pressure, an information source risk activity status evaluation mechanism is constructed. This mechanism calculates the dynamic collection demand value corresponding to different information sources, determines the categories of information sources to be prioritized for monitoring, continuously monitored, and maintained infrequently, and realizes dynamic adjustment of collection resources according to the risk status. Specifically: After assessing the risk contribution capabilities of information sources, further consideration should be given to changes in the current risk status of the bond market. Since the importance of information sources is not fixed in the long term, for example, a regulatory website may only provide general information during the normal operation of a company, but when the company enters the stage of regulatory attention, the information value of that source will significantly increase. Therefore, by further incorporating the current risk activity status, the importance of information sources is dynamically adjusted. Specifically, based on the number of risk events associated with the information source within the current period, the rate of change in risk level, and the increase in risk pressure, a risk activity level is constructed, and the dynamic data collection demand value is further calculated. ; In the formula, This represents the dynamic data collection requirement value corresponding to the j-th information source; This indicates the current risk activity level of the j-th information source; Indicates the risk-enhancing adjustment parameter; It should be noted that the level of risk activity The following factors are considered when determining the risk level: the number of associated risk events (if the same information source is associated with multiple bond risk events recently, it indicates that the source currently has a high risk information density); changes in risk level (if the associated events of the information source rapidly escalate from Level 1 to Level 3 warning, it indicates that the risk is changing rapidly); and the trend of risk pressure growth (if the corresponding entity's repayment pressure index continues to increase, the importance of the source is increased). For example, if an issuer has recently been subject to a series of regulatory penalties, judicial enforcement actions, and rating adjustments, the risk activity level of relevant regulatory, judicial, and rating information sources will increase, leading to a greater demand for dynamic data collection. Furthermore, a pre-set first demand threshold is set. Second demand threshold ( The information sources are categorized based on the dynamic data collection requirements. when When this occurs, it indicates that the information source currently has high risk value and is set as a key monitoring source; when The j-th information source is divided into periodic monitoring information sources, and information is continuously acquired according to a preset regular collection cycle; when When the j-th information source is classified as a low-frequency maintenance information source, the data access frequency of this information source is reduced and it is only used to maintain basic risk coverage. S43. Based on the dynamic collection requirements of each information source, the original information acquisition cycle is adaptively adjusted. By shortening the collection interval for high-risk information sources and extending the collection interval for low-risk information sources, and setting constraints on the collection cycle, the timeliness of risk information acquisition is ensured while reducing the amount of invalid data collection. This achieves dynamic public opinion collection control oriented towards risk changes. Specifically: The data acquisition requirements are converted into actual acquisition control parameters, enabling the system to automatically adjust the data acquisition frequency. Specifically, the original acquisition cycle is corrected based on the dynamic acquisition requirements, using the following adjustment formula: ; In the formula, This indicates the adjusted collection period for the j-th information source; This indicates the collection period of the j-th information source before the adjustment; Specifically, when the information source has a high risk contribution capability and the current risk activity level increases: Increase the collection cycle: The decrease in frequency means the system will acquire new information from that source more frequently. For example, once a bond issuer enters the Level 2 early warning stage, the collection period for related regulatory announcements, exchange announcements, and judicial information sources is automatically shortened in order to obtain timely information on changes in risk. When the risk level decreases: The decline leads to a longer recovery period for the collection cycle; Meanwhile, to avoid abnormal changes in the data collection task due to dynamic adjustments, a periodic constraint is set: ; In the formula, Indicates the minimum allowed acquisition period; This indicates the maximum allowed data collection period; by setting boundaries, the system can ensure a fast risk response while avoiding excessive data collection that would waste computing resources. S44. Introduce manual review results and user browsing behavior as feedback signals to continuously adjust the risk contribution capability of information sources. Increase the evaluation weight of effective information sources based on actual risk identification results, reduce the impact of low-value information sources, and enable the information source evaluation model to continuously optimize with changes in the bond market environment and risk propagation patterns. Specifically: Further, the results of manual review and user behavior feedback are introduced to continuously optimize the information source evaluation model. Specifically, when manual review confirms that the information provided by a certain information source is true and valid and can support the risk level judgment, the subsequent evaluation weight of the information source is increased. Conversely, if an information source consistently generates repetitive, low-value, or misleading information, its evaluation weight will be reduced. At the same time, the level of market attention is judged by combining users' actual browsing behavior; For example, if a risk warning message is viewed by a large number of users, it indicates that the risk event has high market attention value; if a certain type of information is ignored for a long time, it indicates that its actual monitoring value is low. Therefore, a feedback correction factor is established: ; In the formula, This indicates the information source feedback correction factor; This indicates the value provided by the manual reviewer. This represents the user behavior feedback value; and These represent the corresponding feedback impact coefficients; Further update the information source risk contribution capability: ; In the formula, This indicates the revised risk contribution capability of the information source; The updated risk contribution capability is re-involved in the next round of dynamic data collection demand calculation, enabling the system to continuously adjust the information source evaluation results based on actual operational performance.
[0023] Example 2: This invention proposes a bond public opinion monitoring system based on dynamic priority of risk perception, which is used to execute the bond public opinion monitoring method based on dynamic priority of risk perception proposed in Example 1, and includes: Memory; processor; A computer program stored in the memory and capable of running on the processor; The processor executes a computer program to implement a bond sentiment monitoring method based on risk perception dynamic priority, as described in Embodiment 1 above.
[0024] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.
Claims
1. A bond sentiment monitoring method based on dynamic prioritization of risk perception, characterized in that, The specific implementation steps include the following: S1. Establish a multi-source data collection system to standardize the format, clean the fields, calibrate the time, and map the subject of public opinion data from exchange announcements, regulatory disclosures, judicial information, media reports, and investor interaction channels, and aggregate public opinion from different sources to the corresponding bond issuers to form risk event units; S2. Extract the degree of risk impact, information credibility, risk time change characteristics, and risk association and diffusion characteristics from the risk event unit, calculate the dynamic risk heat value, and construct a dynamic priority queue based on the dynamic risk heat value and its changing trend to prioritize the risk events. S3. Conduct in-depth analysis of high-priority risk events in the dynamic priority queue, extract risk factors such as debt repayment, judicial constraints, credit changes and operational liquidity, calculate the bond repayment pressure index, generate a three-level early warning result based on the repayment pressure index, and continuously update the early warning status in combination with new public opinion, risk disposal status and risk change trends. S4. Convert the early warning results into risk feedback intensity, evaluate the value of the information source by combining the risk contribution capability of the information source, the current risk activity status and feedback information, and dynamically adjust the information source collection cycle according to the evaluation results to complete the feedback optimization of the risk identification results to the public opinion collection process.
2. The bond public opinion monitoring method based on dynamic priority of risk perception according to claim 1, characterized in that, Step S1 specifically includes: The system parses webpage text, announcement documents, and structured interface data, converts public opinion data from different sources and in different formats into a unified data object, cleans up invalid characters, duplicate content, and non-business-related information in the text, retains information fields related to bond risk analysis, and performs time calibration on data from different sources to form a basic information object containing the release time, information source identifier, text content, and information of the releasing entity. Establish entity mapping relationships, calculate the degree of entity association based on enterprise name, bond identifier and entity relationship information, and when the degree of entity association meets the preset matching conditions, the corresponding public opinion information is attributed to the same bond entity; Based on a lightweight semantic recognition model, standardized public opinion texts are analyzed to identify types of risks, including abnormal bond repayment, debt default, major litigation, regulatory penalties, credit rating adjustments, and abnormal corporate operations. Risk labels containing risk categories and risk levels are generated by combining the intensity of risk triggers. Based on the relationships between the main entities, the continuity of time, and the evolution of risks, multiple public opinions are analyzed for correlation. Related public opinions formed by the same bond issuer within a preset time range are aggregated to construct a dynamic risk event unit that includes entity information, event evolution information, a set of related public opinions, and a set of risk tags. The event status is continuously updated based on new public opinions.
3. The bond public opinion monitoring method based on dynamic priority of risk perception according to claim 2, characterized in that, In step S2, the specific steps for calculating the dynamic risk thermal value include: Using risk event units as the analysis objects, we extract characteristics of risk impact degree, information credibility degree, risk time change characteristics, and risk association and diffusion characteristics. Among them, the risk impact degree characteristics are determined based on the relationship between risk category and bond risk impact; the information credibility characteristics are determined based on the information source type, formal disclosure status, historical information accuracy, and multi-source verification status; the risk time change characteristics are determined based on the number of new risk information, risk label changes, and risk level change trends within a preset time range; and the risk association diffusion characteristics are determined based on the degree of impact of risk events on related entities and the spread of related risks. The dynamic risk heat value is generated by comprehensively calculating the weights corresponding to the characteristics of each risk status, reflecting the current urgency of the risk event. The dynamic risk heat value corresponding to the risk event is continuously updated based on newly added public opinion information.
4. The bond public opinion monitoring method based on dynamic priority of risk perception according to claim 3, characterized in that, In step S2, the specific steps for constructing a dynamic priority queue based on dynamic risk heat values and their changing trends include: Based on the dynamic risk heat value corresponding to each risk event, the risk events are sorted and combined with the relationship between the dynamic risk heat value and the preset processing threshold, and the risk events are divided into high-risk real-time processing channel, key attention processing channel and ordinary processing channel. Calculate the trend of risk thermal changes, and determine whether the risk is increasing or decreasing based on the risk thermal changes between the current monitoring period and the previous monitoring period; The order of risk events in the dynamic priority queue is adjusted based on the dynamic risk heat value and the trend of risk heat change. When the risk heat value continues to increase, the processing priority is increased, and when the risk heat value continues to decrease, the processing priority is decreased, so as to realize the dynamic adjustment of the processing order of risk events.
5. A bond public opinion monitoring method based on dynamic priority of risk perception according to claim 4, characterized in that, In step S3, the specific steps for conducting in-depth analysis of high-priority risk events in the dynamic priority queue and extracting risk factors related to debt repayment, legal constraints, credit changes, and operational liquidity include: The high-priority risk event units output by the dynamic priority queue are used as the analysis objects. The event text, historical related events, risk tags and subject information are jointly analyzed to identify key entities such as issuers, bond names, related companies, creditor institutions, regulatory agencies and judicial institutions, and generate structured risk information based on the behavioral relationships between entities. Extract debt repayment pressure factors, which include bond principal repayment status, interest payment status, debt extension status, and overdue amount information; Extract judicial constraint factors, which include litigation amount, enforcement amount, asset freezing status and court enforcement status; Extract factors that cause credit changes, including the direction of rating adjustments, the circumstances of rating changes, the reasons for rating adjustments, and matters of concern to rating agencies. Extract operating liquidity factors, which include changes in main business, cash flow, changes in financing channels, and abnormal operating information.
6. The bond public opinion monitoring method based on dynamic priority of risk perception according to claim 5, characterized in that, In step S3, the specific steps for calculating the bond repayment pressure index and generating a three-level early warning result based on the repayment pressure index include: Based on factors such as debt repayment pressure, legal constraints, credit changes, and operational liquidity, corresponding risk impact factors are generated, and the basic repayment pressure is calculated based on the degree of impact of each risk impact factor. Based on the simultaneous occurrence of risk factors, the timing of their occurrence, and the degree of correlation between risk factors, the synergistic impact of risks is determined, the basic repayment pressure is adjusted, and a bond repayment pressure index is generated. Based on the relationship between the bond repayment pressure index and the preset early warning threshold, a three-level early warning result is generated. When the risk pressure reaches the first-level early warning condition, a first-level early warning is generated and continuous tracking is implemented. When the risk pressure reaches the second-level early warning condition, a second-level early warning is generated and the monitoring frequency is increased. When the risk pressure reaches the third-level early warning condition, a third-level early warning is generated and a high-risk warning is output. The warning level will be continuously updated based on the changing trends of risk and pressure.
7. A bond public opinion monitoring method based on dynamic priority of risk perception according to claim 6, characterized in that, The specific steps for continuously updating the alert status include: When new public opinion information is received, it is determined whether the new public opinion belongs to an existing risk event based on the consistency of the risk subject, the correlation of the risk type, and the correlation of the event time. When a new public opinion event belongs to an existing risk event, the new public opinion event will be added to the corresponding risk event unit, and the risk factors will be extracted again, the repayment pressure index will be calculated, and the warning level will be updated. The status of risk events is updated based on the progress of corporate debt resolution, changes in judicial status, and credit recovery. The level of risk concern is reduced when the risk resolution is completed.
8. A bond public opinion monitoring method based on dynamic priority of risk perception according to claim 7, characterized in that, Step S4, which involves evaluating the value of an information source by combining its risk contribution capability, current risk activity status, and feedback information, includes the following specific steps: The warning level is converted into risk feedback intensity, and the risk contribution capability of the information source is evaluated based on the correlation between the risk event and the information source. The contribution of information sources is determined based on the initial discovery of risk events, the provision of risk fields, and the early discovery of risks. The risk activity status of information sources is determined based on the number of risk events associated with information sources, changes in risk levels, and changes in risk pressure within the current period. Dynamic data collection needs are determined by combining the risk contribution capabilities and risk activity status of information sources, and information source collection strategies are adjusted according to these dynamic needs.
9. A bond public opinion monitoring method based on dynamic priority of risk perception according to claim 8, characterized in that, In step S4, the specific steps for dynamically adjusting the information source collection cycle include: The original information collection cycle is adjusted according to dynamic collection needs. When the dynamic collection needs of information sources increase, the collection cycle is shortened, and when the dynamic collection needs of information sources decrease, the collection cycle is extended. Set the data collection period limit to ensure that the adjusted data collection period is within the preset data collection range; Obtain the results of manual review and feedback information on user browsing behavior. Determine the validity of the information provided by the information source based on the results of manual review, determine the degree of attention to risk information based on user browsing behavior, and update the risk contribution capability of the information source based on the feedback results, so that the updated risk contribution capability can be used in subsequent dynamic collection demand calculations.
10. A bond public opinion monitoring system based on risk perception and dynamic prioritization, characterized in that, include: Memory; processor; A computer program stored in the memory and capable of running on the processor; When the processor executes the computer program, it implements a bond public opinion monitoring method based on risk perception dynamic priority as described in any one of claims 1 to 9.
Citation Information
Patent Citations
Intelligent monitoring method and device for bond risk
CN115936835A
