Library copyright risk intelligent early warning method based on multi-source heterogeneous data and dynamic risk threshold

Abstract

The invention provides a library copyright risk intelligent early warning method based on multi-source heterogeneous data and a dynamic risk threshold, and relates to the technical field of intellectual property management. According to the method, judicial referee data, library business system data and copyright database data are acquired, environmental factor data including hot topics and policy changes are introduced, and a fusion feature data set and copyright risk index system is constructed; and training a machine learning time sequence model based on the historical index data and the environmental factor data to generate a dynamic risk threshold value, performing early warning judgment in combination with the real-time index data, and generating an interpretable early warning report. According to the method, self-adaption and accurate early warning of the library copyright risk are realized, and the risk of misinformation and missing report is reduced.

Description

Technical Field

[0001] This invention relates to the field of intellectual property management technology, and in particular to an intelligent early warning method for library copyright risks based on multi-source heterogeneous data and dynamic risk thresholds. Background Technology

[0002] Existing library copyright risk management mainly relies on business system ledgers, authorization contracts, and manual spot checks. Some systems introduce fixed rules or static thresholds to alert users to situations such as "use beyond the scope, expired authorization, and dissemination of suspected infringing content." Data sources are usually limited to the library's internal business systems or a single copyright database, and cross-system verification depends on manual comparison. Warning results are mostly simple outputs of "whether to issue an alert."

[0003] With the increasing procurement of digital resources, inter-library sharing, online dissemination, and public rights protection channels, copyright risk signals are becoming more multi-sourced and time-varying. The industry is gradually evolving towards "multi-source data access, intelligent judgment, closed-loop management, and audit traceability," and is beginning to focus on incorporating judicial rulings, policy changes, and public opinion hotspots into risk assessment to improve the coverage and response speed of early warnings.

[0004] Existing solutions generally suffer from problems such as data silos leading to incomplete evidence chains, fixed thresholds being difficult to adapt to risk fluctuations caused by hot topics and policy changes, and the lack of transparency in the triggering reasons of models or rules making it difficult for administrators to review and handle them. These problems result in false alarms and omissions, affecting the operability and traceability of copyright risk governance. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide an intelligent early warning method for library copyright risks based on multi-source heterogeneous data and dynamic risk thresholds. By fusing multi-source heterogeneous data and incorporating a dynamic risk threshold mechanism that incorporates environmental factors, this invention achieves comprehensive, accurate, and adaptive early warning of library copyright risks, effectively overcoming the problems of false alarms and missed alarms caused by data fragmentation and static thresholds in existing technologies.

[0006] To achieve the above objectives, the present invention provides the following solution: A library copyright risk intelligent early warning method based on multi-source heterogeneous data and dynamic risk thresholds includes: Acquire multi-source heterogeneous data of management objects used to characterize library copyright risks and environmental factor data used for risk assessment of the management objects; the multi-source heterogeneous data includes at least judicial judgment data, library business system data and copyright database data; the environmental factor data includes hot topic data and policy change data, and the environmental factor data carries a time identifier and is continuously updated according to a preset update cycle to form an environmental factor sequence; Based on a preset fusion data model, the multi-source heterogeneous data is mapped to fields, associated with entities, and unified in structure. A fusion feature dataset is output, and a source identifier is written to the records in the fusion feature dataset to characterize the data source of the multi-source heterogeneous data. A copyright risk indicator system is constructed based on the fused feature dataset, and historical copyright risk indicator data and real-time copyright risk indicator data are generated for each of the managed objects according to a preset granularity. The historical copyright risk indicator data and the historical environmental factor data extracted from the environmental factor sequence according to a preset historical time window are input into a machine learning time series model for training, and the threshold adjustment result is output. The threshold adjustment result is combined with the real-time environmental factor data obtained from the environmental factor sequence according to the preset update cycle to output a dynamic risk threshold. The real-time copyright risk indicator data is matched with the dynamic risk threshold, and an early warning judgment result is output. An early warning event is generated when the early warning judgment result meets the preset early warning triggering conditions. An interpretable early warning report is generated for the aforementioned early warning event, and the early warning report is sent to the management terminal to assist in decision-making. The early warning report includes a list of abnormal indicators, the source identifiers corresponding to the abnormal indicators, the triggered early warning rules, and an explanation of the association between the dynamic risk thresholds.

[0007] Preferably, the judicial ruling data includes judgment documents from the China Trial Network, the library business system data includes resource usage records and authorization management information related to the managed object, and the copyright database data includes ownership information and authorization scope information related to the managed object.

[0008] Preferably, the hot topic data includes hot entries and popularity information of copyright-related topics; the policy change data includes the release information and effective time information of copyright-related policy documents, and the time identifier is used to represent the occurrence time or effective time of the hot topic data and the policy change data.

[0009] Preferably, the field mapping includes unifying similar fields from different sources into fused fields according to a preset field mapping relationship; the entity association includes associating multiple records pointing to the same managed object according to a preset entity association rule; and the structure unification includes writing the associated records into the fused feature dataset according to a unified data structure.

[0010] Preferably, the source identifier includes at least a judicial judgment source identifier, a library business system source identifier, and a copyright database source identifier, and is used to locate the data source corresponding to the abnormal indicator in the warning report.

[0011] Preferably, the copyright risk indicator system includes authorization status indicators, usage behavior indicators, dispute record indicators, and dissemination attention indicators, and both the historical copyright risk indicator data and the real-time copyright risk indicator data include the indicator values ​​of the authorization status indicators, usage behavior indicators, dispute record indicators, and dissemination attention indicators.

[0012] Preferably, the machine learning time series model is a Long Short-Term Memory (LSTM) network model, and the threshold adjustment result represents the threshold adjustment amount or direction of each indicator in the copyright risk indicator system, and is used to generate the dynamic risk threshold corresponding to each indicator.

[0013] Preferably, the early warning determination result includes an indicator matching result and a rule matching result; wherein the indicator matching result is used to characterize the comparison result between the indicator value in the real-time copyright risk indicator data and the corresponding dynamic risk threshold, and the rule matching result is used to characterize whether the early warning rule corresponding to the preset early warning triggering condition is met.

[0014] Preferably, the early warning report further includes: the abnormal indicator value of each abnormal indicator in the abnormal indicator list, the dynamic risk threshold corresponding to the abnormal indicator value, and the comparison result of the abnormal indicator value relative to the dynamic risk threshold; the triggered early warning rule; the source identifier corresponding to the abnormal indicator; and the time identifier of the real-time environmental factor data used to generate the dynamic risk threshold and the threshold adjustment result.

[0015] The present invention discloses the following technical effects: This invention overcomes the problem in existing technologies where copyright risk information is scattered across different systems and relies on manual comparison by acquiring and integrating judicial judgment data, library business system data, and copyright database data under the same technical framework. This allows the assessment of copyright risks in libraries to no longer be limited to a single data source, thereby forming a complete information foundation covering ownership, usage behavior, and dispute situations, significantly improving the comprehensiveness and objectivity of copyright risk identification.

[0016] This invention introduces environmental factor data and constructs it into a time-stamped sequence of environmental factors, enabling copyright risk assessment to perceive changes in the external environment such as hot topics and policy changes. This overcomes the problem in existing technologies where risk judgment relies on fixed rules or static thresholds for a long time and is difficult to adapt to the fluctuation of risk intensity over time. It improves the adaptability of early warning results to changes in the real copyright environment from a mechanism perspective.

[0017] This invention trains a machine learning time series model using historical copyright risk indicator data and historical environmental factor data, and generates a dynamic risk threshold by combining it with real-time environmental factor data. This allows the risk threshold to be automatically adjusted according to historical evolution trends and current environmental conditions, avoiding false alarms and missed alarms caused by traditional threshold settings that rely on experience or manual configuration, thereby improving the accuracy and stability of copyright risk warning.

[0018] This invention achieves refined identification of the risk status of different management objects by matching real-time copyright risk indicator data with dynamic risk thresholds and generating early warning events during the early warning determination stage. This makes early warning triggering no longer a simple binary judgment, but a comprehensive judgment based on multi-dimensional indicators and dynamic thresholds, effectively improving the rationality and controllability of early warning results.

[0019] This invention provides clear triggering basis and data source for early warning results by explicitly providing a list of abnormal indicators, corresponding data sources, triggering early warning rules, and environmental factor information related to dynamic risk thresholds when generating early warning reports. This solves the problems of lack of interpretation path and difficulty in verification and handling of early warning results in the prior art, thereby providing library managers with clear and reliable decision support basis. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart of the method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The purpose of this invention is to provide an intelligent early warning method for library copyright risks based on multi-source heterogeneous data and dynamic risk thresholds. Through an interpretable early warning reporting mechanism, the indicators, rules and data sources on which the early warning is triggered are clearly defined, thereby improving the transparency and operability of copyright risk early warning results and providing reliable decision support for library copyright risk management.

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides an intelligent early warning method for library copyright risks based on multi-source heterogeneous data and dynamic risk thresholds, including: Step 100: Obtain multi-source heterogeneous data of the management objects used to characterize the library's copyright risks and environmental factor data used to assess the risks of the management objects; the multi-source heterogeneous data includes at least judicial judgment data, library business system data and copyright database data; the environmental factor data includes hot topic data and policy change data, and the environmental factor data carries a time stamp and is continuously updated according to a preset update cycle to form an environmental factor sequence; Step 200: Based on the preset fusion data model, perform field mapping, entity association and structure unification on multi-source heterogeneous data, output fusion feature dataset, and write source identifiers for records in the fusion feature dataset to characterize the data source of multi-source heterogeneous data; Step 300: Construct a copyright risk indicator system based on the fused feature dataset, and generate historical copyright risk indicator data and real-time copyright risk indicator data for each managed object according to the preset granularity; Step 400: Input the historical copyright risk indicator data and the historical environmental factor data extracted from the environmental factor sequence according to the preset historical time window into the machine learning time series model for training, output the threshold adjustment result, and combine the threshold adjustment result with the real-time environmental factor data obtained from the environmental factor sequence according to the preset update cycle to output the dynamic risk threshold. Step 500: Match the real-time copyright risk indicator data with the dynamic risk threshold, output the warning judgment result, and generate a warning event when the warning judgment result meets the preset warning triggering conditions; Step 600: Generate an interpretable early warning report for the early warning event and send the early warning report to the management terminal to assist in handling decisions; the early warning report includes a list of abnormal indicators, the source identifiers corresponding to the abnormal indicators, the triggered early warning rules, and a description of the association between dynamic risk thresholds.

[0026] Specifically, in step 100 of this embodiment, in order to obtain multi-source heterogeneous data of management objects used to characterize library copyright risks, this embodiment pre-determines the object scope and object granularity of the management objects. The management objects refer to object units included in the copyright risk assessment, and each object unit includes at least one or a combination of digital resource entries, library collection resource entries, interlibrary shared entries, and reader access sessions. In this embodiment, the object granularity is set at the resource entry level, and the most recent 180 days are used as the data collection time range. Based on the object scope and object granularity, this embodiment obtains judicial judgment data, library business system data, and copyright database data for each management object. At least three types of data sources are collected for the same management object. The judicial judgment data is used to characterize the judgment facts and conclusions related to copyright disputes, and includes judgment document information from the China Trial Network. The library business system data is used to characterize resource usage record information and authorization management information related to the management object. The copyright database data is used to characterize the rights ownership information and authorization scope information related to the management object. This embodiment performs time-based and object-based aggregation on the multi-source heterogeneous data. That is, the data entries corresponding to the same management object under different data sources are aggregated into management object data records according to preset aggregation rules, so as to ensure that a consistent data foundation can be obtained from the same management object dimension when constructing risk indicators in the future.

[0027] In this embodiment, to obtain environmental factor data for risk assessment of managed objects, environmental factor data is defined as a set of time-varying information that has an external impact on the intensity of copyright risk. This environmental factor data is divided into two categories: hot topic data and policy change data. In this embodiment, the update cycle of the environmental factor data is set to 1 day, and the maximum number of hot topics within each update cycle is set to 50. Hot topic data refers to information on changes in attention to copyright-related topics within a certain time frame, including at least hot topics and popularity indicators. Policy change data refers to the release and effective dates of copyright-related policy documents, covering policy documents released or effective within the last 365 days. This embodiment configures time stamps for the environmental factor data. These time stamps represent the occurrence time of the hot topic data and the release or effective date of the policy change data. The time stamps are used as the sorting basis for the environmental factor sequence, enabling the environmental factor data to form a continuously updatable serialized representation in the time dimension. This provides a unified standard for subsequently extracting historical environmental factor data by historical time windows and obtaining real-time environmental factor data by update cycle.

[0028] In this embodiment, to enable the multi-source heterogeneous data and the environmental factor data to jointly serve the risk assessment of the managed object, a time alignment method is used to establish a correspondence between the managed object data records and the environmental factor sequence. In this embodiment, daily alignment is used, and the time tolerance between adjacent statistical periods is set to 1 day. Specifically, this embodiment uses the occurrence time or statistical time of the managed object data records as the alignment benchmark, and selects hot topic data and policy change data from the environmental factor sequence that are in the same or adjacent statistical periods as the alignment benchmark as the environmental factor data subset associated with the risk assessment of the managed object. For each managed object, the environmental factor data subset associated with each statistical period contains at least one hot topic item or one policy change record. The environmental factor data subset is used to characterize the external environmental background when the managed object experiences usage behavior, changes in authorization status, or the emergence of disputed information, so that the subsequent dynamic adjustment of the risk threshold can simultaneously reflect the managed object's own status and changes in the external environment.

[0029] Specifically, in step 200 of this embodiment, after acquiring the multi-source heterogeneous data described in step 100, to eliminate the differences in field naming, data structure, and semantic expression among different data sources, this embodiment performs field mapping processing on the multi-source heterogeneous data based on a preset fusion data model. Field mapping refers to mapping fields from different data sources that have the same semantics or function to fused fields according to a pre-established field mapping relationship table. For example, the judgment conclusion field in judicial judgment data, the usage status field in library business system data, and the authorization status field in copyright database data are uniformly mapped to fused fields used to characterize the copyright status of the managed object. In this embodiment, each type of fused field corresponds to at least two original fields from different sources, thereby ensuring that the fused fields can fully reflect the copyright risk information of the managed object under different data sources.

[0030] In this embodiment, after completing the field mapping, entity association processing is further performed on the multi-source heterogeneous data to achieve data aggregation around the same management object. Entity association refers to the association and integration of multiple data records pointing to the same management object according to preset entity association rules. These entity association rules include at least matching rules based on resource unique identifiers, authorization credential information, or usage record pointing relationships. In this embodiment, when performing entity association on the same management object, multiple records from different data sources are allowed, and these records are considered as risk characterization information of the same management object in different dimensions. After completing the entity association, this embodiment performs structural unification processing on the associated records, that is, organizing the field content according to a unified data structure and writing the structurally unified records into the fused feature dataset. In this embodiment, the fused feature dataset is organized with management objects as the basic unit, and each management object corresponds to at least one fused record.

[0031] In this embodiment, to enhance the interpretability of subsequent risk warning results, a source identifier is written to each record in the fused feature dataset while generating the dataset. The source identifier characterizes the data source composition of the fused record and includes at least judicial judgment source identifiers, library business system source identifiers, and copyright database source identifiers. In this embodiment, the source identifiers correspond one-to-one with the fused fields in a distinguishable format, enabling accurate location of the data source corresponding to the abnormal indicator when generating the warning report. By writing the source identifier, this embodiment can trace and locate the source of abnormal indicators without changing the structure of the fused feature dataset, thereby providing clear data basis for the management to analyze and handle copyright risks.

[0032] Optionally, in step 300 of this embodiment, after generating the fused feature dataset described in step 200, this embodiment constructs a copyright risk indicator system based on the fused feature dataset to form a unified representation of the copyright risk status of the managed object. The copyright risk indicator system refers to a set of indicators set from different dimensions for the copyright risk of the managed object, used to transform information related to copyright risk in the fused feature dataset into indicator values ​​that can be used for subsequent threshold adjustment and early warning judgment. In this embodiment, the copyright risk indicator system includes at least authorization status indicators, usage behavior indicators, dispute record indicators, and dissemination attention indicators. The authorization status indicators characterize the validity and consistency of the authorization scope of the managed object; the usage behavior indicators characterize the degree to which the actual usage of the managed object conforms to the authorization constraints; the dispute record indicators characterize the occurrence and intensity of disputes related to the managed object; and the dissemination attention indicators characterize the degree of risk exposure of the managed object at the dissemination and attention level. To ensure the coverage of the indicator system, this embodiment generates indicator values ​​for the above four types of indicators for each managed object.

[0033] In this embodiment, to achieve the generation of copyright risk indicator data at a preset granularity as described in step 300, the preset granularity is defined as a combination of time granularity and object granularity for risk statistics and output of managed objects. The object granularity is consistent with step 100, using the resource item level. In this embodiment, the time granularity is set to daily, ensuring that each managed object generates a set of copyright risk indicator values ​​for each day. Based on this preset granularity, this embodiment extracts the fusion fields and their values ​​corresponding to each indicator from the fusion feature dataset, and forms a set of indicator values ​​for the authorization status indicator, the usage behavior indicator, the dispute record indicator, and the dissemination attention indicator according to the indicator definition rules, thereby obtaining copyright risk indicator data corresponding to each managed object. In this embodiment, each managed object generates at least four indicator values ​​in each statistical period, and these values ​​are written into the copyright risk indicator data using a unified field caliber.

[0034] In this embodiment, to meet the different data requirements for subsequent machine learning time-series model training and online early warning judgment, the copyright risk indicator data is further divided into historical copyright risk indicator data and real-time copyright risk indicator data. Both include the indicator values ​​of the authorization status indicator, the usage behavior indicator, the dispute record indicator, and the dissemination attention indicator. The historical copyright risk indicator data refers to the sequence of indicator values ​​accumulated within a preset historical time window according to the preset granularity, used to characterize the risk evolution characteristics of the managed object over a longer period. The real-time copyright risk indicator data refers to the set of indicator values ​​generated within the current statistical period, used to characterize the risk status of the managed object in the current period. In this embodiment, the historical time window is set to the most recent 180 days, and a historical indicator sequence is formed with each day as a statistical period. The real-time copyright risk indicator data corresponds to the indicator values ​​of the most recent statistical period, thereby ensuring that the historical data and real-time data are consistent in terms of indicator scope and meet the needs of training and early warning within the time range.

[0035] Further, in step 400 of this embodiment, after obtaining the historical copyright risk indicator data described in step 300, this embodiment extracts historical environmental factor data from the environmental factor sequence formed in step 100 according to a preset historical time window, and uses the historical copyright risk indicator data and the historical environmental factor data together as training samples to input into the machine learning time series model to learn the temporal correlation between indicator changes and environmental changes. The preset historical time window refers to the time range used to cover the risk evolution law of the managed object, to ensure that the training samples can include different risk stages and different external environmental states; in this embodiment, the preset historical time window is the most recent 180 days, and the historical environmental factor data and the historical copyright risk indicator data adopt the same daily time alignment caliber, so that the indicator value of each day can be associated with hot topic data and policy change data within the same statistical period, avoiding mismatch in the time dimension of the training samples.

[0036] In this embodiment, the machine learning time series model uses a long short-term memory network model to characterize the long-term dependence and short-term fluctuation characteristics of indicator values ​​over time, and to generate threshold adjustment results during training. The threshold adjustment results refer to the threshold adjustment information for each indicator in the copyright risk indicator system, reflecting the trend of indicator threshold changes relative to the baseline threshold. In this embodiment, the threshold adjustment results at least characterize the threshold adjustment direction for each indicator, including both upward and downward adjustments. Furthermore, they can characterize the adjustment magnitude level to distinguish the strength of the impact of environmental changes on the threshold. The adjustment magnitude level is divided into three levels, from weak to strong, to characterize slight adjustment, general adjustment, and significant adjustment, so as to achieve dynamic adaptation of the threshold to environmental changes without introducing complex calculation processes.

[0037] In this embodiment, after obtaining the threshold adjustment result, real-time environmental factor data is obtained from the environmental factor sequence according to a preset update cycle. This real-time environmental factor data is then combined with the threshold adjustment result to generate dynamic risk thresholds corresponding to each indicator. The dynamic risk threshold refers to the set of indicator thresholds used for early warning determination within the current statistical period. These thresholds can be adjusted according to the timeliness of real-time hot topics and policy changes. In this embodiment, the preset update cycle is 1 day, ensuring that the dynamic risk thresholds are updated once per statistical period. Corresponding dynamic risk thresholds are generated for the authorization status indicator, usage behavior indicator, dispute record indicator, and dissemination attention indicator in the copyright risk indicator system. This ensures that the matching determination in subsequent step 500 can compare the thresholds of different indicators under a unified standard and form an early warning determination result.

[0038] As an example, in this embodiment, to improve the threshold adaptation capability under the combined effects of "indicator evolution over time" and "abrupt changes in environmental factors," this embodiment employs a threshold learning structure coupled with dual-channel temporal encoding and environmental gating in the machine learning time series model. This structure includes an indicator temporal encoding channel and an environmental temporal encoding channel: the indicator temporal encoding channel is used to encode the time dependency of historical copyright risk indicator data, and the environmental temporal encoding channel is used to encode the time changes of historical environmental factor data; the encoding results of the two channels are coupled through an environmental gating unit, which is used to adaptively modulate the temporal representation of the indicator channel in situations such as an increase in the intensity of hot topics or the effectiveness of policy changes, enabling the model to distinguish between "indicator self-fluctuations" and "environment-driven fluctuations." In this embodiment, the input time length is the most recent 180 days, aligned to the day level; the environmental gating unit sets influence weights for hot topic data and policy change data respectively and completes coupling within the same statistical period, thereby forming a joint temporal representation that is more sensitive to threshold adjustments.

[0039] In this embodiment, to enhance the ability to capture sudden hotspots and phased policy adjustments, a combined structure of multi-scale temporal feature extraction and key moment attention focusing is adopted within each channel. The multi-scale temporal feature extraction is used to simultaneously extract short-term fluctuation features and medium-to-long-term trend features. The key moment attention focusing is used to automatically select several statistical periods from historical time windows that contribute more significantly to threshold changes, and use these as the key basis for threshold learning. This allows the model to maintain stable threshold adjustment logic in both scenarios of "short-term surges in popularity" and "long-term risk accumulation." To maintain model interpretability, this embodiment sets "indicator contribution output" as an accompanying result of training and inference, so that the relative contribution of each indicator to threshold adjustment in the current statistical period can be recorded for subsequent warning reports. In this embodiment, key moment attention focusing selects a maximum of 14 statistical periods as key segments, and multi-scale extraction covers 3 time scales, thereby achieving compatibility with risk signals at different time scales without introducing complex calculations.

[0040] In this embodiment, the training process is based on a joint sample of historical copyright risk indicator data and historical environmental factor data. It employs a training flow of "benchmark threshold alignment—threshold adjustment learning—robust constraint calibration" to output threshold adjustment results and further generate dynamic risk thresholds corresponding one-to-one with each indicator. Specifically, this embodiment first constructs a supervision signal in the training data consistent with the historical early warning results of the managed object, enabling the model to learn the adjustment direction and magnitude of each indicator threshold under different environmental backgrounds. Then, robustness constraints are introduced to prevent frequent threshold jumps when environmental factors fluctuate slightly, and to allow interpretable step adjustments to the thresholds when the number of hot items rises rapidly or policies take effect. Finally, the threshold adjustment results after training are calibrated for consistency, ensuring that the dynamic risk thresholds of the authorization status indicator, usage behavior indicator, dispute record indicator, and dissemination attention indicator have consistent time signatures and update rhythms within the same statistical period. In this embodiment, the training samples are divided into a 7:2:1 training segment, validation segment, and test segment in chronological order, and the update cycle of the dynamic risk thresholds is set to 1 day, thereby ensuring consistency in the model training, calibration, and online updates and meeting the needs of subsequent early warning judgments.

[0041] Further, in step 500 of this embodiment, after obtaining the dynamic risk threshold output in step 400 and the real-time copyright risk indicator data generated in step 300, this embodiment performs early warning judgment processing to determine the current copyright risk status of the managed object. Specifically, this embodiment matches each indicator value in the real-time copyright risk indicator data with its corresponding dynamic risk threshold to form an indicator matching result. The indicator matching result is used to characterize whether each indicator value reaches or exceeds its corresponding dynamic risk threshold within the current statistical period, thereby reflecting the real-time risk level of the managed object under different risk dimensions. In this embodiment, the matching process is performed once in each statistical period, and corresponding indicator matching results are formed for the authorization status indicator, usage behavior indicator, dispute record indicator, and dissemination attention indicator, so that the risk status of each indicator can be independently characterized and retained for subsequent comprehensive judgment.

[0042] In this embodiment, based on the obtained indicator matching results, the indicator matching results are further processed by rule matching according to preset early warning triggering conditions to generate rule matching results. The preset early warning triggering conditions refer to a set of rules used to describe the early warning triggering logic, specifying under what indicator matching conditions an early warning event should be triggered; the rule matching results are used to characterize whether the corresponding early warning rule is satisfied. In this embodiment, the early warning rules include at least single-indicator triggering rules and multi-indicator combination triggering rules. The single-indicator triggering rule characterizes the early warning condition when any indicator matching result indicates that its value exceeds the corresponding dynamic risk threshold. The multi-indicator combination triggering rule characterizes the early warning condition when multiple indicators simultaneously meet the preset matching conditions within the same statistical period. In this embodiment, at least two types of early warning rules are evaluated in each statistical period to ensure that the early warning judgment results can cover different risk triggering scenarios.

[0043] In this embodiment, after generating the indicator matching result and the rule matching result, a comprehensive early warning judgment result is formed, and it is determined whether the early warning judgment result meets the preset early warning triggering condition; when it does, a corresponding early warning event is generated. The early warning event is used to characterize the copyright risk status of the managed object that needs attention or handling within the current statistical period, and serves as the input basis for generating an interpretable early warning report in step 600. In this embodiment, the early warning event is generated in statistical period as the smallest unit, and each managed object generates at most one early warning event within the same statistical period, thereby avoiding repeated triggering and ensuring the stability and manageability of the early warning result; at the same time, early warning judgment results that do not meet the preset early warning triggering condition are only saved as historical judgment records for subsequent risk evolution analysis without triggering early warning events.

[0044] Furthermore, in step 600 of this embodiment, after generating the early warning event in step 500, this embodiment generates an interpretable early warning report for the early warning event to present the formation process, triggering basis, and related data background of the early warning event in a structured manner. The early warning report refers to an explanatory data set formed around a single early warning event, used to show the management end the risk status corresponding to the early warning event and its judgment basis; in this embodiment, the early warning report includes at least a list of abnormal indicators, the source identifiers corresponding to the abnormal indicators, the triggered early warning rules, and explanatory content associated with dynamic risk thresholds, so that the management end can directly obtain key information related to the early warning event without tracing back the original data processing flow.

[0045] In this embodiment, regarding the generation of the abnormal indicator list, this embodiment filters out indicators whose values ​​reach or exceed the corresponding dynamic risk thresholds from the indicator matching results formed in step 500, and identifies them as abnormal indicators. For each abnormal indicator, this embodiment records its abnormal indicator value, the dynamic risk threshold corresponding to that abnormal indicator value, and the comparison result of the abnormal indicator value relative to the dynamic risk threshold in the early warning report, which is used to characterize the risk level of the indicator in the current statistical period. In this embodiment, the abnormal indicator list contains at least one abnormal indicator, and the number of abnormal indicators included in the same early warning event does not exceed four, to ensure that the early warning report is clear and readable while being information-rich.

[0046] In this embodiment, to clarify the triggering logic of the warning event, the warning report further includes a warning rule that triggers the warning event, and the matching conditions of the corresponding indicator are marked in conjunction with the list of abnormal indicators. Simultaneously, this embodiment utilizes the source identifier written in step 200 to associate each abnormal indicator with its corresponding data source, and clearly marks the source identifier corresponding to the abnormal indicator in the warning report, indicating that the abnormal indicator originates from at least one of judicial judgment data, library business system data, or copyright database data. This method enables the management end to directly identify the source dimension of the abnormal indicator, providing data guidance for subsequent manual verification and handling.

[0047] In this embodiment, to further illustrate the background of the dynamic risk threshold formation, the early warning report includes a timestamp of the real-time environmental factor data used to generate the dynamic risk threshold and the corresponding threshold adjustment result. This is used to characterize the impact of changes in hot topics or policies on the threshold adjustment within the current statistical period. The threshold adjustment result is consistent with the real-time environmental factor data in the time dimension and is associated with the dynamic risk threshold corresponding to the abnormal indicator. This allows the management end to understand that the early warning event not only originates from changes in the indicators of the managed object itself but is also related to changes in the external environment. Finally, this embodiment sends the generated early warning report to the management end to assist managers in analyzing, judging, and making decisions regarding the early warning event.

[0048] As an optional implementation method, corresponding to the above method, such as Figure 2 As shown, this embodiment also provides a library copyright risk intelligent early warning system based on multi-source heterogeneous data and dynamic risk thresholds, including: The multi-source data acquisition unit is used to acquire multi-source heterogeneous data of the management objects used to characterize the copyright risks of the library, as well as environmental factor data used for risk assessment of the management objects; the multi-source heterogeneous data includes at least judicial judgment data, library business system data, and copyright database data; the environmental factor data includes hot topic data and policy change data, and the environmental factor data carries a time identifier and is continuously updated according to a preset update cycle to form an environmental factor sequence; The data fusion processing unit is used to perform field mapping, entity association and structure unification on the multi-source heterogeneous data based on a preset fusion data model, output a fusion feature dataset, and write source identifiers for the records in the fusion feature dataset to characterize the data source of the multi-source heterogeneous data. The risk indicator construction unit is used to construct a copyright risk indicator system based on the fused feature dataset, and generate historical copyright risk indicator data and real-time copyright risk indicator data for each of the managed objects according to a preset granularity. The dynamic threshold generation unit is used to input the historical copyright risk indicator data and the historical environmental factor data extracted from the environmental factor sequence according to a preset historical time window into a machine learning time series model for training, output the threshold adjustment result, and combine the threshold adjustment result with the real-time environmental factor data obtained from the environmental factor sequence according to the preset update cycle to output a dynamic risk threshold. The early warning determination and event generation unit is used to match the real-time copyright risk indicator data with the dynamic risk threshold, output the early warning determination result, and generate an early warning event when the early warning determination result meets the preset early warning triggering conditions. The early warning report generation and output unit is used to generate an interpretable early warning report for the early warning event and send the early warning report to the management terminal to assist in decision-making. The early warning report includes a list of abnormal indicators, the source identifiers corresponding to the abnormal indicators, the triggered early warning rules, and an explanation of the association between the dynamic risk thresholds.

[0049] The beneficial effects of this invention are as follows: This invention acquires judicial judgment data, library business system data, and copyright database data within the same process, and generates a fused feature dataset through field mapping, entity association, and unified structure. This enables the copyright status, usage behavior, and rights boundaries of managed objects from different sources to be uniformly collected and consistently represented. It overcomes the problems of incomplete risk assessment information and high manual verification costs caused by data dispersion and data silos in the prior art, thereby improving the coverage and objectivity of copyright risk identification.

[0050] This invention introduces hot topic data and policy change data that carry time stamps and are continuously updated to form a sequence of environmental factors. It also uses historical time windows and update cycles to extract historical and real-time environmental factor data with a unified standard. This enables risk assessment to be synchronized in both the time and environmental dimensions, overcoming the problem that existing technologies rely solely on static business data and are unable to respond to fluctuations in risk intensity caused by hot topics and policy changes. This enhances the adaptability of early warning systems to changes in the external environment.

[0051] This invention obtains threshold adjustment results by inputting historical copyright risk indicator data and historical environmental factor data into a machine learning time series model for training. It then combines real-time environmental factor data to output dynamic risk thresholds that correspond one-to-one with each indicator. This allows the thresholds to adaptively adjust with historical evolution trends and current environmental conditions, avoiding false alarms and missed alarms caused by fixed thresholds in different stages and scenarios. This improves the accuracy, stability, and scenario robustness of early warning judgments.

[0052] This invention achieves a hierarchical judgment chain from "indicator layer - rule layer - event layer" by matching real-time copyright risk indicator data with dynamic risk thresholds to form indicator matching results during the early warning judgment stage. This makes the early warning triggering logic clear and controllable, and overcomes the problems of inconsistent early warning triggering criteria and difficulty in distinguishing between single indicator anomalies and multi-indicator linkage risks in the prior art. This improves the pertinence of risk handling and the operability of management.

[0053] This invention generates an interpretable early warning report after an early warning event is triggered. It clearly provides a list of abnormal indicators, the values ​​of the abnormal indicators and their corresponding dynamic risk thresholds and comparison results, the source identifiers of the abnormal indicators, the triggering early warning rules, and an explanation related to the time identifiers and threshold adjustment results of real-time environmental factor data. This solves the problem of "black box" early warning results and difficulty in locating the source of anomalies and the triggering basis in the prior art. Thus, it provides the management with evidence clues and decision support that can be directly verified and handled.

[0054] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0055] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for intelligent early warning of library copyright risks based on multi-source heterogeneous data and dynamic risk thresholds, characterized in that, include: Acquire multi-source heterogeneous data of management objects used to characterize library copyright risks and environmental factor data used for risk assessment of the management objects; the multi-source heterogeneous data includes at least judicial judgment data, library business system data and copyright database data; the environmental factor data includes hot topic data and policy change data, and the environmental factor data carries a time identifier and is continuously updated according to a preset update cycle to form an environmental factor sequence; Based on a preset fusion data model, the multi-source heterogeneous data is mapped to fields, associated with entities, and unified in structure. A fusion feature dataset is output, and a source identifier is written to the records in the fusion feature dataset to characterize the data source of the multi-source heterogeneous data. A copyright risk indicator system is constructed based on the fused feature dataset, and historical copyright risk indicator data and real-time copyright risk indicator data are generated for each of the managed objects according to a preset granularity. The historical copyright risk indicator data and the historical environmental factor data extracted from the environmental factor sequence according to a preset historical time window are input into a machine learning time series model for training, and the threshold adjustment result is output. The threshold adjustment result is combined with the real-time environmental factor data obtained from the environmental factor sequence according to the preset update cycle to output a dynamic risk threshold. The real-time copyright risk indicator data is matched with the dynamic risk threshold, and an early warning judgment result is output. An early warning event is generated when the early warning judgment result meets the preset early warning triggering conditions. An interpretable early warning report is generated for the aforementioned early warning event, and the early warning report is sent to the management terminal to assist in decision-making. The early warning report includes a list of abnormal indicators, the source identifiers corresponding to the abnormal indicators, the triggered early warning rules, and an explanation of the association between the dynamic risk thresholds.

2. The intelligent early warning method for library copyright risks based on multi-source heterogeneous data and dynamic risk thresholds according to claim 1, characterized in that, The judicial ruling data includes judgment documents from the China Trial Network; the library business system data includes resource usage records and authorization management information related to the managed object; and the copyright database data includes ownership information and authorization scope information related to the managed object.

3. The intelligent early warning method for library copyright risks based on multi-source heterogeneous data and dynamic risk thresholds according to claim 1, characterized in that, The hot topic data includes hot entries and popularity information related to copyright-related topics; the policy change data includes the release information and effective time information of copyright-related policy documents, and the time identifier is used to represent the occurrence time or effective time of the hot topic data and the policy change data.

4. The intelligent early warning method for library copyright risks based on multi-source heterogeneous data and dynamic risk thresholds according to claim 1, characterized in that, The field mapping includes unifying similar fields from different sources into fused fields based on preset field mapping relationships; the entity association includes associating multiple records pointing to the same managed object based on preset entity association rules; and the structure unification includes writing the associated records into the fused feature dataset according to a unified data structure.

5. The intelligent early warning method for library copyright risk based on multi-source heterogeneous data and dynamic risk thresholds according to claim 1, characterized in that, The source identifier includes at least a judicial ruling source identifier, a library business system source identifier, and a copyright database source identifier, and is used to locate the data source corresponding to the abnormal indicator in the warning report.

6. The intelligent early warning method for library copyright risks based on multi-source heterogeneous data and dynamic risk thresholds according to claim 1, characterized in that, The copyright risk indicator system includes authorization status indicators, usage behavior indicators, dispute record indicators, and dissemination attention indicators. Both the historical copyright risk indicator data and the real-time copyright risk indicator data include the indicator values ​​of the authorization status indicators, usage behavior indicators, dispute record indicators, and dissemination attention indicators.

7. The intelligent early warning method for library copyright risks based on multi-source heterogeneous data and dynamic risk thresholds according to claim 1, characterized in that, The machine learning time series model is a Long Short-Term Memory (LSTM) network model, and the threshold adjustment result represents the threshold adjustment amount or direction of each indicator in the copyright risk indicator system, and is used to generate the dynamic risk threshold corresponding to each indicator.

8. The intelligent early warning method for library copyright risk based on multi-source heterogeneous data and dynamic risk thresholds according to claim 1, characterized in that, The early warning determination result includes the indicator matching result and the rule matching result; The indicator matching result is used to characterize the comparison result between the indicator value in the real-time copyright risk indicator data and the corresponding dynamic risk threshold, and the rule matching result is used to characterize whether the warning rule corresponding to the preset warning trigger condition is met.

9. The intelligent early warning method for library copyright risk based on multi-source heterogeneous data and dynamic risk thresholds according to claim 1, characterized in that, The early warning report also includes: the abnormal indicator values ​​of each abnormal indicator in the abnormal indicator list, the dynamic risk threshold corresponding to the abnormal indicator values, and the comparison results of the abnormal indicator values ​​relative to the dynamic risk threshold; the triggered early warning rules; the source identifier corresponding to the abnormal indicator; and the time identifier of the real-time environmental factor data used to generate the dynamic risk threshold and the threshold adjustment results.

Images