Metadata exception prediction method and device, medium and product

Abstract

The invention provides a metadata anomaly prediction method and device, a medium and a product, and relates to the technical field of metadata analys.The method comprises the steps that by obtaining current metadata attribute time sequence data of all metadata entities, a preset LSTM model is input, and metadata attributes of a next time sequence are predicted; obtaining an attention score of each sequence element in the next time sequence by using a Transform model; according to the time sequence change characteristics of the attention scores, predicting the probability that the metadata attributes are changed and abnormal in the next time sequence; when the probability exceeds a preset probability threshold value, determining that the metadata attribute is changed abnormally; and determining the influence range of the change exception through tracing the associated metadata entity in the knowledge graph. The problems that in the prior art, prospective prediction is lacked for metadata change exception, and the exception influence range is difficult to determine are solved.

Description

Technical Field

[0001] This invention relates to the field of metadata analysis technology, and in particular to a method, apparatus, medium, and product for predicting metadata anomalies. Background Technology

[0002] Driven by the wave of digital transformation, metadata, as the "data gene" of data, is deeply embedded in the core of various information systems and business processes. Its accuracy, consistency, and stability directly determine the scientific nature of data-driven decision-making and the continuity of business operations. Especially in the airport industry, with its complex business scenarios and highly integrated information systems, metadata management faces unprecedented challenges due to the collaborative operation of dozens of business systems such as flight scheduling, passenger services, and security management. On the one hand, the fragmentation of metadata standards among various systems leads to frequent semantic ambiguities and format conflicts; on the other hand, the accelerated pace of business iteration causes the scale of metadata to grow exponentially, and issues such as dynamic changes in field attributes and complex intertwined permission systems exacerbate the fragility of the metadata ecosystem.

[0003] In practice, systemic risks caused by metadata anomalies have become a pain point in the industry. If changes to metadata in upstream systems are not promptly synchronized downstream, it can easily lead to disruptions in the business chain. For example, if an airline's ticketing system adjusts field lengths without timely notification to the settlement system, it can directly cause bill generation failures, impacting financial settlement efficiency and customer experience. Furthermore, risk warning mechanisms built on metadata lineage are constrained by the integrity of the lineage. If data flow records are missing or the link is interrupted, the impact of changes to critical systems may be missed, leading to unrecognized potential risks and ultimately triggering a chain reaction of data errors and task scheduling failures. Faced with these complex challenges, traditional rule-based metadata management methods are insufficient to meet the needs of real-time monitoring and intelligent early warning, necessitating the introduction of innovative technologies. Summary of the Invention

[0004] The embodiments of the present invention provide a method, apparatus, medium and product for predicting metadata anomalies, which aims to solve the problem of lack of forward-looking prediction of metadata change anomalies and inability to determine the scope of the impact of anomalies, making it difficult to plan metadata system maintenance or upgrade strategies in advance.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a metadata anomaly prediction method, comprising the following steps: Obtain the current metadata attribute time series data for each metadata entity, wherein the metadata attributes include structural attributes, data quality indicators, permission attributes, and / or change records; The metadata attribute time series data is input into a predetermined LSTM model to predict the metadata attributes of the next time series. The metadata attributes of the next time series are input into a predetermined Transformer model. The attention score of each sequence element in the next time series is obtained by using the self-attention mechanism of the Transformer model. The attention score is used to represent the degree of correlation between the metadata attributes and other sequence elements in the next time series. Based on the temporal variation characteristics of the attention scores of each sequence element in the next time series, predict the probability of metadata attributes changing abnormally in the next time series; the abnormal changes include metadata conflicts and / or quality degradation. When the probability of an abnormal change in a metadata attribute exceeds a preset probability threshold, the metadata attribute is determined to have undergone an abnormal change. For metadata entities whose metadata attributes have undergone abnormal changes, trace their upstream and downstream metadata entities in their respective knowledge graphs to determine the scope of the impact of the abnormal changes.

[0006] Furthermore, the structural attributes include field name, field type, field length, and / or whether the field can be nullable; the data quality indicators include field missing rate, field outlier ratio, and / or field data consistency; the permission attributes include access permission level and / or changes in the number of permission holders; and the change records include change time interval and / or change frequency.

[0007] Furthermore, the degree of association is obtained by: generating a query vector, a key vector, and a value vector for each sequence element of the metadata attribute in the next time series through a linear transformation; for each sequence element, calculating the dot product between the query vector and the key vectors of all sequence elements in the next time series to obtain an attention score; and converting the attention score into a probability distribution through scaling and softmax operations, and then weighting and summing the value vectors to obtain an output representation of the degree of association between each sequence element and other sequence elements.

[0008] Furthermore, the Transformer model also identifies the correlation between metadata attribute changes between different metadata entities through a self-attention mechanism. This is used to predict abnormal changes in the same metadata attribute of other metadata entities based on the correlation when an abnormal change occurs in the metadata attribute of a certain metadata entity.

[0009] Furthermore, the knowledge graph uses metadata entities as nodes and the relationships between metadata entities as edges. The relationships between metadata entities include subordinate relationships, association relationships, permission relationships, business relationships, and / or change relationships.

[0010] Furthermore, it also includes one or more of the following: By pruning techniques, connections or neurons in the LSTM and Transformer models that have a lower impact on the prediction results than the preset conditions are removed. Quantization techniques are used to convert high-precision values ​​in the model into low-precision values. In the feature extraction process, the Transformer model uses depthwise separable convolutions to replace some fully connected layers or convolutional layers. Data augmentation techniques are used to generate synthetic metadata anomaly samples to expand the training dataset; In a large-scale metadata scenario, LSTM and Transformer models are pre-trained, and the model parameters are obtained as initialization parameters. Then, the initialization parameters are adjusted according to the predetermined low-resource environment goal.

[0011] Secondly, the present invention provides a metadata anomaly prediction device, including a memory and a processor, wherein the memory stores at least one program, and the at least one program is executed by the processor to implement the metadata anomaly prediction method as described above.

[0012] Thirdly, the present invention provides a computer-readable storage medium storing at least one program, wherein the at least one program is executed by a processor to implement the metadata anomaly prediction method as described above.

[0013] Fourthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the metadata anomaly prediction method as described above.

[0014] The above technical solution has the following technical effects: By acquiring the current time-series data of metadata attributes for each metadata entity and inputting it into a predefined LSTM model, the metadata attributes for the next time series are predicted. The metadata attributes for the next time series are then input into a predefined Transformer model, and the attention score of each sequence element of the metadata attribute in the next time series is obtained using the Transformer model's self-attention mechanism. Based on the temporal change characteristics of the attention scores of each sequence element in the next time series, the probability of an abnormal change in the metadata attribute is predicted. When the probability of an abnormal change in the metadata attribute exceeds a preset probability threshold, the abnormal change in the metadata attribute is determined. For the metadata entity whose metadata attribute has an abnormal change, its upstream and downstream metadata entities are traced in its knowledge graph to determine the scope of the impact of the abnormal change. This invention solves the problem that existing technologies lack forward-looking prediction of metadata change anomalies and cannot determine the scope of the anomaly's impact, making it difficult to plan metadata system maintenance or upgrade strategies in advance.

[0015] In a further embodiment, the present invention uses a knowledge graph to treat metadata entities as nodes and the relationships between metadata entities as edges. The relationships between metadata entities include subordinate relationships, association relationships, permission relationships, business relationships, and / or change relationships. When performing anomaly prediction, the information in the knowledge graph can be combined to analyze the potential impact of metadata changes on related business processes and associated metadata entities.

[0016] In a further embodiment, this invention removes connections or neurons in the LSTM and Transformer models whose impact on prediction results is below a preset condition through pruning techniques, and converts high-precision values ​​in the models to low-precision values ​​through quantization techniques. During feature extraction, the Transformer model uses depthwise separable convolutions to replace some fully connected layers or convolutional layers. Data augmentation techniques are used to generate synthetic metadata anomaly samples to expand the training dataset. The LSTM and Transformer models are pre-trained in a large-scale metadata scenario to obtain model parameters as initialization parameters, which are then adjusted according to a predetermined low-resource environment objective. This achieves reduced computational complexity and resource consumption while maintaining prediction performance, overcoming the limitations of low-resource environments. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a metadata anomaly prediction method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the metadata anomaly prediction device in one embodiment of the present invention. Detailed Implementation

[0018] To further illustrate the various embodiments, the present invention provides accompanying drawings. These drawings are part of the disclosure of the present invention, primarily used to illustrate the embodiments and to explain the operating principles of the embodiments in conjunction with the relevant descriptions in the specification. With reference to these drawings, those skilled in the art should be able to understand other possible implementations and the advantages of the present invention. Components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0019] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments.

[0020] Example 1: Figure 1 This is a flowchart illustrating a metadata anomaly prediction method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method of this embodiment includes the following steps: Obtain the current time-series data of metadata attributes for each metadata entity. Metadata attributes include structural attributes, data quality indicators, permission attributes, and / or change records. In one specific implementation, airport metadata is used as the specific scenario, and the aforementioned metadata entities are airport metadata entities. In one specific implementation, structural attributes include field name, field type, field length, and / or whether a field can be nullable, such as airport name, flight information type, information length limit, and whether it is a required field; data quality indicators include field missing rate, field outlier ratio, and / or field data consistency, such as the completeness of flight information, the proportion of abnormal flights, and data consistency verification; permission attributes include access permission level and / or changes in the number of permission holders, involving the management of airport metadata access permissions and dynamic statistics of permission users; change records include change time intervals and / or change frequencies, recording the update history and update frequency of airport metadata attributes.

[0021] In one specific implementation, before using a temporal neural network to analyze metadata incorporating airport operation scenarios, this metadata needs to be preprocessed and its features extracted to convert it into a format suitable for model input. This metadata is typically stored in various structured and semi-structured forms, such as database table structure definitions, configuration files, and log records, documenting all aspects of airport operations. First, the raw metadata is cleaned and normalized, unifying the data format and encoding methods, and removing invalid or duplicate data records. Then, key features are extracted as input variables for the model; these features comprehensively cover the metadata attributes. To better capture the temporal characteristics of this metadata, the features are arranged in chronological order to form a sequence of data. For example, for each metadata entity, the aforementioned feature values ​​are extracted at fixed time intervals, such as daily or weekly, forming a feature sequence. Simultaneously, the feature sequence is normalized, scaling the numerical features to a specific range, such as 0~1 or -1~1, thereby improving the stability and convergence speed of model training.

[0022] The time series data, which incorporates metadata attributes from airport operations, is input into a pre-defined LSTM model to accurately predict the metadata attributes of the next time series, providing strong support for the airport's future operations.

[0023] In this embodiment, the LSTM model is a variant of the Recurrent Neural Network (RNN) specifically designed to handle long-term dependencies in sequential data. Its core structure includes a cell state, a forget gate, an input gate, and an output gate. The cell state stably transmits information along time steps, effectively mitigating the vanishing or exploding gradient problems that traditional RNNs encounter when processing long sequences. The forget gate determines which information needs to be discarded from the cell state, outputting a value between 0 and 1 using the sigmoid function; values ​​closer to 0 indicate greater need to discard the corresponding information. The input gate controls the degree to which new information enters the cell state, also based on multiplying the result of the sigmoid function with the element-wise of the candidate cell state before updating the cell state. The output gate determines which information to output based on the current cell state; its output is combined with the activated cell state as the hidden state for the current time step, used to pass to the next time step and generate the final output. In metadata anomaly prediction, LSTM can model sequential data where metadata changes over time, capturing long-term dependency features such as the gradual changing trends of metadata field values ​​and periodic fluctuations in data update frequency. For example, when considering the data type change process of a field in a database table, LSTM can learn about potential abnormal fluctuation patterns in the field's data before the change. For instance, numeric fields frequently showing values ​​close to boundary values ​​or unexpected special values. These subtle changes may foreshadow a subsequent abrupt change in the field's type. Therefore, LSTM can learn about these patterns in an airport database table, providing crucial early warning information for airport metadata management.

[0024] In one specific implementation, during the initial training of the LSTM model, the input feature sequence is first received. Through its internal forget gate, input gate, and output gate structure, the hidden state representation of each time step in the sequence is gradually learned. These hidden states encode the changing trends and local patterns of metadata over time. For example, for a feature sequence of permission change frequency of a permission object, LSTM can learn the rising or falling trends of permission change frequency in different time periods, as well as periodic change patterns such as more frequent permission adjustments at the beginning of each month.

[0025] The metadata attributes of the next time series are input into a predefined Transformer model. The attention score of each sequence element in the next time series is obtained by using the self-attention mechanism of the Transformer model. The attention score is used to represent the degree of correlation between the metadata attributes and other sequence elements in the next time series. In this embodiment, the self-attention mechanism of the Transformer model allows the model to pay attention to all other elements in the sequence while processing each element, thereby capturing the global dependencies between elements without having to process the sequence sequentially like an RNN, which greatly improves the efficiency of parallel computing.

[0026] In one specific implementation, in the application scenario of airport industry metadata, the degree of correlation is obtained as follows: Metadata attributes, such as flight information and passenger flow, are transformed linearly into query vectors, key vectors, and value vectors for each sequence element in the next time series. For each sequence element, such as flight information for a certain time period, the dot product of its query vector and the key vectors of all sequence elements in the next time series, such as flight information and passenger flow in subsequent time periods, is calculated to obtain an attention score. The attention score is then converted into a probability distribution through scaling and softmax operations, and the value vectors are then weighted and summed to finally obtain an output representation of the degree of correlation between each sequence element (i.e., flight information for a specific time period) and other sequence elements (i.e., flight information and passenger flow in other time periods). This method helps the airport industry to more accurately understand and predict flight operations and passenger flow trends in different time periods.

[0027] In one specific implementation, particularly in the metadata application scenario of the airport industry, during the Transformer model training process, the sequence features processed by the LSTM layer are passed to the Transformer layer. The Transformer layer utilizes a self-attention mechanism to globally readjust and combine feature information. This feature information may include time-series data of flight information, passenger flow changes, etc. By calculating the attention scores between different time steps, the model can determine which time step features are more important for the prediction of the current time step. For example, when predicting flight delays or cancellations, the Transformer layer may find a strong correlation between recent flight anomalies and similar historical events, or a correlation between a surge in passenger flow and increased airport operational pressure, thus giving these relevant time steps higher attention weights to fully utilize this key historical information to assist in prediction. The training objective of the model is to optimize the network parameters by minimizing the loss function between the predicted value and the real metadata anomaly labels, such as whether flight conflicts have occurred or whether airport service quality has deteriorated. A cross-entropy loss function combined with the Adam optimization algorithm is used to iteratively update the model's weights, allowing the model to gradually learn the mapping relationship between airport metadata change patterns and anomalies, thereby more effectively predicting and responding to various challenges in airport operations.

[0028] In one specific implementation, the Transformer model also uses a self-attention mechanism to identify the correlation between metadata attribute changes between different metadata entities. This is used to predict abnormal changes in the same metadata attribute of other metadata entities based on the correlation when an abnormal change occurs in the metadata attribute of a certain metadata entity.

[0029] Based on the temporal variation characteristics of the attention scores of each sequence element in the next time series, predict the probability of metadata attributes changing abnormally in the next time series, where abnormal changes include metadata conflicts and / or quality degradation. When the probability of an abnormal change in a metadata attribute exceeds a preset probability threshold, the metadata attribute is determined to have undergone an abnormal change. In one specific implementation, when the predicted probability exceeds a set threshold, the system triggers an anomaly warning mechanism to promptly notify data management personnel or relevant business personnel of potential metadata anomaly risks.

[0030] For metadata entities whose metadata attributes have undergone abnormal changes, trace their upstream and downstream metadata entities in their respective knowledge graphs to determine the scope of the impact of the abnormal changes.

[0031] In one specific implementation, the knowledge graph uses metadata entities such as database tables, fields, indexes, and user permissions as nodes, and the relationships between metadata entities as edges. These relationships include subordinate relationships (e.g., a field belongs to a table), association relationships (e.g., foreign key associations between tables), permission relationships (e.g., user permissions on tables or fields), and business relationships (e.g., the contextual relationships and / or change relationships of metadata entities in business processes). In another specific implementation, change relationships are obtained from recorded metadata change history events. These change history events include the time of the change, the type of change (e.g., field type modification, permission addition and / or permission deletion), the reason for the change (e.g., business requirement adjustments, system optimization), and the scope of the change (e.g., affected user groups, other related metadata entities).

[0032] In one specific implementation, when performing anomaly prediction, information from the knowledge graph is combined, especially in the application scenario of airport industry metadata, to deeply analyze the potential impact of metadata changes on related business processes and associated metadata entities. For example, if the field types of the flight information table in the airport operations system change, the system can quickly identify the key position of this table in the business process and its close relationship with other tables and business links based on the knowledge graph. This analytical capability enables the system to predict that this change may trigger a series of upstream and downstream metadata conflicts or quality degradation issues, including but not limited to anomalies in the generation of related flight dynamic reports, errors in passenger data statistics, and adaptation failures of airport-related applications, thereby ensuring the efficiency and accuracy of airport operations.

[0033] In one specific implementation, to address the limitations of low-resource environments, this embodiment develops a model with lightweight anomaly detection properties. While maintaining predictive performance, it significantly reduces computational complexity and resource consumption through model compression and optimization techniques, including one or more of the following: Pruning techniques are used to remove connections or neurons in LSTM and Transformer models that have a lower impact on prediction results than a preset condition. For example, sensitivity analysis is performed on trained LSTM and Transformer models to identify which connections have small absolute weight values ​​and contribute little to the model output, and these connections are then pruned to reduce the model size. Quantization techniques are used to convert high-precision values, such as 32-bit floating-point numbers, into low-precision values, such as 8-bit or 16-bit integers, reducing storage space and computational cost while maintaining basic model performance. In the feature extraction process, the Transformer model uses depthwise separable convolutions to replace some fully connected layers or convolutional layers. Data augmentation techniques are used to synthesize metadata anomalous samples, such as simulating field type mutations and abnormal permission diffusion, to expand the training dataset and improve the model’s generalization ability under limited data. In large-scale metadata scenarios, LSTM and Transformer models are pre-trained to obtain model parameters as initialization parameters. These initialization parameters are then adjusted according to predetermined low-resource environment goals, enabling the model to quickly adapt to anomaly detection requirements in different low-resource scenarios.

[0034] Example 2: Figure 2 This is a schematic diagram of the structure of a metadata anomaly prediction device according to an embodiment of the present invention, as shown below. Figure 2 As shown, the device includes a processor 201, a memory 202, a bus 203, and a computer program stored in the memory 202 and executable on the processor 201. The processor 201 includes one or more processing cores. The memory 202 is connected to the processor 201 via the bus 203. The memory 202 is used to store program instructions. When the processor executes the computer program, it implements the steps in the above-described method embodiment of Embodiment 1 of the present invention.

[0035] Furthermore, as an executable solution, the metadata anomaly prediction device can be a computer unit, which can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The computer unit may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above-described computer unit structure is merely an example and does not constitute a limitation on the computer unit. It may include more or fewer components, or combine certain components, or use different components. For example, the computer unit may also include input / output devices, network access devices, buses, etc., and this embodiment of the invention does not limit this.

[0036] Furthermore, as an executable solution, the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the computer unit, connecting various parts of the entire computer unit via various interfaces and lines.

[0037] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the computer unit by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital card (SD card), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0038] Example 3: The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method described in the embodiments of the present invention.

[0039] If the modules / units integrated in the computer unit are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), and software distribution media, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0040] Example 4: The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the metadata anomaly prediction method as described above.

[0041] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.

Claims

1. A metadata anomaly prediction method characterized by, The method comprises the following steps: obtaining current metadata attribute time series data of each metadata entity, the metadata attributes including structure attributes, data quality indicators, permission attributes and / or change records; inputting the metadata attribute time series data into a predetermined LSTM model to predict the metadata attributes of the next time series; inputting the metadata attributes of the next time series into a predetermined Transformer model, and using the self-attention mechanism of the Transformer model to obtain the attention scores of each sequence element in the next time series, the attention scores representing the correlation degree between each sequence element and other sequence elements in the next time series; predicting the probability of change anomaly of the metadata attributes in the next time series according to the time sequence variation characteristics of the attention scores of each sequence element in the next time series; the change anomaly includes metadata conflict and / or quality degradation; determining that the metadata attributes have change anomaly when the probability of change anomaly of the metadata attributes exceeds a preset probability threshold; for the metadata entity whose metadata attributes have change anomaly, tracing other metadata entities upstream and downstream thereof in the knowledge graph to which it belongs, and determining the influence range of the change anomaly. 2.The metadata anomaly prediction method of claim 1, wherein, The structure attributes include field name, field type, field length and / or whether the field is null; the data quality indicators include field missing rate, field abnormal value proportion and / or field data consistency; the permission attributes include access permission level and / or number change of permission holders; the change records include change time interval and / or change frequency. 3.The metadata anomaly prediction method of claim 1, wherein, The correlation degree is obtained by the following method: generating a query vector, a key vector and a value vector from each sequence element in the metadata attributes in the next time series through linear transformation; for each sequence element, calculating the dot product of the query vector and the key vectors of all sequence elements in the next time series to obtain the attention score; converting the attention score into a probability distribution through scaling and softmax operation, and performing weighted summation on the value vector to obtain the output representation of the correlation degree between each sequence element and other sequence elements. 4.The metadata anomaly prediction method of claim 1, wherein, The Transformer model also identifies the correlation of metadata attribute changes between different metadata entities through the self-attention mechanism, which is used to predict the change anomaly of the same metadata attribute of other metadata entities when the metadata attribute change of a certain metadata entity is abnormal.

5. The metadata anomaly prediction method of claim 1, wherein, The knowledge graph takes metadata entities as nodes and the relationships between the metadata entities as edges, and the relationships between the metadata entities include subordinate relationship, correlation relationship, permission relationship, business relationship and / or change relationship.

6. The metadata anomaly prediction method of claim 1, wherein, It also comprises one or more of the following: Through pruning technology, the connections or neurons in the LSTM model and the Transformer model that have an impact on the prediction results below a preset condition are removed, and through quantization technology, high-precision values in the model are converted into low-precision values. In the feature extraction process of the Transformer model, deep separable convolution is used to replace part of the fully connected layers or convolutional layers. Synthetic metadata anomaly samples are generated by data augmentation techniques to augment the training dataset; The LSTM model and the Transformer model are pre-trained in a predetermined large-scale metadata scenario, and model parameters are obtained as initialization parameters, and the initialization parameters are adjusted according to a predetermined low-resource environment target.

7. A metadata anomaly prediction apparatus characterized by comprising: The metadata anomaly prediction method comprises a memory and a processor, the memory stores at least one program, and the at least one program is executed by the processor to realize the metadata anomaly prediction method according to any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that, The storage medium stores at least one program, and the at least one program is executed by the processor to realize the metadata anomaly prediction method according to any one of claims 1 to 6.

9. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the metadata anomaly prediction method according to any one of claims 1 to 6.

Images