Report auditing method and device and storage medium

Abstract

The invention discloses a report auditing method and device and a storage medium. Relates to the field of artificial intelligence, and the method comprises the steps: determining a to-be-audited report and a reference table, the reference table being a basis for judging whether the to-be-audited report has abnormal index data; converting the to-be-audited report into a target report in a preset format; and inputting the target report and the reference table into a deep learning model, establishing a data consanguinity map according to the target report and the reference table through the deep learning model, auditing the target report according to the data consanguinity map, and outputting an auditing result for the to-be-audited report. Through the method and the device, the problem of low auditing efficiency depending on manual report auditing in related technologies is solved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and more specifically, to a method, apparatus, and storage medium for reviewing reports. Background Technology

[0002] In today's fast-paced business environment, the review of various financial reports is crucial to ensuring the transparency and compliance of corporate operations. However, traditional report review processes rely heavily on manual labor, especially when handling complex, data-intensive annual reports. Business experts or auditors must meticulously check the financial data in the reports page by page, line by line, to verify their accuracy, logical coherence, and compliance. This process is not only time-consuming and labor-intensive but also susceptible to individual cognitive biases and fatigue, leading to low efficiency and difficulty in guaranteeing review quality. Furthermore, manual review faces enormous challenges when dealing with thousands of pages of annual reports and massive amounts of Excel spreadsheets. The large amount of tabular data, key indicators, and complex interrelationships contained in financial reports require auditors to possess deep financial knowledge and extremely high levels of attention to identify potential numerical errors, logical inconsistencies, or compliance issues.

[0003] There is currently no effective solution to the problem of low review efficiency caused by relying on manual review reports in related technologies. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, and storage medium for reviewing reports, in order to solve the problem of low review efficiency in related technologies that rely on manual review of reports.

[0005] To achieve the above objectives, according to one aspect of this application, a method for reviewing reports is provided. The method includes: determining a report to be reviewed and a reference table, wherein the reference table is the basis for determining whether the report to be reviewed contains abnormal indicator data; converting the report to be reviewed into a target report in a preset format; inputting the target report and the reference table into a deep learning model; using the deep learning model to establish a data lineage graph based on the target report and the reference table; reviewing the target report based on the data lineage graph; and outputting the review result for the report to be reviewed.

[0006] Optionally, converting the report to be reviewed into a target report in a preset format includes: traversing each cell of the table in the report to be reviewed and reading the text content and attribute information of each cell; skipping merged cells that have already been read if they are already merged cells in the table in the report to be reviewed; reconstructing the table structure of the table in the report to be reviewed based on the text content and attribute information of each cell to generate a target table; and generating a target report based on the target table and the content in the report to be reviewed other than the table.

[0007] Optionally, inputting the target report and reference table into the deep learning model includes: determining the business logic for reviewing the target report, and determining prompt words based on the business logic for reviewing the target report; inputting the target report, reference table, and prompt words into the deep learning model.

[0008] Optionally, establishing a data lineage graph based on the target report and reference table using a deep learning model includes: extracting indicators, indicator data, and relationships between indicators from the target report; extracting indicators, indicator data, and relationships between indicators from the reference table; and constructing a data lineage graph by using the indicators in the target report and the indicators in the reference table as nodes, and the relationships between indicators in the target report and the relationships between indicators in the reference table as edges. The node information in the data lineage graph includes indicator data from the target report and indicator data from the reference table.

[0009] Optionally, reviewing the target report based on the data lineage graph includes: traversing the edges in the data lineage graph to determine whether each indicator data meets the preset rules; identifying indicator data in the target report that does not meet the preset rules as abnormal indicator data, and determining the source of the abnormal indicator data by tracing back the edges in the data lineage graph.

[0010] Optionally, reviewing the target report based on the data lineage graph includes: identifying common fields in the target report and the reference table, and associating the common fields in the target report and the reference table; defining a verification query statement based on the correlation relationship; determining whether the indicator data in the target report meets the data logic based on the query statement and the common fields; identifying indicator data in the target report that does not meet the data logic as abnormal indicator data, and determining the source of the abnormal indicator data by tracing back the edges in the data lineage graph.

[0011] Optionally, after reviewing the target report based on the data lineage map and outputting the review results for the report to be reviewed, the method further includes: dividing the target report into multiple paragraphs, performing position encoding on each paragraph, and generating multiple paragraph embedding vectors containing position encodings; integrating the multiple paragraph embedding vectors into a comprehensive vector representation; reviewing the comprehensive vector representation to determine abnormal information in the target report; and adding the abnormal information to the review results to update the review results.

[0012] To achieve the above objectives, according to another aspect of this application, a report review apparatus is provided. The apparatus includes: a first determining unit, configured to determine a report to be reviewed and a reference table, wherein the reference table is the basis for determining whether the report to be reviewed contains abnormal indicator data; a conversion unit, configured to convert the report to be reviewed into a target report in a preset format; and a review unit, configured to input the target report and the reference table into a deep learning model, establish a data lineage graph based on the target report and the reference table through the deep learning model, review the target report based on the data lineage graph, and output the review result for the report to be reviewed.

[0013] Optionally, the conversion unit includes: a reading module, used to traverse each cell of the table in the report to be reviewed and read the text content and attribute information in each cell; a skipping module, used to skip merged cells that have already been read if they are already merged cells in the table of the report to be reviewed; a reconstruction module, used to reconstruct the table structure of the table in the report to be reviewed based on the text content and attribute information of each cell to generate a target table; and a generation module, used to generate a target report based on the target table and the content in the report to be reviewed other than the table.

[0014] Optionally, the review unit includes: a first determining module, used to determine the business logic for reviewing the target report, and to determine prompt words based on the business logic for reviewing the target report; and an input module, used to input the target report, reference table, and prompt words into the deep learning model.

[0015] Optionally, the review unit includes: a first extraction module for extracting indicators, indicator data, and relationships between indicators from the target report; a second extraction module for extracting indicators, indicator data, and relationships between indicators from the reference table; and a construction module for constructing a data lineage graph by using the indicators in the target report and the indicators in the reference table as nodes, and the relationships between indicators in the target report and the relationships between indicators in the reference table as edges, wherein the node information in the data lineage graph includes indicator data from the target report and indicator data from the reference table.

[0016] Optionally, the audit unit includes: a first judgment module, used to traverse the edges in the data lineage graph and judge whether each indicator data meets the preset rules; and a backtracking module, used to identify indicator data in the target report that do not meet the preset rules as abnormal indicator data, and to determine the source of the abnormal indicator data by tracing the edges in the data lineage graph.

[0017] Optionally, the audit unit includes: an association module, used to identify common fields in the target report and the reference table, and to associate the common fields in the target report and the reference table; a definition module, used to define the verification query statement based on the reconciliation relationship; a second judgment module, used to determine whether the indicator data in the target report meets the data logic based on the query statement and the common fields; and a second determination module, used to determine the indicator data in the target report that does not meet the data logic as abnormal indicator data, and to determine the source of the abnormal indicator data by tracing the edges in the data lineage graph.

[0018] Optionally, the device further includes: a segmentation unit, used to segment the target report into multiple segments after reviewing the target report according to the data lineage map and outputting the review result for the report to be reviewed, and to perform position encoding on each segment to generate multiple segment embedding vectors containing position encoding; an integration unit, used to integrate the multiple segment embedding vectors into a comprehensive vector representation; a second determination unit, used to review the comprehensive vector representation and determine the abnormal information in the target report; and an update unit, used to add the abnormal information to the review result to update the review result.

[0019] According to another aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform any kind of report auditing method.

[0020] According to another aspect of this application, an electronic device is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include an auditing method for performing any type of report.

[0021] According to another aspect of this application, a computer program product is provided, including computer instructions, which, when executed by a processor, implement the steps of a method for reviewing reports according to any one of the foregoing.

[0022] In this embodiment, by determining the report to be reviewed and a reference table (where the reference table serves as the basis for judging whether the report to be reviewed contains abnormal indicator data), the report to be reviewed is converted into a target report in a preset format. The target report and the reference table are then input into a deep learning model, which constructs a data lineage graph based on the target report and the reference table. The target report is reviewed based on the data lineage graph, and the review result for the report to be reviewed is output. This solves the technical problem of low review efficiency when relying on manual review of reports. In this application, the report to be reviewed and the reference table are first determined. The reference table serves as the basis for judging whether the indicator data in the report to be reviewed is abnormal. The report to be reviewed is converted into a target report in a preset format. The deep learning model receives the target report and the reference table and constructs a data lineage graph based on them. By analyzing the relationship between indicator data in the data lineage graph, it determines whether there is abnormal indicator data and outputs the review result for the report to be reviewed, thus achieving the technical effect of improving the efficiency and accuracy of report review. Attached Figure Description

[0023] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0024] Figure 1 A hardware block diagram of a computer terminal for implementing a report review method is shown.

[0025] Figure 2 This is a flowchart of the report review method provided according to the embodiments of this application;

[0026] Figure 3 This is a schematic diagram of the report review method provided according to the embodiments of this application;

[0027] Figure 4 This is a schematic diagram of a report review device provided according to an embodiment of this application;

[0028] Figure 5 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] It should be noted that all relevant information (including but not limited to report information, reference table information, etc.) and data (including but not limited to data used for review, data used for analysis, etc.) involved in this application are information and data authorized by the user or fully authorized by all parties. For example, if there is an interface between this system and the relevant user or organization, before obtaining the relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain the relevant information after receiving the consent information from the aforementioned user or organization.

[0032] Example 1

[0033] According to an embodiment of this application, a method embodiment for reviewing a report is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0034] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a report review method is shown. Figure 1As shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a Universal Serial Bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0035] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0036] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the report review method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby implementing the aforementioned report review method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0037] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0038] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0039] Under the aforementioned operating environment, this application provides the following: Figure 2 The report review method is shown. Figure 2 This is a flowchart of the report review method according to Embodiment 1 of this application.

[0040] Step S201: Determine the report to be audited and the reference table. The reference table is the basis for judging whether there are abnormal indicator data in the report to be audited.

[0041] Optionally, the report to be audited refers to documents such as financial reports and annual reports that require auditing, containing a large amount of financial data, indicators, and related textual descriptions. The reference table serves as the basis for judging whether the indicator data in the report to be audited is abnormal. The reference table can be a structured dataset, such as an Excel spreadsheet, containing historical data, industry standards, compliance requirements, etc., from multiple dimensions, used to compare and verify whether the data in the report to be audited conforms to the expected format, numerical range, and logical relationships.

[0042] Step S202: Convert the report to be reviewed into a target report in a preset format.

[0043] Optionally, the report to be reviewed can be converted into a target report in a preset format. The conversion process may include parsing the report structure, extracting key data, and reorganizing the table structure, so that the deep learning model can directly read and understand the information in the report. For example, if the table in the report to be reviewed is a multidimensional table with many attributes, this multidimensional table can be converted into a two-dimensional data table to facilitate subsequent comparison.

[0044] Step S203: Input the target report and reference table into the deep learning model, build a data lineage map based on the target report and reference table, review the target report based on the data lineage map, and output the review results for the report to be reviewed.

[0045] Optionally, a deep learning model is a machine learning model based on neural networks that can automatically learn features from large amounts of data and make predictions or judgments, suitable for complex data analysis and pattern recognition tasks. By inputting a pre-formatted target report and a reference table into the deep learning model, a data lineage graph is constructed. Based on the data relationships in the lineage graph, the deep learning model can automatically review the target report, identify inconsistent or abnormal indicator data compared to the reference table, and finally output detailed review results. For example, the review results can be an annotated Word report, embedding details (difference points, difference items, difference amounts, custodian data) and a concise summary of conclusions at the points of difference.

[0046] The report review method provided in this application, through determining the report to be reviewed and a reference table (where the reference table serves as the basis for judging whether there is abnormal indicator data in the report to be reviewed), converts the report to be reviewed into a target report in a preset format; inputs the target report and the reference table into a deep learning model, which then constructs a data lineage graph based on the target report and the reference table, reviews the target report based on the data lineage graph, and outputs the review result for the report to be reviewed. This solves the technical problem of low review efficiency when relying on manual report review. In this application, the report to be reviewed and the reference table are first determined, with the reference table serving as the basis for judging whether the indicator data in the report to be reviewed is abnormal. The report to be reviewed is converted into a target report in a preset format. The deep learning model receives the target report and the reference table, and constructs a data lineage graph based on them. By analyzing the relationship between indicator data in the data lineage graph, it determines whether there is abnormal indicator data and outputs the review result for the report to be reviewed, thus achieving the technical effect of improving the efficiency and accuracy of report review.

[0047] To improve the efficiency of report review, optionally, in the report review method provided in this application embodiment, converting the report to be reviewed into a target report in a preset format includes:

[0048] The first step is to iterate through each cell in the table of the report to be reviewed and read the text content and attribute information of each cell.

[0049] Optionally, in the initial stage of format conversion, each table in the report to be reviewed will be traversed and the text content and attribute information of each cell will be read, such as the table subject, time information, indicator name, main project name, and sub-project name.

[0050] The second step is to skip merged cells in the table of the report to be reviewed if they have already been read.

[0051] Optionally, during the traversal, to avoid data duplication and redundancy, if a merged cell is encountered and its contents have already been read, subsequent reading operations for that merged cell can be automatically skipped.

[0052] The third step is to reconstruct the table structure in the report to be reviewed based on the text content and attribute information of each cell, and generate the target table.

[0053] Optionally, by analyzing the text content and attribute information of cells, the logical structure of the table can be reconstructed, transforming the complex structure of the original multi-level header table into a clear and standardized table format.

[0054] The fourth step is to generate the target report based on the target table and the contents of the report to be reviewed, excluding the table.

[0055] Alternatively, by combining the target table with other content in the report to be reviewed (such as text descriptions, titles, etc.), a target report with a clear structure and uniform format can be generated.

[0056] In summary, by following the steps above, the complex, multi-dimensional table structure in the report to be reviewed is reconstructed into a simple and clear target table. Combining the target table with the content in the report to be reviewed (excluding the table itself) generates a target report with a clear structure and uniform format, which helps improve the accuracy and efficiency of subsequent report reviews.

[0057] To improve the accuracy of report review, optionally, in the report review method provided in this application embodiment, inputting the target report and reference table into the deep learning model includes:

[0058] The first step is to determine the business logic used to review the target report, and then determine the prompt words based on the business logic.

[0059] Optionally, business logic refers to the specific rules and standards that must be followed during the audit process, including compliance requirements regarding the consistency of financial data, numerical ranges, and data formats. Before inputting the target report and reference tables into the deep learning model, it is necessary to first clarify the business logic upon which the target report is based. Based on this business logic, prompts can be constructed to guide the deep learning model on how to understand and process the data in the target report, how to compare it with the reference tables, and how to locate and identify potential data anomalies or non-compliance.

[0060] The second step is to input the target report, reference table, and prompt words into the deep learning model.

[0061] Optionally, the target report, reference table, and customized prompts can be fed as input to the deep learning model. The deep learning model can perform complex data comparison, logical analysis, and anomaly detection tasks, automatically identify inconsistencies or anomalies between the target report and the reference table, and generate a detailed audit report.

[0062] In summary, by following the steps outlined above and designing prompts for reviewing target reports, the combination of deep learning models and these prompts can improve review efficiency and accuracy.

[0063] To improve the accuracy of report review, optionally, in the report review method provided in this application embodiment, establishing a data lineage map based on the target report and reference table using a deep learning model includes:

[0064] The first step is to extract the indicators, indicator data, and relationships between indicators from the target report.

[0065] Optionally, financial report indicators include total assets, net profit, debt ratio, etc., representing the financial condition and operating results of a company or fund. Indicator data are the specific numerical values ​​of the indicators, used to reflect the company's financial status at a specific point in time. Relationships are the connections between indicators, including but not limited to correlation, proportional, and causal relationships. Deep learning models, through in-depth analysis of the target report, can identify all indicators in the target report, their specific values, and the complex relationships between them.

[0066] The second step is to extract the indicators, indicator data, and relationships between indicators from the reference table.

[0067] Alternatively, similar to the parsing of the target report, the deep learning model can identify the metrics, metric data, and relationships between them in the reference table.

[0068] The third step is to construct a data lineage graph by using the indicators in the target report and the reference table as nodes, and the relationships between the indicators in the target report and the relationships between the indicators in the reference table as edges. The node information in the data lineage graph includes the indicator data in the target report and the indicator data in the reference table.

[0069] Optionally, the deep learning model uses the indicators, indicator data and their relationships extracted from the target report and reference table as nodes and edges to construct a data lineage graph. The data lineage graph can intuitively display the flow and relationships of data and is a visual data structure for the deep learning model to perform intelligent auditing.

[0070] In summary, by extracting the indicators, indicator data, and their relationships from the target report and reference table through the above steps, a data lineage graph reflecting data correlation is constructed, providing a clear data structure and logical clues for intelligent auditing.

[0071] To improve the efficiency and accuracy of report review, optionally, in the report review method provided in this application embodiment, reviewing the target report based on the data lineage map includes:

[0072] The first step is to traverse the edges in the data lineage graph and determine whether each indicator data satisfies the preset rules.

[0073] Optionally, preset rules refer to predefined data review standards, including numerical ranges, correlation relationships, industry standards, regulatory requirements, etc., used to determine the compliance of indicator data. The deep learning model begins to traverse the edges in the data lineage graph, that is, the relationships between indicator data, and verifies each indicator data according to the preset rules.

[0074] The second step is to identify the indicator data in the target report that does not meet the preset rules as abnormal indicator data, and to determine the source of the abnormal indicator data by tracing back the edges in the data lineage graph.

[0075] Optionally, if the deep learning model discovers indicator data that does not meet the preset rules, it can further trace back the edges in the data lineage graph to find the source of the abnormal data, that is, the starting point of the abnormal data generation or propagation, providing clear clues for the correction of the problem.

[0076] In summary, by following the steps above and constructing a data lineage graph to review target reports, abnormal indicator data can be detected efficiently. Furthermore, by tracing back the edges in the graph, the source of these abnormal data can be accurately located, thus improving the efficiency of the review process.

[0077] To improve the accuracy of report review, optionally, in the report review method provided in this application embodiment, reviewing the target report based on the data lineage map includes:

[0078] The first step is to identify the common fields in the target report and the reference table, and then link the common fields in the target report and the reference table together.

[0079] Optionally, by identifying common fields or indicator names in the target report and the reference table, data associations between the target report and the reference table can be established through common fields.

[0080] The second step is to define the verification query statement based on the cross-reference relationship.

[0081] Optionally, reconciliation relationships refer to the mathematical logic and compliance requirements that should be followed between various indicators in financial data, such as sums, differences, and multiple relationships. Based on industry standards and compliance requirements for financial reporting, a series of reconciliation relationships can be predefined, and corresponding query statements can be generated accordingly to verify whether the logical relationships between specific data are valid in the data lineage graph.

[0082] The third step is to determine whether the indicator data in the target report meets the data logic based on the query statement and common fields.

[0083] Optionally, the constructed query statement can be used to perform data verification in the data lineage graph, comparing indicator data and checking whether the correlation relationships are valid.

[0084] The fourth step is to identify the indicator data in the target report that does not meet the data logic as abnormal indicator data, and to determine the source of the abnormal indicator data by tracing back the edges in the data lineage graph.

[0085] Optionally, when indicator data fails to pass data logic verification, it can be marked as abnormal indicator data, and the source of the abnormal data can be found by tracing the data flow path through the edge in the data lineage graph, that is, the field or calculation process where the abnormality first occurred.

[0086] In summary, by following the steps above and reviewing the target report using the defined query statements, the relevant fields in the target report and the reference table can be accurately compared, thus improving the accuracy of data comparison.

[0087] To improve the accuracy of report review, optionally, in the report review method provided in this application embodiment, after reviewing the target report based on the data lineage map and outputting the review result for the report to be reviewed, the method further includes:

[0088] The first step is to divide the target report into multiple paragraphs, perform position encoding on each paragraph, and generate multiple paragraph embedding vectors containing the position encoding.

[0089] Optionally, the target report can be segmented into multiple paragraphs. This breaks down long documents into smaller processing units, and positional encoding is performed on each paragraph to generate paragraph embedding vectors that contain positional information. Positional encoding enables deep learning models to understand the relative position of each paragraph within the original document when processing paragraph embedding vectors, which is crucial for contextual understanding.

[0090] The second step is to integrate the embedding vectors of multiple paragraphs into a single comprehensive vector representation.

[0091] Optionally, all paragraphs can be embedded into vectors and integrated into a single vector to form a comprehensive vector representation of the entire report, capturing global information from the entire document.

[0092] The third step is to review the comprehensive vector representation and identify any abnormal information in the target report.

[0093] Optionally, based on comprehensive vector representation, deep learning models can identify and label illogical or irregular anomalies in target reports, including format errors, unreasonable numerical values, and contradictory descriptions.

[0094] The fourth step is to add the abnormal information to the audit results to update the audit results.

[0095] Optionally, if new anomalies are discovered during the review of the comprehensive vector representation, they can be added to the previous review results to form an updated review report.

[0096] In summary, by taking the above steps, the target report is segmented into multiple paragraphs, paragraph embedding vectors containing positional encoding are generated, and then integrated into a comprehensive vector representation. This provides a review perspective for deep learning models that considers both data and context, significantly enhancing the depth, accuracy, and efficiency of the review process.

[0097] For example, Figure 3 The diagram illustrates a report review method according to this embodiment. First, the user uploads the annual report and reference table to be reviewed via a webpage, or via a scheduled task. Next, the system checks if the annual report is in doc format; if so, it converts it to docx format. The docx format annual report and reference table are then input into a large language model, which performs intelligent review. The model can also predefine rules to review the annual report, obtaining the review results, which include cell annotations and comments. Finally, the review results are output, and the user can view and download them.

[0098] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0099] Example 2

[0100] This application also provides a report review apparatus. It should be noted that the report review apparatus of this application can be used to execute the report review method provided in this application. The report review apparatus provided in this application is described below.

[0101] According to an embodiment of this application, an apparatus for implementing the above-described report review method is also provided, such as... Figure 4As shown, the device includes: a first determining unit 401, a conversion unit 402, and an auditing unit 403.

[0102] Specifically, the first determining unit 401 is used to determine the report to be audited and the reference table, wherein the reference table is the basis for judging whether there are abnormal indicator data in the report to be audited;

[0103] The conversion unit 402 is used to convert the report to be reviewed into a target report in a preset format.

[0104] The review unit 403 is used to input the target report and reference table into the deep learning model, build a data lineage map based on the target report and reference table through the deep learning model, review the target report based on the data lineage map, and output the review results for the report to be reviewed.

[0105] The report review device provided in this application embodiment determines the report to be reviewed and a reference table through a first determining unit 401, wherein the reference table is the basis for judging whether there are abnormal indicator data in the report to be reviewed; a conversion unit 402 converts the report to be reviewed into a target report in a preset format; and a review unit 403 inputs the target report and the reference table into a deep learning model, establishes a data lineage graph based on the target report and the reference table through the deep learning model, reviews the target report based on the data lineage graph, and outputs the review result for the report to be reviewed. This solves the technical problem of low review efficiency due to reliance on manual review of reports, and achieves the effect of improving the efficiency and accuracy of report review.

[0106] Optionally, in the report review apparatus provided in this application embodiment, the conversion unit includes: a reading module, used to traverse each cell of the table in the report to be reviewed and read the text content and attribute information in each cell; a skipping module, used to skip the merged cells that have already been read if the cells in the table in the report to be reviewed are merged cells; a reconstruction module, used to reconstruct the table structure of the table in the report to be reviewed based on the text content and attribute information of each cell to generate a target table; and a generation module, used to generate a target report based on the target table and the content in the report to be reviewed other than the table.

[0107] Optionally, in the report review device provided in this application embodiment, the review unit includes: a first determining module, used to determine the business logic for reviewing the target report, and to determine prompt words according to the business logic for reviewing the target report; and an input module, used to input the target report, reference table and prompt words into a deep learning model.

[0108] Optionally, in the report review device provided in this application embodiment, the review unit includes: a first extraction module, used to extract indicators, indicator data, and relationships between indicators in the target report; a second extraction module, used to extract indicators, indicator data, and relationships between indicators in a reference table; and a construction module, used to construct a data lineage graph by using the indicators in the target report and the indicators in the reference table as nodes, and the relationships between indicators in the target report and the relationships between indicators in the reference table as edges, wherein the node information in the data lineage graph includes indicator data in the target report and indicator data in the reference table.

[0109] Optionally, in the report review device provided in this application embodiment, the review unit includes: a first judgment module, used to traverse the edges in the data lineage graph and judge whether each indicator data meets the preset rules; and a backtracking module, used to determine the indicator data in the target report that does not meet the preset rules as abnormal indicator data, and determine the source of the abnormal indicator data by backtracking the edges in the data lineage graph.

[0110] Optionally, in the report review device provided in this application embodiment, the review unit includes: an association module, used to determine the common fields in the target report and the reference table, and to associate the common fields in the target report and the common fields in the reference table; a definition module, used to define a verification query statement based on the correlation relationship; a second judgment module, used to determine whether the indicator data in the target report meets the data logic based on the query statement and the common fields; and a second determination module, used to determine the indicator data in the target report that does not meet the data logic as abnormal indicator data, and to determine the source of the abnormal indicator data by tracing back the edges in the data lineage graph.

[0111] Optionally, in the report review apparatus provided in this application embodiment, the apparatus further includes: a segmentation unit, used to segment the target report into multiple paragraphs after reviewing the target report according to the data lineage map and outputting the review result for the report to be reviewed, and to perform position encoding on each paragraph to generate multiple paragraph embedding vectors containing position encoding; an integration unit, used to integrate the multiple paragraph embedding vectors into a comprehensive vector representation; a second determination unit, used to review the comprehensive vector representation and determine the abnormal information in the target report; and an update unit, used to add the abnormal information to the review result to update the review result.

[0112] It should be noted that the first determining unit 401, the conversion unit 402, and the review unit 403 mentioned above correspond to steps S201 to S203 in Embodiment 1. The three units and the corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules or units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above modules can also be part of a device and run in the computer terminal 10 provided in Embodiment 1.

[0113] Example 3

[0114] Embodiments of this application may provide a computer terminal, which may be any computer terminal device in a group of computer terminals. Optionally, in this embodiment, the aforementioned computer terminal may also be replaced with a mobile terminal or an electronic device, etc.

[0115] Optionally, in this embodiment, the computer terminal may be located in at least one of a plurality of network devices in a computer network.

[0116] In this embodiment, the computer terminal described above can execute the program code for the following steps in the report review method: determining the report to be reviewed and a reference table, wherein the reference table is the basis for judging whether there is abnormal indicator data in the report to be reviewed; converting the report to be reviewed into a target report in a preset format; inputting the target report and the reference table into a deep learning model, establishing a data lineage graph based on the target report and the reference table through the deep learning model, reviewing the target report based on the data lineage graph, and outputting the review result for the report to be reviewed.

[0117] Optionally, the aforementioned computer terminal can execute the program code for the following steps in the report review method: converting the report to be reviewed into a target report in a preset format includes: traversing each cell of the table in the report to be reviewed and reading the text content and attribute information in each cell; skipping the already read merged cells if they are already read; reconstructing the table structure of the table in the report to be reviewed based on the text content and attribute information of each cell to generate a target table; and generating a target report based on the target table and the content in the report to be reviewed other than the table.

[0118] Optionally, the computer terminal described above can execute program code for the following steps in the report review method: inputting the target report and reference table into the deep learning model, including: determining the business logic for reviewing the target report, and determining prompt words based on the business logic for reviewing the target report; inputting the target report, reference table, and prompt words into the deep learning model.

[0119] Optionally, the aforementioned computer terminal can execute the program code for the following steps in the report review method: establishing a data lineage graph based on the target report and reference table using a deep learning model, including: extracting indicators, indicator data, and relationships between indicators from the target report; extracting indicators, indicator data, and relationships between indicators from the reference table; constructing a data lineage graph by using the indicators in the target report and the indicators in the reference table as nodes, and the relationships between indicators in the target report and the relationships between indicators in the reference table as edges, wherein the node information in the data lineage graph includes indicator data from the target report and indicator data from the reference table.

[0120] Optionally, the computer terminal described above can execute the program code for the following steps in the report review method: reviewing the target report based on the data lineage graph includes: traversing the edges in the data lineage graph and determining whether each indicator data meets the preset rules; identifying the indicator data in the target report that does not meet the preset rules as abnormal indicator data, and determining the source of the abnormal indicator data by tracing back the edges in the data lineage graph.

[0121] Optionally, the aforementioned computer terminal can execute program code for the following steps in the report review method: reviewing the target report based on the data lineage graph, including: identifying common fields in the target report and the reference table, and associating the common fields in the target report and the reference table; defining a verification query statement based on the correlation relationship; determining whether the indicator data in the target report meets the data logic based on the query statement and the common fields; identifying indicator data in the target report that does not meet the data logic as abnormal indicator data, and determining the source of the abnormal indicator data by tracing back the edges in the data lineage graph.

[0122] Optionally, the aforementioned computer terminal can execute the program code for the following steps in the report review method: after reviewing the target report based on the data lineage map and outputting the review result for the report to be reviewed, the method further includes: dividing the target report into multiple paragraphs, performing position encoding on each paragraph, and generating multiple paragraph embedding vectors containing position encodings; integrating the multiple paragraph embedding vectors into a comprehensive vector representation; reviewing the comprehensive vector representation to determine abnormal information in the target report; and adding the abnormal information to the review result to update the review result.

[0123] Optionally, Figure 5This is a structural block diagram of an electronic device according to an embodiment of this application. Figure 5 As shown, the electronic device may include: one or more ( Figure 5 (Only one is shown) processor 502, memory 504, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0124] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the report review method and apparatus in this embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the aforementioned report review method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0125] The processor can access the information and application programs stored in the memory via the transmission device to perform the steps described above in the report review method.

[0126] This application provides a scheme for reviewing reports. By determining the report to be reviewed and a reference table (where the reference table serves as the basis for judging whether the report to be reviewed contains abnormal indicator data), the report to be reviewed is converted into a target report in a preset format. The target report and the reference table are input into a deep learning model, which then establishes a data lineage graph based on the target report and the reference table. The target report is reviewed based on the data lineage graph, and the review results for the report to be reviewed are output. This solves the technical problem of low efficiency in relying on manual report review, and achieves the technical effect of improving the efficiency and accuracy of report review.

[0127] Those skilled in the art will understand that Figure 5 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 5 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 5 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 5 The different configurations shown.

[0128] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0129] Example 4

[0130] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the report review method provided in Embodiment 1.

[0131] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0132] This application also provides a computer program product that, when executed on a data processing device, is adapted to perform the steps of a report auditing method.

[0133] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0134] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0135] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0136] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0137] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0138] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0139] The above are merely preferred embodiments of this application. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for reviewing reports, characterized in that, include: Identify the report to be reviewed and the reference table, wherein the reference table is the basis for determining whether there are abnormal indicator data in the report to be reviewed; Convert the report to be reviewed into a target report in a preset format; The target report and the reference table are input into a deep learning model. The deep learning model builds a data lineage graph based on the target report and the reference table. The target report is reviewed based on the data lineage graph, and the review result for the report to be reviewed is output.

2. The method according to claim 1, characterized in that, Converting the report to be reviewed into a target report in a preset format includes: Iterate through each cell of the table in the report to be reviewed, and read the text content and attribute information of each cell; If a cell in the table of the report to be reviewed is a merged cell and has already been read, skip the merged cell that has already been read; Based on the text content and attribute information of each cell, the table structure of the table in the report to be reviewed is reconstructed to generate the target table; The target report is generated based on the target table and the contents of the report to be reviewed, excluding the table.

3. The method according to claim 1, characterized in that, Inputting the target report and the reference table into the deep learning model includes: Determine the business logic used to review the target report, and determine prompt words based on the business logic for reviewing the target report; The target report, the reference table, and the prompt words are input into the deep learning model.

4. The method according to claim 1, characterized in that, The data lineage map is constructed using the deep learning model based on the target report and the reference table, including: Extract the indicators, indicator data, and relationships between indicators from the target report; Extract the indicators, indicator data, and relationships between the indicators from the reference table; The data lineage graph is constructed by using the indicators in the target report and the indicators in the reference table as nodes, and the relationships between the indicators in the target report and the relationships between the indicators in the reference table as edges. The node information in the data lineage graph includes the indicator data in the target report and the indicator data in the reference table.

5. The method according to claim 4, characterized in that, The review of the target report based on the aforementioned kinship data includes: Traverse the edges in the data lineage graph and determine whether each indicator data satisfies the preset rules; The indicator data in the target report that do not meet the preset rules are identified as abnormal indicator data, and the source of the abnormal indicator data is determined by tracing back the edges in the data lineage graph.

6. The method according to claim 4, characterized in that, The review of the target report based on the aforementioned kinship data includes: Identify the common fields in the target report and the reference table, and perform data association between the common fields in the target report and the common fields in the reference table; Query statements for verification based on cross-reference relationships; Based on the query statement and the common fields, determine whether the indicator data in the target report meets the data logic; The indicator data in the target report that does not meet the data logic is identified as abnormal indicator data, and the source of the abnormal indicator data is determined by tracing back the edges in the data lineage graph.

7. The method according to claim 1, characterized in that, After reviewing the target report based on the pedigree data and outputting the review result for the report to be reviewed, the method further includes: The target report is divided into multiple paragraphs, and each paragraph is position-encoded to generate multiple paragraph embedding vectors containing the position codes. The multiple paragraph embedding vectors are integrated into a single comprehensive vector representation; The comprehensive vector representation is reviewed to identify any abnormal information in the target report; The abnormal information is added to the audit result to update the audit result.

8. A report review device, characterized in that, include: The first determining unit is used to determine the report to be reviewed and the reference table, wherein the reference table is the basis for determining whether there are abnormal indicator data in the report to be reviewed; A conversion unit is used to convert the report to be reviewed into a target report in a preset format; The review unit is used to input the target report and the reference table into a deep learning model, establish a data lineage graph based on the target report and the reference table through the deep learning model, review the target report based on the data lineage graph, and output the review result for the report to be reviewed.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the computer-readable storage medium is located to perform the report review method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the report review method according to any one of claims 1 to 7.

11. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the report review method according to any one of claims 1 to 7.

Images