Method and system for identifying and tracing false enterprise information based on structured field propagation chain

Abstract

The application provides a false enterprise information identification and tracing method and system based on a structured field propagation chain, comprising: constructing a field propagation chain graph based on the relationship between fields in an enterprise database; calculating the change trend of the field value and the deviation relative to the industry benchmark or historical benchmark for the field value of the target field in the field propagation chain graph in different time periods to obtain a field value evolution score; obtaining a plurality of related fields associated with the target field based on the field propagation chain graph, and generating a rule gate decision result for the target field through a configured structured rule engine; and performing multi-factor fusion calculation based on the field value evolution score, field confidence and rule gate decision result of the target field to obtain a false information identification result of the target field and generate a field-level evidence chain corresponding to the identification result.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and natural language processing, and more specifically, to a method and system for identifying and tracing fake enterprise information based on a structured field propagation chain. Background Technology

[0002] With the rapid development of the market economy, corporate information serves as a core basis for market supervision, credit approval, and business cooperation, making its authenticity and reliability crucial. However, current corporate information is characterized by its multi-source heterogeneity, frequent updates, and complex dissemination paths. Some companies, seeking illicit gains, submit false information by altering registered capital, falsifying operating data, and fabricating equity structures. This not only disrupts market order but also causes significant losses to regulatory authorities, financial institutions, and other relevant parties.

[0003] Existing technologies for identifying fraudulent enterprise information largely rely on single-field verification, manual review, or simple rule matching, which have significant limitations. Firstly, they struggle to capture potential relationships and propagation patterns between fields, failing to effectively identify hidden fraudulent information where "a single field is compliant, but multiple fields are abnormal in combination." Secondly, they lack dynamic analysis of field value evolution, making it difficult to distinguish between reasonable data fluctuations and malicious tampering. Furthermore, after identifying fraudulent information, they cannot accurately trace its source and scope of influence, hindering efficient regulatory and accountability efforts. Therefore, there is an urgent need for a technical solution that can achieve accurate identification and tracing based on field relationships and dynamic evolution characteristics to overcome the shortcomings of existing technologies.

[0004] Patent document CN120217163A (application number: 202510585927.4) discloses a method and apparatus for detecting misinformation based on a cross-social media propagation network. The method includes: using news data as the initial key nodes of the propagation structure in the network graph; setting a learnable node selector for the network graph, introducing a multi-expert model framework, where each expert model obtains a key propagation subgraph based on the importance weights of the nodes output by the node selector; using an aggregation network to aggregate the nodes of the key propagation subgraph; calculating the mutual information loss of the key propagation subgraph based on graph information bottleneck theory; simultaneously introducing a gating network to assign weights to the outputs of each expert model and performing misinformation detection and classification after aggregating the node aggregation results; constructing a classification loss, and also introducing comprehensiveness loss and diversity loss for the multi-expert model framework, combined with the mutual information loss to jointly optimize the node selector, gating network, and aggregation network, thereby improving the accuracy and effectiveness of misinformation detection. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for identifying and tracing fake enterprise information based on a structured field propagation chain.

[0006] A method for identifying and tracing fake enterprise information based on a structured field propagation chain, provided by the present invention, includes:

[0007] Step S1: Construct a field propagation chain graph based on the relationships between fields in the enterprise database;

[0008] Step S2: For the field values ​​of the target field in the field propagation chain graph at different time periods, calculate the changing trend of the field values ​​and the deviation relative to the industry benchmark or historical benchmark, and calculate the field value evolution score based on the calculated changing trend and the deviation relative to the benchmark.

[0009] Step S3: Based on the field propagation chain graph, obtain multiple related fields that have a propagation path association with the target field, and perform joint logical judgment on the field values ​​corresponding to the multiple related fields through the configured structured rule engine to generate a rule gating judgment result for the target field;

[0010] Step S4: Perform multi-factor fusion calculation based on the field value evolution score, field confidence, and rule gating judgment result of the target field to obtain the false information identification result of the target field, and generate a field-level evidence chain corresponding to the identification result.

[0011] Preferably, step S1 includes:

[0012] Step S1.1: Obtain the enterprise field set based on multiple heterogeneous databases of the enterprise, and convert the original fields into standard fields using a preset field semantic mapping table;

[0013] Step S1.2: Select the most reliable field value from the standard field set based on time priority and weight;

[0014] Step S1.3: Normalize the selected field values ​​of the converted standard fields to obtain normalized field values;

[0015] Step S1.4: Construct an enterprise database based on standard fields and their normalized field values;

[0016] Step S1.5: Construct a directional field propagation chain graph based on the computational dependencies, logical references, or semantic dependencies between fields in the enterprise database.

[0017] Preferably, the nodes of the field propagation chain graph represent standard fields; the edges of the field propagation chain graph represent the relationships between fields; and the edge weights of the field propagation chain graph represent the value propagation sensitivity or derivation relationship between fields.

[0018] The directional field propagation chain graph is used to represent the causal / dependency direction and supports truncation settings for path length to limit propagation depth.

[0019] Preferably, step S2 includes:

[0020] Step S2.1: For the field values ​​of the same target field in different time periods, calculate its changing trend in different time periods based on time series analysis, and calculate its relative benchmark deviation based on industry / historical benchmarks;

[0021] Step S2.2: Based on the calculated trend and relative benchmark deviation, a linear weighted average is used to generate field value evolution scores;

[0022] Specifically, the step of calculating the changing trend of the same target field in different time periods based on time series analysis includes: dynamically calculating the year-on-year / month-on-month changing trend of the field using a time window sliding mechanism; wherein the changing trend includes: the magnitude of change and volatility.

[0023] Preferably, step S3 includes:

[0024] Step S3.1: Based on the field propagation chain graph, obtain multiple related fields that have a propagation path association with the target field;

[0025] Step S3.2: Configure the structured rule engine according to preset expert rules or preset regulatory logic;

[0026] Step S3.3: Through the configured structured rule engine, including using Boolean logic gates to jointly judge the field values ​​corresponding to multiple related fields, the system identifies whether the path connecting multiple fields is a risk path as the rule gating judgment result of the target field.

[0027] A system for identifying and tracing fake enterprise information based on a structured field propagation chain, provided by the present invention, includes:

[0028] Module M1: Constructs a field propagation chain graph based on the relationships between fields in the enterprise database;

[0029] Module M2: For the field values ​​of the target field in the field propagation chain graph at different time periods, calculate the changing trend of the field values ​​and the deviation relative to the industry benchmark or historical benchmark, and calculate the field value evolution score based on the calculated changing trend and the deviation relative to the benchmark.

[0030] Module M3: Based on the field propagation chain graph, obtain multiple related fields that have a propagation path associated with the target field, and perform joint logical judgment on the field values ​​corresponding to the multiple related fields through the configured structured rule engine to generate a rule gating judgment result for the target field;

[0031] Module M4: Based on the field value evolution score, field confidence and rule gating judgment result of the target field, perform multi-factor fusion calculation to obtain the false information identification result of the target field, and generate a field-level evidence chain corresponding to the identification result.

[0032] Preferably, the module M1 includes:

[0033] Module M1.1: Obtains enterprise field sets based on multiple heterogeneous databases, and converts the original fields into standard fields using a preset field semantic mapping table;

[0034] Module M1.2: Selects the most reliable field value from a standard field set based on time priority and weight;

[0035] Module M1.3: Normalizes the selected field values ​​of the converted standard fields to obtain normalized field values;

[0036] Module M1.4: Constructs an enterprise database based on standard fields and their normalized values;

[0037] Module M1.5: Constructs a directional field propagation chain graph based on the computational dependencies, logical references, or semantic dependencies between fields in the enterprise database.

[0038] Preferably, the nodes of the field propagation chain graph represent standard fields; the edges of the field propagation chain graph represent the relationships between fields; and the edge weights of the field propagation chain graph represent the value propagation sensitivity or derivation relationship between fields.

[0039] The directional field propagation chain graph is used to represent the causal / dependency direction and supports truncation settings for path length to limit propagation depth.

[0040] Preferably, the module M2 includes:

[0041] Module M2.1: For the field values ​​of the same target field in different time periods, calculate the changing trend of the field in different time periods based on time series analysis, and calculate the relative benchmark deviation based on industry / historical benchmarks;

[0042] Module M2.2: Based on the calculated trend of change and the relative deviation from the benchmark, a linear weighted average is used to generate field value evolution scores;

[0043] Specifically, the step of calculating the changing trend of the same target field in different time periods based on time series analysis includes: dynamically calculating the year-on-year / month-on-month changing trend of the field using a time window sliding mechanism; wherein the changing trend includes: the magnitude of change and volatility.

[0044] Preferably, the module M3 includes:

[0045] Module M3.1: Based on the field propagation chain graph, obtain multiple related fields that have a propagation path association with the target field;

[0046] Module M3.2: A structured rule engine configured based on preset expert rules or preset regulatory logic;

[0047] Module M3.3: Through the configured structured rule engine, including the use of Boolean logic gates to jointly judge the field values ​​corresponding to multiple related fields, the module identifies whether the path connecting multiple fields is a risky path as the rule gating judgment result of the target field.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. This invention focuses on the dynamic evolution and semantic dependency path construction between fields in a structured enterprise database. It uses field semantic mapping, time series trajectory and rule logic gating mechanism to analyze the abnormal evolution behavior of key field values ​​within the enterprise, constructs a field-level propagation chain graph, and generates multi-factor false identification conclusions by combining confidence.

[0050] 2. This invention is applicable to scenarios such as financial fraud identification, corporate information disclosure review, and automatic risk control assessment, and has higher interpretability and auditability. Attached Figure Description

[0051] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0052] Figure 1 This is an architecture diagram of a fake information identification system based on a dynamic propagation chain.

[0053] Figure 2 A flowchart for constructing the propagation chain and mining the node behavior graph.

[0054] Figure 3 This is a diagram showing the internal structure of the multimodal feature extraction module.

[0055] Figure 4 This is a structural diagram of the rule gating and reasoning module.

[0056] Figure 5Visualize the structure of the evidence chain for each field.

[0057] Figure 6 This is a flowchart of a multimodal data fusion and recognition model. Detailed Implementation

[0058] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0059] Example 1

[0060] A method for identifying and tracing fake enterprise information based on a structured field propagation chain, provided by the present invention, includes:

[0061] Step S1: Construct a field propagation chain graph based on the relationships between fields in the enterprise database;

[0062] Specifically, step S1 includes:

[0063] Step S1.1: Obtain the enterprise field set based on multiple heterogeneous databases of the enterprise, and convert the original fields into standard fields using a preset field semantic mapping table;

[0064] Step S1.2: Select the most reliable field value from the standard field set based on time priority and weight;

[0065] Step S1.3: Normalize the selected field values ​​of the converted standard fields to obtain normalized field values;

[0066] Step S1.4: Construct an enterprise database based on standard fields and their normalized field values;

[0067] Step S1.5: Construct a directional field propagation chain graph based on the computational dependencies, logical references, or semantic dependencies between fields in the enterprise database.

[0068] The nodes of the field propagation chain graph represent standard fields; the edges of the field propagation chain graph represent the relationships between fields; the edge weights of the field propagation chain graph represent the value transmission sensitivity or derivation relationship between fields; the edge weights are constructed based on historical value cooperation, computational logic chains, or semantic dependencies. During field propagation chain analysis, the influence of field anomaly evolution scores on the propagation path is weighted and adjusted based on the edge weights of the field propagation chain graph to reflect the impact of different field dependency strengths on the anomaly propagation results.

[0069] The directional field propagation chain graph is used to represent the causal / dependency direction and supports truncation settings for path length to limit propagation depth.

[0070] This embodiment achieves traceable and causal analysis of enterprise information anomalies through a structured field propagation chain graph; interpretability and causal consistency are significantly enhanced, enabling regulatory and auditing scenarios to directly see the abnormal causal chain.

[0071] Step S2: For the field values ​​of the target field in the field propagation chain graph at different time periods, calculate the changing trend of the field values ​​and the deviation relative to the industry benchmark or historical benchmark, and calculate the field value evolution score based on the calculated changing trend and the deviation relative to the benchmark.

[0072] Specifically, step S2 includes:

[0073] Step S2.1: For the field values ​​of the same target field in different time periods, calculate its changing trend in different time periods based on time series analysis, and calculate its relative benchmark deviation based on industry / historical benchmarks;

[0074] Step S2.2: Based on the calculated trend and relative benchmark deviation, a linear weighted average is used to generate field value evolution scores;

[0075] Specifically, the step of calculating the changing trend of the same target field in different time periods based on time series analysis includes: dynamically calculating the year-on-year / month-on-month changing trend of the field using a time window sliding mechanism; wherein the changing trend includes: the magnitude of change and volatility.

[0076] Step S3: Based on the field propagation chain graph, obtain multiple related fields that have a propagation path association with the target field, and use the configured structured rule engine to perform joint logical judgment on the field values ​​corresponding to the multiple related fields to generate a rule gating judgment result for the target field;

[0077] Specifically, step S3 includes:

[0078] Step S3.1: Based on the field propagation chain graph, obtain multiple related fields that have a propagation path association with the target field;

[0079] Step S3.2: Configure the structured rule engine according to preset expert rules or preset regulatory logic;

[0080] Step S3.3: Using the configured structured rule engine, including the use of Boolean logic gates to jointly judge the field values ​​corresponding to multiple related fields, the system identifies whether the path connecting multiple fields is a risky path, which is then used as the rule gating judgment result for the target field. The Boolean logic gates include: AND, OR, and XOR.

[0081] In this embodiment, for example, if the net cash flow is less than 0 and the accounts receivable ratio is greater than 35%, the "potential financial risk" node is activated.

[0082] Step S4: Based on the field value evolution score, field confidence score, and rule gating judgment result of the target field, perform multi-factor fusion calculation to obtain the false information identification result of the target field, and generate a field-level evidence chain corresponding to the identification result. The field confidence score is determined by multiplying the source credibility and field importance weight; the source credibility is allocated according to the data source level; the field importance is obtained based on domain knowledge or historical samples; the false information identification result of the target field contains a structured JSON format, including at least a conclusion label, trigger rule number, key field path, and confidence score; the field-level evidence chain is used to generate a graphical link diagram, displaying the source table, field name, timestamp, and value of each field.

[0083] In this embodiment, when the "potential financial risk" node is activated, the triggered field link is labeled with a risk label; the false information identification result of the target field is obtained by integrating the field evolution score, field credibility, and rule weight calculation path confidence, and the node field, value, timestamp and triggering reason are recorded.

[0084] In this embodiment, the method further includes: setting a field confidence threshold, and suspending the corresponding link propagation calculation when the confidence of any field is greater than the field confidence threshold, for example, when the source data is too old or the source is unreliable.

[0085] The present invention also provides a system for identifying and tracing fake enterprise information based on a structured field propagation chain. The system for identifying and tracing fake enterprise information based on a structured field propagation chain can be implemented by executing the process steps of the method for identifying and tracing fake enterprise information based on a structured field propagation chain. That is, those skilled in the art can understand the method for identifying and tracing fake enterprise information based on a structured field propagation chain as a preferred embodiment of the system for identifying and tracing fake enterprise information based on a structured field propagation chain.

[0086] This invention employs a series of technical features, including structured field propagation chain graph construction, dynamic evolution scoring, Boolean rule gating mechanism, and field-level evidence chain generation, to achieve causal identification, path-level reasoning, and interpretable output for enterprise information anomalies. Compared with existing schemes based on user behavior graphs or single-field threshold judgments, this invention significantly improves the accuracy, traceability, and system robustness of anomaly identification, and can provide unified and auditable false information identification results in multi-source heterogeneous data environments.

[0087] Example 2

[0088] Example 2 is a preferred example of Example 1.

[0089] According to the present invention, a method for identifying and tracing fake enterprise information based on a structured field propagation chain is provided, such as... Figure 1-2 As shown, it includes:

[0090] Acquire multimodal input data, including: text, images, and tabular information from multiple sources;

[0091] Structured field extraction is performed based on the acquired multimodal input data, including: using a multimodal parsing model that combines text and image models to extract fields from the multimodal input data; wherein, the text model can be any pre-trained language model, including BERT, RoBERTa or equivalent models in this field, and extracts structured field names and corresponding data through named entity recognition, keyword localization and dependency parsing, etc.

[0092] For scanned documents or PDFs with layout information, OCR and layout analysis can be used to extract the two-dimensional structure; this embodiment allows the use of a CNN+Transformer hybrid model or other equivalent architectures.

[0093] The extracted structured fields are standardized to unify field format and semantic tags;

[0094] By utilizing semantics, source, and data pattern recognition to identify the relationships between fields, a field propagation chain graph is constructed based on the standardized fields and their corresponding values. A time dimension and upstream and downstream propagation path information are added to the field propagation chain graph.

[0095] like Figure 3 As shown, based on the target field in the constructed field propagation chain graph, a score is assigned according to the field change behavior. Specifically, the changes in field values ​​at different time points are analyzed, including: anomaly identification and consistency verification. Anomaly identification includes detecting sudden changes, trend drift, and outliers in field values. Consistency verification includes: comparison with external trusted sources (such as official website data) for confirmation. Based on anomaly identification and consistency verification, a scoring index is generated to determine whether the data is false.

[0096] like Figure 4 As shown, based on the field propagation chain graph, multiple related fields that have a propagation path associated with the target field are obtained, and the field values ​​corresponding to the multiple related fields are jointly judged by the configured structured rule engine to generate a rule gating judgment result for the target field;

[0097] Specifically, load preset rules from the rule base, such as AND and OR; determine whether the field value satisfies the logical gate condition; if it satisfies the rule, mark the field as an exception chain; otherwise, maintain the original state.

[0098] Based on the field value evolution score, field confidence, and rule gating judgment result of the target field, a multi-factor fusion calculation is performed to obtain the false information identification result of the target field, and a field-level evidence chain corresponding to the identification result is generated.

[0099] like Figure 5 As shown, this demonstrates how multiple source fields, after propagation path analysis, form risk paths pointing to different target fields, and the final identification result is output through the evidence chain scoring module. Source Field 1, Source Field 2, and Source Field 3 represent the field data captured in the initial stage of information propagation; intermediate nodes (propagation paths) represent the evolutionary relationship, transformation logic, or propagation trajectory between fields; target field 1 (suspicious) / target field 2 (normal) represent the field nodes identified as having or not having risk after rule gating and scoring; the evidence chain scoring module weights and fuses the propagation chain structure and scoring factors of the target field path to output the risk confidence level; the display module visually presents the suspicious field link and risk level to users or regulators.

[0100] like Figure 6 The diagram illustrates the entire process of identifying misinformation using a multimodal feature fusion and inference model, encompassing structured field data, unstructured text data, and propagation path information. Specifically, it includes: structured field data (such as field values, numerical anomalies, and source credibility); unstructured text data (including news texts, comments, and forum discussions); a propagation path graph (extracting multi-hop propagation structures through propagation chain mining); a multimodal feature extraction module (encoding features from different data types to form a unified vector); a fusion encoding module (aligning and fusing multi-source information to generate a unified representation); a time-series inference module (introducing time series modeling to identify false signals in the time-series evolution); and the final risk prediction result and its confidence level.

[0101] Example 3

[0102] Example 3 is a preferred example of Example 1.

[0103] This embodiment is applied to the scenario of verifying false information about enterprises in the credit approval process of commercial banks. It focuses on identifying false information in core fields such as registered capital, operating income, debt-to-asset ratio, and tax payment for SMEs applying for operating loans. The data sources involved include four types of heterogeneous databases: the commercial bank's internal credit system, the market supervision department's business registration database, the tax department's electronic declaration database, and the People's Bank of China's credit reporting system. These databases cover dynamic data of enterprises over the past three years, achieving fully automated processing of the entire process from identification to tracing and evidence consolidation, meeting the compliance and efficiency requirements of credit approval.

[0104] In this embodiment, the core standard fields for filtering cover basic enterprise information, including: registered capital, establishment time, equity ratio, etc.; financial data, including: operating revenue, asset-liability ratio, accounts receivable, net cash flow, etc.; and compliance information, including: tax amount, number of social security contributors, and administrative license status, etc.

[0105] The time period is set to quarterly, the data traceability period is 3 years, there are 12 time nodes, the time window sliding step is 1 quarter, and year-on-year and month-on-month calculations are supported.

[0106] The field propagation chain graph includes:

[0107] Nodes consist of multiple standard fields, and edges represent the dependencies between these fields, including computational dependencies (e.g., tax amount = operating revenue × VAT rate), logical dependencies (e.g., debt-to-asset ratio = total liabilities / total assets), and semantic dependencies (e.g., registered capital is positively correlated with the number of people paying social security).

[0108] Edge weight calculation: Based on the Pearson correlation coefficient and the co-operation of historical data, the edge weight calculation ranges are [0.85, 0.95] for dependence, [0.7, 0.85] for logical dependence, and [0.4, 0.65] for semantic dependence. For example, the edge weight of operating revenue - tax amount is 0.92, and the edge weight of registered capital - number of social security contributors is 0.53.

[0109] Propagation depth: The path length truncation threshold is set to 3, meaning that only related fields within 3 hops of the target field are analyzed.

[0110] This embodiment provides a method for identifying and tracing fake enterprise information based on a structured field propagation chain, including:

[0111] Step 1: Construct a field propagation chain graph based on the relationships between fields in the enterprise database;

[0112] In this embodiment, raw data from 1,000 loan applicant companies were collected from four types of heterogeneous databases, resulting in 78 raw fields. The fields were standardized and converted using a preset field semantic mapping table. For example, “registered capital” and “subscribed capital” were uniformly mapped to “registered capital”, and “annual tax payment” and “annual total tax payment” were uniformly mapped to “tax payment”.

[0113] Based on time priority, data updated within the last 6 months is selected and weighted, and the field value credibility is combined with the data update time decay coefficient to filter the credibility value of each standard field.

[0114] Normalization is performed on the selected confidence values;

[0115] A directed propagation chain graph is constructed based on the dependencies between fields. An example of a propagation chain with "operating revenue" as the core node is as follows: Operating revenue (node ​​A) → Tax amount (node ​​B, edge weight 0.92) → Net cash flow (node ​​C, edge weight 0.81); Operating revenue (node ​​A) → Accounts receivable (node ​​D, edge weight 0.78) → Allowance for doubtful accounts (node ​​E, edge weight 0.89).

[0116] Step 2: For the field values ​​of the target field in the field propagation chain graph at different time periods, calculate the changing trend of the field values ​​and the deviation relative to the industry benchmark or historical benchmark. Based on the calculated changing trend and the deviation relative to the benchmark, calculate the field value evolution score.

[0117] In this embodiment, the target field "Operating Revenue" (Company Code: A2024001, Industry: General Equipment Manufacturing) is used as an example:

[0118] Time series data extraction: The normalized values ​​of the company's operating revenue for 12 time periods from Q1 2021 to Q4 2023 were extracted: [0.32, 0.35, 0.38, 0.42, 0.45, 0.49, 0.51, 0.53, 0.56, 0.58, 0.62, 0.85].

[0119] Trend calculation: The year-on-year / quarter-on-quarter change magnitude and volatility are calculated using a time window sliding mechanism. The quarter-on-quarter change magnitude in Q4 2023 is approximately 37.1% (0.85-0.62) / 0.62, and the year-on-year change magnitude is approximately 102.4% (0.85-0.42) / 0.42. The volatility is 0.18 obtained by calculating the standard deviation of data over 12 time periods.

[0120] Calculation of relative benchmark deviation: The industry benchmark range for the year-on-year growth rate of operating revenue in the general equipment manufacturing industry is [8%, 25%], and the historical benchmark is 15%; the relative industry benchmark deviation = (102.4% - 25%) / 25% ≈ 309.6%, and the relative historical benchmark deviation = (102.4% - 15%) / 15% ≈ 582.7%.

[0121] Evolutionary score generation: The linear weighted formula S=α×St+β×Sb is used, where α=0.4 (trend weight) and β=0.6 (deviation weight); St (trend score) is calculated as 89 points based on the magnitude and volatility of change, and Sb (deviation score) is calculated as 94 points based on the deviation from the benchmark. The final evolutionary score is 0.4×89+0.6×94=92.2 points.

[0122] Step 3: Based on the field propagation chain graph, obtain multiple related fields that have a propagation path associated with the target field, and use the configured structured rule engine to perform joint logical judgment on the field values ​​corresponding to the multiple related fields to generate a rule gating judgment result for the target field;

[0123] In this embodiment, relevant fields are obtained as follows: Based on the propagation chain graph, relevant fields within 3 hops of operating revenue are obtained, such as tax amount, accounts receivable, bad debt provision, and value-added tax rate.

[0124] Rule matching: The structured rule engine is invoked to perform joint logical judgment on the relevant field values. The company's tax payment growth rate in Q4 2023 was 8%, and the VAT rate was 13% (legal range). This meets all the conditions of rule R01: revenue growth rate > 50% ∧ tax payment growth rate < 10% ∧ VAT rate within the legal range. The rule gating judgment result is that rule R01 is triggered, and it is marked as a path of revenue overstatement risk.

[0125] Step 4: Perform multi-factor fusion calculation based on the field value evolution score, field confidence, and rule gating judgment result of the target field to obtain the false information identification result of the target field, and generate a field-level evidence chain corresponding to the identification result.

[0126] In this embodiment, the field confidence score is calculated as follows: the operating revenue data comes from the tax database, the data source weight is 0.95, the field importance weight is 0.95, the data update time is January 2024, the time decay coefficient λ=0.98, and the field confidence score is approximately 0.95×0.95×0.98≈0.886.

[0127] Multi-factor fusion calculation: The weighted summation formula F=w1×S+w2×C+w3×R is used, where w1=0.4 (evolutionary score weight), w2=0.3 (confidence weight), w3=0.3 (rule weight); S=92.2 (evolutionary score), C=0.886×100=88.6 (confidence converted to percentage), R=0.9×100=90 (rule weight converted to percentage). The final risk score is 0.4×92.2+0.3×88.6+0.3×90=90.46. A risk threshold of 75 points is set, and this field is judged as high-risk for false information.

[0128] Finally, the risk propagation path is presented in the form of a directed graph, with nodes labeled with field names, confidence levels, and anomaly markers. Weights and dependency types are assigned to each node, and the causal link of operating revenue (high risk) → tax amount (abnormal association) → net cash flow (association impact) is highlighted. A time-series change curve of field values ​​and a comparison chart of the baseline interval are also attached.

[0129] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0130] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for identifying and tracing fake enterprise information based on a structured field propagation chain, characterized in that, include: Step S1: Construct a field propagation chain graph based on the relationships between fields in the enterprise database; Step S2: For the field values ​​of the target field in the field propagation chain graph at different time periods, calculate the changing trend of the field values ​​and the deviation relative to the industry benchmark or historical benchmark, and calculate the field value evolution score based on the calculated changing trend and the deviation relative to the benchmark. Step S3: Based on the field propagation chain graph, obtain multiple related fields that have a propagation path association with the target field, and use the configured structured rule engine to perform joint logical judgment on the field values ​​corresponding to the multiple related fields to generate a rule gating judgment result for the target field; Step S4: Perform multi-factor fusion calculation based on the field value evolution score, field confidence, and rule gating judgment result of the target field to obtain the false information identification result of the target field, and generate a field-level evidence chain corresponding to the identification result.

2. The method for identifying and tracing fake enterprise information based on structured field propagation chains according to claim 1, characterized in that, Step S1 includes: Step S1.1: Obtain the enterprise field set based on multiple heterogeneous databases of the enterprise, and convert the original fields into standard fields using a preset field semantic mapping table; Step S1.2: Select reliable field values ​​that meet preset requirements based on time priority and weight for the standard field set; Step S1.3: Normalize the selected field values ​​of the converted standard fields to obtain normalized field values; Step S1.4: Construct an enterprise database based on standard fields and their normalized field values; Step S1.5: Construct a directional field propagation chain graph based on the computational dependencies, logical references, or semantic dependencies between fields in the enterprise database.

3. The method for identifying and tracing fake enterprise information based on structured field propagation chains according to claim 2, characterized in that, The nodes of the field propagation chain graph represent standard fields; the edges of the field propagation chain graph represent the relationships between fields; and the edge weights of the field propagation chain graph represent the value propagation sensitivity or derivation relationship between fields. The directional field propagation chain graph is used to represent the causal / dependency direction and supports truncation settings for path length to limit propagation depth.

4. The method for identifying and tracing fake enterprise information based on structured field propagation chains according to claim 1, characterized in that, Step S2 includes: Step S2.1: For the field values ​​of the same target field in different time periods, calculate its changing trend in different time periods based on time series analysis, and calculate its relative benchmark deviation based on industry / historical benchmarks; Step S2.2: Based on the calculated trend and relative benchmark deviation, a linear weighted average is used to generate field value evolution scores; Specifically, the step of calculating the changing trend of the same target field in different time periods based on time series analysis includes: dynamically calculating the year-on-year / month-on-month changing trend of the field using a time window sliding mechanism; wherein the changing trend includes: the magnitude of change and volatility.

5. The method for identifying and tracing fake enterprise information based on structured field propagation chains according to claim 1, characterized in that, Step S3 includes: Step S3.1: Based on the field propagation chain graph, obtain multiple related fields that have a propagation path association with the target field; Step S3.2: Configure the structured rule engine according to preset expert rules or preset regulatory logic; Step S3.3: Through the configured structured rule engine, including using Boolean logic gates to jointly judge the field values ​​corresponding to multiple related fields, the system identifies whether the path connecting multiple fields is a risk path as the rule gating judgment result of the target field.

6. A system for identifying and tracing fake enterprise information based on a structured field propagation chain, characterized in that, include: Module M1: Constructs a field propagation chain graph based on the relationships between fields in the enterprise database; Module M2: For the field values ​​of the target field in the field propagation chain graph at different time periods, calculate the changing trend of the field values ​​and the deviation relative to the industry benchmark or historical benchmark, and calculate the field value evolution score based on the calculated changing trend and the deviation relative to the benchmark. Module M3: Based on the field propagation chain graph, obtain multiple related fields that have a propagation path associated with the target field, and perform joint logical judgment on the field values ​​corresponding to the multiple related fields through the configured structured rule engine to generate a rule gating judgment result for the target field; Module M4: Based on the field value evolution score, field confidence and rule gating judgment result of the target field, perform multi-factor fusion calculation to obtain the false information identification result of the target field, and generate a field-level evidence chain corresponding to the identification result.

7. The fake enterprise information identification and tracing system based on structured field propagation chain according to claim 6, characterized in that, The module M1 includes: Module M1.1: Obtains enterprise field sets based on multiple heterogeneous databases, and converts the original fields into standard fields using a preset field semantic mapping table; Module M1.2: Selects reliable field values ​​that meet preset requirements based on time priority and weight for a standard field set; Module M1.3: Normalizes the selected field values ​​of the converted standard fields to obtain normalized field values; Module M1.4: Constructs an enterprise database based on standard fields and their normalized values; Module M1.5: Constructs a directional field propagation chain graph based on the computational dependencies, logical references, or semantic dependencies between fields in the enterprise database.

8. The fake enterprise information identification and tracing system based on structured field propagation chain according to claim 7, characterized in that, The nodes of the field propagation chain graph represent standard fields; the edges of the field propagation chain graph represent the relationships between fields; and the edge weights of the field propagation chain graph represent the value propagation sensitivity or derivation relationship between fields. The directional field propagation chain graph is used to represent the causal / dependency direction and supports truncation settings for path length to limit propagation depth.

9. The fake enterprise information identification and tracing system based on structured field propagation chain according to claim 6, characterized in that, The module M2 includes: Module M2.1: For the field values ​​of the same target field in different time periods, calculate the changing trend of the field in different time periods based on time series analysis, and calculate the relative benchmark deviation based on industry / historical benchmarks; Module M2.2: Based on the calculated trend of change and the relative deviation from the benchmark, a linear weighted average is used to generate field value evolution scores; Specifically, the step of calculating the changing trend of the same target field in different time periods based on time series analysis includes: dynamically calculating the year-on-year / month-on-month changing trend of the field using a time window sliding mechanism; wherein the changing trend includes: the magnitude of change and volatility.

10. The fake enterprise information identification and tracing system based on structured field propagation chain according to claim 6, characterized in that, The module M3 includes: Module M3.1: Based on the field propagation chain graph, obtain multiple related fields that have a propagation path association with the target field; Module M3.2: A structured rule engine configured based on preset expert rules or preset regulatory logic; Module M3.3: Through the configured structured rule engine, including the use of Boolean logic gates to jointly judge the field values ​​corresponding to multiple related fields, the module identifies whether the path connecting multiple fields is a risky path as the rule gating judgment result of the target field.

Images