National-owned enterprise operation evaluation method and system based on total event traceability, and computer readable medium

Abstract

The invention relates to the technical field of enterprise operation management evaluation, and discloses a national-owned enterprise operation evaluation method and system based on total event traceability and a computer readable medium, and the method comprises the steps: collecting original multi-source heterogeneous data in an enterprise operation process, converting the original multi-source heterogeneous data into a standard structure, and constructing a standardized three-dimensional event data set; performing weight attribute configuration and dynamic weight calibration on each standardized event number to obtain a dynamic event value weight table; performing political, economic and social event subset classification on the standardized event data, generating an initial score of each dimension, and generating a national enterprise three-dimensional evaluation result by using an event impact factor diffusion algorithm and nonlinear weighted fusion; and finally, determining an abnormal standard based on evaluation data of the same-industry and same-scale national-owned enterprises, carrying out abnormal identification and reverse traceability analysis, and constructing a complete event chain of a reason event, an intermediate event and a result event.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of enterprise management evaluation, in particular to a management method based on event value quantification evaluation and tracing, and specifically proposes a state-owned enterprise operation evaluation method, system and computer readable medium based on full event tracing, which is suitable for state-owned asset supervision and enterprise performance evaluation. BACKGROUND

[0002] The management of modern enterprises, especially state-owned enterprises, has a special market position with both economic and public attributes. The evaluation of its operation has shifted from traditional single financial performance assessment to a comprehensive evaluation system covering economic efficiency, governance capability, and social responsibility fulfillment. How to scientifically, objectively, and quantitatively evaluate the contribution of non-financial and non-economic governance activities to the overall value of the enterprise has become a technical problem that needs to be solved in the field of state-owned asset supervision.

[0003] The main methods for evaluating the operation of state-owned enterprises in the prior art include evaluation based on financial indicator system, qualitative evaluation based on expert subjective opinion, and comprehensive evaluation based on partial informatization:

[0004] The evaluation method based on the financial indicator system is to build a financial data warehouse, collect core financial data such as balance sheet, profit statement, and cash flow statement, and use financial ratios such as return on net assets and asset-liability ratio, as well as mathematical models such as Dupont analysis system to calculate the evaluation results. However, this method has a single evaluation dimension, focusing only on economic performance, and cannot structurally collect and quantify non-financial events such as party building and social responsibility. Moreover, the data is highly lagging, only reflecting historical operating results, making it difficult to dynamically monitor and warn the decision-making execution process.

[0005] To overcome the shortcomings of the financial indicator method, the qualitative evaluation method based on expert subjective opinion proposes to organize expert review meetings, combined with report review and on-site interviews, to qualitatively evaluate the governance level and social responsibility fulfillment of enterprises. However, the core problem lies in the strong subjectivity of the evaluation standard, the lack of a unified quantitative model, and the large influence of human factors on the results, making it difficult to reproduce. The evaluation efficiency is low, making it difficult to meet the actual needs of rapid and regular evaluation of a large number of state-owned enterprises.

[0006] The partial informatization comprehensive evaluation system mainly uses management tools such as balanced scorecard to add non-financial indicators such as customers and internal processes to the financial data layer, and then combines the scores of experts and questionnaire surveys to evaluate party building and social responsibility, and then weights and aggregates the financial scores. However, this method treats each indicator as an isolated point, cannot build a causal relationship execution chain from decision-making to execution to results, and does not clearly define the value transmission mechanism of non-economic events, making it difficult to scientifically quantify their dynamic impact on enterprise value.

[0007] The existing management evaluation methods do not treat enterprise operations as a dynamic process composed of interrelated events. They fail to achieve automatic event collection, standardized classification, in-depth correlation analysis, and scientific quantification, resulting in fundamental defects in the evaluation, such as one-sided dimensions, isolated and static data, and lagging and superficial results. They lack forward-looking and in-depth early warning capabilities and cannot meet the core requirements of state-owned asset supervision for comprehensiveness, accuracy, and objectivity. Summary of the Invention

[0008] In view of the defects and shortcomings of existing state-owned enterprise operation evaluation methods, the purpose of this invention is to provide a state-owned enterprise operation evaluation method based on full event traceability, which solves the problems of traditional evaluation having a single dimension, difficulty in quantifying non-financial events, lack of full-process traceability capability, and static evaluation. It realizes a comprehensive, quantitative, dynamic, and traceable integrated evaluation of the political, economic, and social three-dimensional value of state-owned enterprises, and improves the accuracy, depth and timeliness of state-owned asset supervision.

[0009] According to a first aspect of the present invention, a method for evaluating the operation of state-owned enterprises based on full event tracing is proposed, comprising the following steps:

[0010] Step 1: Collect original multi-source heterogeneous data during the enterprise's operation, and transform the multi-source heterogeneous data into standardized event data in a unified format to construct a standardized three-dimensional event dataset;

[0011] Step 2: Traverse the standardized three-dimensional event dataset, configure the weight attributes for each standardized event data according to its event characteristics, and perform dynamic weight calibration based on the event value transmission model to obtain a dynamic event value weight table.

[0012] Step 3: Based on the standardized three-dimensional event dataset and the dynamic event value weight table, classify all standardized event data into event subsets in three dimensions: political, economic, and social, and generate initial scores for each dimension. Then, use the event impact factor diffusion algorithm to iteratively calculate the comprehensive scores for each dimension, and generate the three-dimensional evaluation results of state-owned enterprises through nonlinear weighted fusion.

[0013] Step 4: Determine the anomaly criteria based on the assessment data of state-owned enterprises of the same industry and scale, compare the generated three-dimensional assessment results with the anomaly criteria to determine dimensional anomalies and comprehensive anomalies, and conduct reverse source tracing analysis based on the event association diagram determined by the event value transmission model, and construct a complete event chain of cause event → intermediate event → result event according to the anomaly.

[0014] According to a second aspect of the present invention, a computer system is provided, comprising:

[0015] One or more processors;

[0016] The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the aforementioned process of performing the state-owned enterprise operation evaluation method based on full event tracing.

[0017] In a third aspect of the present invention, a computer-readable medium for storing software is provided, the software comprising instructions executable by one or more computers, the instructions, when executed by the one or more computers, performing the aforementioned process of the state-owned enterprise operation evaluation method based on full event tracing.

[0018] The state-owned enterprise operation evaluation method based on full event tracing in the above embodiments of the present invention, through a three-dimensional event warehouse of political, economic and social aspects, places non-financial events such as Party committee decisions and social responsibilities on an equal footing with economic performance for quantitative evaluation, comprehensively reflecting the comprehensive value and mission of state-owned enterprises; for non-financial events, through a triple mechanism of policy guidance weight + AHP weight + dynamic calibration, abstract governance activities (such as Party committee pre-research and inspection rectification) are transformed into calculable and comparable quantitative indicators, which greatly reduces the subjective error of traditional expert evaluation and provides data-driven objective decision-making basis for state-owned asset supervision.

[0019] The method of this invention achieves dynamic updates of assessment weights and results through real-time policy analysis, timeliness decay factors, and real-time data collection, enabling pre-emptive warnings and in-process monitoring of risks, thus overcoming the shortcomings of static and lagging traditional assessments. During the assessment process, relying on event association graphs and reverse DFS algorithms, a causal event chain of decision-making-execution-result is constructed, realizing the transformation from result assessment to process assessment. This can accurately locate the root causes of problems, providing clear action paths for enterprise rectification and regulatory intervention, improving the efficiency of state-owned asset supervision and risk prevention capabilities. It can be applied to the quantitative assessment of the operating conditions of state-owned enterprises by state-owned asset supervision and management departments at all levels, the optimization of the operation and management of state-owned enterprises themselves, and the accounting of the effectiveness of the integration of Party building and operation. At the same time, it can realize the tracking and monitoring of the implementation effect of state-owned asset supervision policies at the enterprise level.

[0020] It should be understood that all combinations of the foregoing concepts and the additional concepts described in more detail below may be considered part of the inventive subject matter of this disclosure, provided that such concepts do not contradict each other. Furthermore, all combinations of the claimed subject matter are considered part of the inventive subject matter of this disclosure.

[0021] The foregoing and other aspects, embodiments, and features of the teachings of the present invention will be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the invention, such as features and / or beneficial effects of exemplary embodiments, will become apparent from the following description or may be learned through practice of specific embodiments according to the teachings of the present invention. Attached Figure Description

[0022] Figure 1 This is a flowchart of a state-owned enterprise operation evaluation method based on full event tracing according to an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of the process for event acquisition and construction of a three-dimensional event repository of standard events according to an embodiment of the present invention.

[0024] Figure 3 This is a flowchart illustrating the generation of a dynamic event value weight table based on event value quantification calculation according to an embodiment of the present invention.

[0025] Figure 4 This is a schematic diagram of the process of obtaining three-dimensional evaluation results through iterative calculations of each dimension according to an embodiment of the present invention.

[0026] Figure 5 This is a schematic diagram of an anomaly identification and tracing process based on an event chain according to an embodiment of the present invention. Detailed Implementation

[0027] To better understand the technical content of the present invention, specific embodiments are described below in conjunction with the accompanying drawings.

[0028] Various aspects of the invention are described in this disclosure with reference to the accompanying drawings, which illustrate numerous illustrative embodiments. The embodiments of this disclosure are not necessarily intended to encompass all aspects of the invention. It should be understood that the various concepts and embodiments described above, as well as those described in more detail below, can be implemented in any of many ways, because the concepts and embodiments disclosed herein are not limited to any particular implementation. Furthermore, some aspects of the invention disclosed may be used alone or in any suitable combination with other aspects of the invention disclosed.

[0029] Combination Figure 1 The state-owned enterprise operation evaluation method based on full event tracing shown in the embodiment includes the following steps:

[0030] Step 1: Collect original multi-source heterogeneous data during the enterprise's operation, and transform the multi-source heterogeneous data into standardized event data in a unified format to construct a standardized three-dimensional event dataset;

[0031] Step 2: Traverse the standardized three-dimensional event dataset, configure the weight attributes for each standardized event data according to its event characteristics, and perform dynamic weight calibration based on the event value transmission model to obtain a dynamic event value weight table.

[0032] Step 3: Based on the standardized three-dimensional event dataset and the dynamic event value weight table, classify all standardized event data into event subsets in three dimensions: political, economic, and social, and generate initial scores for each dimension. Then, use the event impact factor diffusion algorithm to iteratively calculate the comprehensive scores for each dimension, and generate the three-dimensional evaluation results of state-owned enterprises through nonlinear weighted fusion.

[0033] Step 4: Determine the anomaly criteria based on the assessment data of state-owned enterprises of the same industry and scale, compare the generated three-dimensional assessment results with the anomaly criteria to determine dimensional anomalies and comprehensive anomalies, and conduct reverse source tracing analysis based on the event association diagram determined by the event value transmission model, and construct a complete event chain of cause event → intermediate event → result event according to the anomaly.

[0034] In step 1, the aim is to collect heterogeneous event data and construct a standardized three-dimensional event dataset.

[0035] As an optional implementation method, original multi-source heterogeneous data is collected during the enterprise's operation, including:

[0036] For structured data, periodic incremental collection is used; for unstructured data, web crawlers are used to collect network file data and file parsing services are used to read and parse file data, and raw multi-source heterogeneous data is obtained concurrently.

[0037] Data buffering is achieved by configuring an asynchronous Kafka message queue. Three topics are created to store raw data for political, economic, and social events, respectively. Data collected from each data source is categorized and stored according to data source, collection time, and data type, generating a raw dataset. The data format includes structured data tables, unstructured file indexes, and API return data cache. Furthermore, each topic can be configured with three partitions and two replicas to ensure no data loss.

[0038] In specific examples, data interfaces can be used to connect to a company's business management system, such as financial systems, ERP systems, equity registration systems, investment management systems, and OA approval systems.

[0039] For structured data, connect to relational databases such as MySQL, Oracle, and SQL Server via JDBC database connector, configure database connection pool (maximum number of connections 50, minimum number of idle connections 10, connection timeout 30s), and collect data incrementally according to preset scheduling cycle (e.g., hourly).

[0040] For unstructured data, PDF, Word, and TXT format files (such as meeting minutes and work reports) are read through a file parsing service. At the same time, policy documents and social responsibility disclosure reports from public channels are collected through web crawlers. The crawler is configured with a crawling depth of 3 layers and a request interval of 1 second to avoid triggering the anti-crawling mechanism of the target website.

[0041] As an optional implementation, in step 1, multi-source heterogeneous data is transformed into standardized event data in a unified format to construct a standardized three-dimensional event dataset, including: entity type system definition and entity extraction, relation type definition and relation extraction, standardized encapsulation, and database entry operation.

[0042] As a concrete example, the pre-trained BERT-base model can be fine-tuned using a state-owned enterprise event annotation dataset (annotated samples covering scenarios such as Party building, investment, and environmental protection). The objective function can be optimized to the cross-entropy loss function, with 20 training rounds, a learning rate of 2e-5, and a batch size of 32, making the model suitable for entity recognition and relationship extraction tasks in specific fields of state-owned enterprises.

[0043] Five entity type systems are defined, including event subject (e.g., XX Group Office, Finance Department), event type (e.g., Party Committee Pre-research, Project Funding), occurrence time (e.g., 2023-03-15), key content (e.g., New Energy Battery Industrial Park Investment Proposal), and related entities (e.g., 200 million yuan, Environmental Assessment Report). Each record in the original dataset is then split into sentences and parsed using the BERT-base model to output entity boundaries and type labels.

[0044] In the entity extraction process, unstructured text is split into sentences and input into a fine-tuned BERT model. The output is entity boundaries and type labels. Ambiguous entities (such as mid-March in terms of time) are standardized and converted into ISO format timestamps (such as 2023-03-15T00:00:00Z) through a time parsing algorithm. The entity extraction confidence score is calculated. Entities with a confidence score below 0.8 are marked as low confidence and trigger manual review.

[0045] Three types of relationships are defined, including subject-action-object (e.g., Party Committee of Group A - passed - proposal), event-related entity (e.g., project investment - 200 million yuan), and event-time (e.g., environmental impact assessment report release - 2023-06-10). Then, for each record in the original dataset, the relationship is first clarified by matching regular rules, and then the complex semantic relationships are extracted by the BERT-base model.

[0046] In the relationship extraction process, a rule-model hybrid strategy is adopted. First, the relationship is clarified by matching regular rules (such as the proposal name after review and approval). Then, the complex semantic relationship is extracted by the BERT model. The rationality of the extracted relationship is verified. For example, the subject of the Party Committee's preliminary research must be the Party organization. Otherwise, it is marked as abnormal.

[0047] Then, the extracted entities and relationships are encapsulated into a unified JSON Schema data object to achieve standardized event object encapsulation.

[0048] As an example, the encapsulated field constraints are as follows:

[0049] event_id: Uses a prefix + timestamp + random number format (e.g., E_20230315_0001) to ensure global uniqueness;

[0050] event_source: an enumeration value, such as Party Building System, ERP System, OA_System), which cannot be empty;

[0051] event_time: ISO8601 format timestamp, accurate to the second;

[0052] event_subject: A non-empty string with a length not exceeding 100 characters;

[0053] event_type: Uses a two-level code, such as GOV_DECISION corresponding to the political-decision category, and is associated with the three-dimensional classification system;

[0054] key_content: A non-empty string with a length not exceeding 500 characters;

[0055] related_entities: An array type, with elements being strings, containing a maximum of 10 related entities;

[0056] extraction_confidence: Floating-point type, value range 0-1, rounded to two decimal places.

[0057] Finally, the extracted elements are mapped to a predefined three-dimensional political-economic-social event classification system to generate standardized event data with a unified JSON format. This standardized event data includes event ID, type code, timestamp, responsible party, and description fields. After verifying the integrity and format of the standardized event data, it is stored in a standardized three-dimensional event dataset to realize the construction of a three-dimensional event repository.

[0058] As an optional implementation, the 3D event repository uses the ClickHouse distributed columnar database, and the data table structure is designed as follows:

[0059] Table name: event_warehouse, divided into tables by first-level dimension (event_political, event_economic, event_social).

[0060] Core fields: event_id (primary key), event_source, event_time, event_subject, event_type, key_content, related_entities, extraction_confidence, classify_result (classification result), create_time (database entry time);

[0061] Index design: A primary key index is created for event_id, and a composite index is created for event_time, event_subject, and classify_result, supporting fast queries by time range, subject, and category;

[0062] Storage strategy: Set a 3-year data retention period, automatically archive historical data to low-cost storage nodes, and store hot data (the last year) on high-performance nodes.

[0063] As an example, the data collection and standardization process is as follows:

[0064] (1) Multi-source data access: The system collected an unstructured minutes of a Party Committee resolution from the Party building system, which read: "A Party Committee meeting was held on March 15, 2023, and the 'Proposal on Investing in the Construction of a New Energy Battery Industrial Park' was reviewed and approved."

[0065] (2) Event structuring: The built-in NLP engine (BERT model) processes the minutes. Through Named Entity Recognition (NER) technology, the event subject is extracted as follows: Party Committee of Group A, event type: pre-event research of the Party Committee, occurrence time: 2023-03-15, key content: approval of the investment proposal for the new energy battery industrial park;

[0066] (3) Standardization and storage: The system maps these elements to a predefined three-dimensional event classification system of "politics-economy-society", generates a standardized event object, and assigns a unique IDE001; the object is classified as a "political-decision" event and stored in the three-dimensional event repository in a unified JSON format;

[0067] Other example events processed concurrently:

[0068] E002 (Economy-Investment): On April 1, 2023, data was collected from the ERP system showing that "amounting to the new energy battery industrial park project was 200 million yuan".

[0069] E003 (Social - Environmental Protection): On June 10, 2023, "Release the Environmental Impact Assessment Report of the Industrial Park" was collected from the project management system.

[0070] (4) Output: A standardized event dataset containing events such as E001, E002, E003, etc.

[0071] Further, in step 2, it aims to quantitatively evaluate the contribution of non - financial events through event value quantification calculation and generate a dynamic event value weight table. By obtaining the policy texts of the regulatory authorities through real - time updates, using algorithms to extract keywords and perform semantic analysis, a policy weight knowledge base is constructed and maintained to quantify the emphasis of policies on each dimension.

[0072] As an optional implementation method, in step 2, weight attributes are configured for each standardized event data according to its event characteristics, including:

[0073] Use the TF - IDF algorithm to extract keywords from policy documents, filter stop words and retain core words related to political, economic, and social attributes;

[0074] Construct a dimension - keyword mapping dictionary, calculate the emphasis of policy documents on each dimension through word frequency statistics, and achieve dimension correlation calculation;

[0075] Calculate the dimension correlation for all collected policy documents, and take the average value as the initial dimension weight of each dimension under the first - level dimension, denoted as the first - level dimension weight; and

[0076] Under the first - level dimension, use the analytic hierarchy process to determine the event type weight with the contribution degree of the event to the dimension goal as the criterion, which is used as the second - level event type weight.

[0077] In a specific embodiment, the official websites of departments such as the State - owned Assets Supervision and Administration Commission of the State Council, the Ministry of Finance, and the National Development and Reform Commission, as well as the policy columns of authoritative media such as People's Daily and Economic Daily, can be continuously monitored through web crawlers. The crawlers are configured to execute regularly every day to collect the released policy documents; duplicate the collected policy documents (based on the file title + release time) and store them in the policy database.

[0078] Then use the TF - IDF algorithm to extract keywords from policy documents, filter stop words (such as "of", "and", "for"), and retain core words related to politics, economy, and society (such as Party building, reform, development, environmental protection, people's livelihood, etc.).

[0079] Then construct a dimension - keyword mapping dictionary, for example, the political dimension is associated with Party building, Party committees, policies, the economic dimension is associated with economy, investment, revenue, etc., and the social dimension is associated with environmental protection, people's livelihood, responsibility), and calculate the emphasis degree R of policy documents on each dimension through word frequency statistics. d:

[0080] R d =W d / W p ;

[0081] Among them, W d W p These represent the total keyword frequency for dimension d and the total keyword frequency for policy documents, respectively. d=1, 2, 3, corresponding to the political, economic, and social dimensions, respectively.

[0082] As an optional implementation, the dimensional correlation degree is calculated for all collected policy documents, and the average value is taken as the initial dimensional weight for each dimension under the first-level dimension, denoted as the first-level dimensional weight, including:

[0083]

[0084] Where n represents the number of policy documents under the corresponding dimension.

[0085] After obtaining the initial dimension weights, further verification can be conducted. For example, 5-7 experts in state-owned asset supervision and enterprise management can review the initial weights. If the experts' opinions differ from the initial weights by more than 10%, the Delphi method should be used for multiple rounds of consultation to finally determine the calibrated first-level dimension weights W1, including the first-level dimension weights of the political dimension. The first-level weight of the economic dimension The first-level weight of the social dimension .

[0086] In this embodiment, the first-level dimension weight of the political dimension The first-level weight of the economic dimension The first-level weight of the social dimension The values ​​are 0.35, 0.45, and 0.20 respectively, and are recorded as follows: Politics (P) = 0.35, Economy (E) = 0.45, Society (S) = 0.20.

[0087] As an optional implementation, the policy weight library is updated monthly, and newly added policy documents are automatically included in the analysis, with dimension weights recalculated to ensure that the weights are synchronized with policy guidance in real time.

[0088] Furthermore, at the first-level dimension, the Analytic Hierarchy Process (AHP) is used to calculate the relative weights of various events based on their contribution to the dimension's objectives. For example, in the political dimension, the weight of the Party Committee's preliminary research event is 25%.

[0089] It should be understood that, in the embodiments of the present invention, based on the causal logic of business operations and decision-making, the ultimate value of an event depends not only on its own type, but also on the dynamic influence of its timeliness and relevance.

[0090] Therefore, based on the event value transmission model, dynamic weight calibration is performed. The dynamic weight correction of this invention is achieved through two core factors: the timeliness decay factor f. t With correlation enhancement factor f r The dynamic event value weight table is dynamically adjusted, specifically through the following process:

[0091] Based on the impact cycle of events involving state-owned enterprises, attenuation coefficients are assigned to political, economic, and social events, and a timeliness attenuation factor f is calculated based on these coefficients. t ;

[0092] Using events as nodes and establishing edges to connect them, an event association graph is constructed.

[0093] Based on the event correlation graph, the correlation enhancement factor f is calculated according to the in-degree and out-degree statistics of events and the correlation gain coefficient. r ;as well as

[0094] Based on primary dimension weights, secondary event type weights, and timeliness decay factor f t and correlation enhancement factor f r Dynamic calibration is performed, and the final impact weight of the event is calculated using the formula: W1×W2×f t ×f r W1 is the weight of the first-level dimension, and W2 is the weight of the second-level event type.

[0095] For example, for event E001 (political-decision-based-Party committee pre-research), based on political (P) = 0.35, social (S) = 0.20, f t =0.11, f r =1.06;

[0096] Further calculations show that the final impact weight W = 0.35 × 0.25 × 0.11 × 1.06 ≈ 0.010.

[0097] Based on the calculation results for each event, an event value weight table is generated, including event ID, primary dimension, secondary category, tertiary category, primary dimension weight, secondary event type weight, and timeliness decay factor f. t and correlation enhancement factor f r Ultimately, this affects the weights, which are then stored in the weight database.

[0098] As an optional implementation method, the aforementioned method configures attenuation coefficients for political, economic, and social events based on the impact cycle of state-owned enterprise events, and calculates a timeliness attenuation factor f based on these attenuation coefficients. t ,include:

[0099] The boundary range for setting the attenuation coefficient is 0.01-0.05; the attenuation coefficient for political events is set to 0.02 (longer impact period), for economic events to 0.03 (medium impact period), and for social events to 0.04 (shorter impact period); these can be customized according to industry characteristics.

[0100] The assessment time difference is determined by calculating the difference between the current assessment date and the event occurrence date, denoted as Δt; and

[0101] Calculate the time-dependent attenuation factor f based on the attenuation coefficient. t (The older the event, the lower its current impact), calculated as follows:

[0102] f t =e -λ×Δt ;

[0103] Where λ represents the attenuation coefficient for each type of event.

[0104] As an optional implementation, the aforementioned method of constructing an event association graph by using events as nodes and establishing edge connections based on event associations includes:

[0105] Using events as nodes, if two events satisfy semantic association and temporal association: semantic association means keyword overlap rate ≥ 30%, and temporal association means time interval ≤ 90 days, then establish an edge connection, construct an event association graph with nodes and edges, and store it through an adjacency list.

[0106] As an optional implementation, the aforementioned correlation enhancement factor f is calculated based on the event correlation graph, according to the event in-degree and out-degree statistics and the correlation gain coefficient. r ,include:

[0107] Based on the event relationship graph, the in-degree and out-degree are calculated. The in-degree refers to the number of previous events referenced by the current event, and the out-degree refers to the number of subsequent events referenced by the current event. For example, if "project funding" references "preliminary research by the Party Committee", then the in-degree of "project funding" increases by 1. If "preliminary research by the Party Committee" is referenced by "project funding" and "environmental impact assessment report", then the out-degree of "preliminary research by the Party Committee" increases by 2.

[0108] The correlation enhancement factor f is calculated based on the correlation gain coefficient. rIf an event is referenced by multiple subsequent events (e.g., there are outgoing edges in the semantic association graph), or if it references multiple previous events (there are incoming edges), its weight will be enhanced. The specific calculation formula is as follows:

[0109] f r =1+β×(in-degree+out-degree).

[0110] In the optional example, the correlation gain coefficient β is set to 0.03, and can be adjusted according to the correlation strength: strong correlation β=0.05, weak correlation β=0.02.

[0111] As a specific example, the dynamic weight adjustment is as follows:

[0112] Event E001 (preliminary research by the Party Committee) occurred relatively recently. t ≈1.0. Meanwhile, the system, through semantic analysis, discovered that E002 (project funding) and E003 (environmental impact assessment report) are both strongly correlated with E001 in terms of content, meaning E001 has outgoing edges in the association graph. Assuming f... r =1.05; therefore, the final weight calculation is: W = W1 × W2 × f t ×f r ;

[0113] Example calculation: The final weight of E001, W001 = 0.35(P) × 0.25 (Party Committee's preliminary research) * 1.0(f) t )×1.05(f r =0.0919 (9.19%).

[0114] Output: Event value weight table, which clearly lists each event ID and its calculated final weight, for example: E001: 9.19%, E002: 12%, E003: 3.5%.

[0115] As an optional implementation, in step 3, based on the standardized three-dimensional event dataset and the dynamic event value weight table, all standardized event data are classified into event subsets according to the political, economic, and social dimensions, and initial scores for each dimension are generated. The event impact factor diffusion algorithm is then used to iteratively calculate the comprehensive scores for each dimension. The three-dimensional evaluation results of state-owned enterprises are generated through nonlinear weighted fusion, including the following processes:

[0116] Step 3.1: Read standardized event data from the standardized three-dimensional event dataset. Based on the first-level dimension information of the event classification result field (classify_result), divide the events into event subsets of three dimensions: political, economic, and social, and store them in three temporary data tables respectively. For example,

[0117] Original cumulative value of political dimension: TotalW p =Σ(final weight W of all political events);

[0118] Original cumulative value of the economic dimension: TotalW e =Σ(final weight W of all economic events);

[0119] Original cumulative value of social dimension: TotalW s =Σ(final weight W of all social events);

[0120] Step 3.2: For all events within the event subset of each dimension, incorporate the first-level dimension weights and sum them to obtain the final weights, thus obtaining the initial scores for each dimension;

[0121] In this embodiment, the initial score is a weighted sum of the original accumulated value and the weights of the first-level dimension:

[0122] Initial score for the political dimension: ;

[0123] Initial score for the economic dimension: ;

[0124] Initial score for the social dimension: ;

[0125] The initial score can be further converted to a percentage system: K is the conversion coefficient, which defaults to 1000 and can be adjusted according to the industry average weight to ensure that the score range is between 0 and 100.

[0126] Step 3.3: Based on the constructed event association graph, supplement the association strength weight w. ij :

[0127] w ij =(semantic similarity × temporal relevance) / max(semantic similarity × temporal relevance);

[0128] Semantic similarity is calculated based on the cosine similarity of event keyword vectors. Temporal relevance is calculated as 1 - |Δt / T|, where T is the maximum time interval threshold, set to 365 days. The association strength weight is w. ij The value range is 0-1;

[0129] Step 3.4: Combine the initial scores from the three dimensions into a vector S0, represented as:

[0130] ;

[0131] Based on the configured damping factor α and the iteration termination threshold as a convergence condition. Diffusion iteration calculation is performed, and the diffusion iteration calculation formula is as follows:

[0132] S k =α×P×S k-1 +(1-α)××S0;

[0133] Among them, S k S k-1 Let represent the score vectors after the kth and (k-1)th iterations, respectively;

[0134] P is the transition matrix, obtained by normalizing the adjacency matrix A of dimension n×n; the adjacency matrix A is constructed based on the event association graph, and for the element A in the i-th row and j-th column of the adjacency matrix A... ij If event i is associated with event j, then A ij =w ij Otherwise A ij =0;

[0135] Step 3.5: Based on the final scores S of each dimension obtained after convergence. p S e S s The corresponding first-level dimension weights are then used to calculate the comprehensive score through weighted summation, as follows:

[0136] Overall score = .

[0137] As a specific example, the damping factor α is set to 0.8 (range 0.7-0.9) to balance the original score with the diffusion effect.

[0138] The convergence condition for iterative calculation is set as: the L2 norm of the difference between the score vectors of two consecutive iterations.

[0139] |S k -S k-1 Iteration stops when | < 0.001, the iteration termination threshold. The configuration is 0.001.

[0140] For example, in the association graph, E001 (political) has edges pointing to E002 (economic) and E003 (social). When the algorithm runs, the value of E001 will partially "diffly" to E002 and E003, thus dynamically improving the final score of the economic and social dimensions based on the initial score.

[0141] As an example, suppose the initial vector S0 = [10, 22, 34]. TIn the transition matrix P, the correlation strength between the political dimension and the economic dimension is 0.5, and the correlation strength between the political dimension and the social dimension is also 0.5. In the first iteration: S1 = 0.8 × P × S0 + 0.2 × S0, we obtain S1 = [11.2, 23.8, 35.6]. T Second iteration: S2 = 0.8 × P × S1 + 0.2 × S0, resulting in S2 = [11.8, 24.5, 36.2]. T If |S3-S2|<0.001 is satisfied after the third iteration, convergence stops.

[0142] Combining the iterative expansion process described above, this approach aims to simulate an iterative diffusion of the impact of an event using a graph-based semi-supervised learning algorithm (LabelPropagation). For example, a political decision event might positively enhance the score of its associated economic project event. Each event node transmits its evaluation value (a function of weights and initial scores) to its neighboring nodes in a proportional manner (determined by the edge weights). This process iterates multiple times until the values ​​of the entire graph network stabilize.

[0143] As an optional implementation, in step 4, anomaly criteria are determined based on assessment data from state-owned enterprises of the same industry and scale. The generated three-dimensional assessment results are compared with the anomaly criteria to determine dimensional anomalies and comprehensive anomalies. Reverse tracing analysis is then performed based on the event correlation diagram determined by the event value transmission model. A complete event chain is constructed based on the anomalies, consisting of causal events → intermediate events → result events, including:

[0144] Step 4.1: Collect evaluation data from state-owned enterprises of the same industry and size, calculate the average score μ and standard deviation σ for each dimension, and set the industry anomaly standard as μ-σ:

[0145] Step 4.2: Traverse the comprehensive scoring results of the 3D evaluation, perform anomaly screening, mark the anomaly type, and store the results in the anomaly result table:

[0146] If the score of any dimension is lower than the abnormality standard and the difference exceeds 10 points, the dimension is judged to be abnormal.

[0147] Those whose overall score ranks in the bottom 10% of the industry or whose overall score is more than 15 points below the industry's abnormal standard are judged as having an overall abnormality.

[0148] Step 4.3: For dimensional anomalies, select the top 10% of events by weight in that dimension as the starting point for tracing the source; for comprehensive anomalies, select the top 5% of events by weight in all three dimensions as the starting point for tracing the source.

[0149] Step 4.4: Start from the source node and perform a reverse depth-first search based on the reverse edges of the event association graph. That is, if event j references event i, then the reverse edge is j→i. Recursively traverse the upstream events and record the occurrence time, key content, and association relationship with downstream events for each event.

[0150] Step 4.5: Sort the upstream events found by time sequence and causal relationship, and construct a complete event chain of cause event → intermediate event → result event. For example, the Party Committee's preliminary research did not emphasize environmental protection requirements (E101) → the environmental impact assessment report was delayed (E102) → low score in the social dimension.

[0151] As an optional implementation, for identified anomalies, such as dimensional anomalies or comprehensive anomalies, an early warning operation is triggered, such as compliance risk warning, operational risk warning, social responsibility risk warning, and policy response delay warning.

[0152] As a concrete example:

[0153] When the political dimension score is more than 15 points lower than the industry benchmark, a compliance risk warning is triggered, and the warning information can be pushed to the state-owned assets supervision and administration department, the enterprise's discipline inspection department, etc. for supervision.

[0154] When the economic dimension score is more than 20 points lower than the industry benchmark, an operational risk warning is triggered, and the warning information is pushed to the state-owned assets supervision and management department and senior management of the enterprise for supervision.

[0155] When the social dimension score is more than 10 points lower than the industry benchmark, a social responsibility risk warning is triggered, and the warning information is pushed to state-owned asset supervision personnel, corporate social responsibility departments, etc.

[0156] When the weight of policy implementation events in the political dimension is less than 5%, a policy response delay warning is triggered, and the warning information is pushed to the enterprise's Party building department, office, etc. for supervision and intervention.

[0157] As an optional implementation method, the early warning information generation includes early warning type, triggering conditions, anomaly score, source tracing summary, rectification suggestions, etc., with rectification suggestions generated based on the event chain tracing results.

[0158] As an optional implementation, the method further includes an operation for generating a source tracing report:

[0159] For each abnormal result, a source tracing report is generated, which includes the abnormality type, abnormality score, industry benchmark, complete event chain, and timeline of events.

[0160] Furthermore, the reasonableness of the event chain for anomaly identification and tracing can be judged, including:

[0161] Logical coherence: Whether the causal relationship between events in the event chain conforms to the business logic, such as decision-making gaps → execution deviations → abnormal results;

[0162] Time matching: Whether the occurrence times of each event in the event chain are arranged in causal order and the time intervals are within a reasonable range, such as the time interval between decision events and execution events being ≤180 days;

[0163] Significance of impact: The weight of the causal event is ≥ 20% of the total weight of the anomaly dimension, or the correlation strength between the causal event and the result event is ≥ 0.6.

[0164] If the event chain meets the above criteria, the evaluation result is deemed reasonable, and a verified evaluation result is generated, along with a traceability report.

[0165] If there are logical breaks, time mismatches, or insignificant impacts in the event chain, it is marked as an evaluation anomaly, triggering the data review process to re-examine the event collection and quantification process.

[0166] In a further embodiment, the traceability results can be used as feedback during the operation process. Using the ECharts visualization library, a three-dimensional radar chart can be generated with political, economic and social dimensions as the three axes and the comprehensive score as the value on the axis for visualization. It can also support filtering and viewing by evaluation period and enterprise type.

[0167] In an optional embodiment, an event propagation path report may further be generated, including:

[0168] The complete event chain is presented using flowcharts combined with necessary explanatory text.

[0169] Successful event chain, such as Party Committee's preliminary research → Project investment → Revenue growth;

[0170] The problem chain is as follows: insufficient emphasis on environmental protection requirements → delay in environmental impact assessment → low social score.

[0171] At the same time, the weight, correlation strength, and time nodes of each event are marked to serve the optimization of state-owned asset supervision and enterprise operation.

[0172] In optional embodiments, a problem tracing list can be further fed back based on the anomaly identification and tracing results. For example, the abnormal problems can be listed by dimension classification, including fields such as problem description, related event chain, impact level (high / medium / low), and priority of rectification suggestions.

[0173] As a specific example, see below:

[0174] In conjunction with the above embodiments of the state-owned enterprise operation evaluation method based on full event tracing, the present invention also proposes a computer system, comprising:

[0175] One or more processors;

[0176] Memory stores instructions that can be operated.

[0177] When the instruction is executed by the one or more processors, it causes the one or more processors to perform an operation, which includes the process of executing the state-owned enterprise operation evaluation method based on full event tracing of any of the foregoing embodiments.

[0178] In conjunction with the above embodiments of the state-owned enterprise operation evaluation method based on full event tracing, the present invention also proposes a computer-readable medium for storing software, the software including instructions executable by one or more computers.

[0179] These instructions, when executed by the one or more computers, perform the process of evaluating the operation of state-owned enterprises based on full event sourcing, as described in any of the preceding embodiments.

[0180] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention shall be determined by the claims.

Claims

1. A method for evaluating the operation of state-owned enterprises based on full event tracing, characterized in that, Includes the following steps: Step 1: Collect original multi-source heterogeneous data during the enterprise's operation, and transform the multi-source heterogeneous data into standardized event data in a unified format to construct a standardized three-dimensional event dataset; Step 2: Traverse the standardized three-dimensional event dataset, configure the weight attributes for each standardized event data according to its event characteristics, and perform dynamic weight calibration based on the event value transmission model to obtain a dynamic event value weight table. Step 3: Based on the standardized three-dimensional event dataset and the dynamic event value weight table, classify all standardized event data into event subsets in three dimensions: political, economic, and social, and generate initial scores for each dimension. Then, use the event impact factor diffusion algorithm to iteratively calculate the comprehensive scores for each dimension, and generate the three-dimensional evaluation results of state-owned enterprises through nonlinear weighted fusion. Step 4: Determine the anomaly criteria based on the assessment data of state-owned enterprises of the same industry and scale, compare the generated three-dimensional assessment results with the anomaly criteria to determine dimensional anomalies and comprehensive anomalies, and conduct reverse source tracing analysis based on the event association diagram determined by the event value transmission model, and construct a complete event chain of cause event → intermediate event → result event according to the anomaly.

2. The method for evaluating the operation of state-owned enterprises based on full event tracing as described in claim 1, characterized in that, In step 1, the original multi-source heterogeneous data collected during the enterprise's operation includes: For structured data, periodic incremental collection is used; for unstructured data, web crawlers are used to collect network file data and file parsing services are used to read and parse file data, and raw multi-source heterogeneous data is obtained concurrently. Data buffering is achieved by configuring Kafka asynchronous message queues. Three topics are created to store raw data of political, economic, and social events respectively. Data collected from each data source is categorized and stored according to data source, collection time, and data type to generate a raw data set. The data format includes structured data tables, unstructured file indexes, and API return data cache.

3. The method for evaluating the operation of state-owned enterprises based on full event tracing as described in claim 2, characterized in that, In step 1, multi-source heterogeneous data is transformed into standardized event data in a unified format to construct a standardized three-dimensional event dataset, including: Five entity type systems are defined, including event subject, event type, occurrence time, key content, and related entities. Then, each record in the original dataset is split into sentences and parsed using the BERT-base model to output entity boundaries and type labels. Three types of relationships are defined, including subject-action-object, event-related entity, and event-time relationship. Then, for each record in the original dataset, the relationship is first clarified by regular expression matching, and then the complex semantic relationship is extracted by the BERT-base model. The extracted entities and relationships are encapsulated into a unified JSON Schema data object to achieve standardized event object encapsulation; The extracted elements are mapped to a predefined three-dimensional political-economic-social event classification system to generate standardized event data with a unified JSON format. This standardized event data includes event ID, type code, timestamp, responsible party, and description field. After verifying the integrity and format of the standardized event data, it is stored in the standardized 3D event dataset.

4. The method for evaluating the operation of state-owned enterprises based on full event tracing as described in claim 2, characterized in that, In step 2, weight attributes are configured for each standardized event data point based on its event characteristics, including: The TF-IDF algorithm is used to extract keywords from policy documents, filter out stop words, and retain core words related to political, economic, and social attributes. Construct a dimension-keyword mapping dictionary, and calculate the degree of emphasis of policy documents on each dimension through word frequency statistics to realize the calculation of dimension relevance; Calculate the dimensional relevance for all collected policy documents, and take the average value as the initial dimensional weight for each dimension under the first-level dimension, denoted as the first-level dimensional weight; and At the first-level dimension, the analytic hierarchy process (AHP) is used to determine the weights of event types based on their contribution to the dimension's objective, which then serve as the weights of the second-level event types.

5. The method for evaluating the operation of state-owned enterprises based on full event tracing as described in claim 4, characterized in that, In step 2, dynamic weight calibration is performed based on the event value transmission model to obtain a dynamic event value weight table, including: Based on the impact cycle of events involving state-owned enterprises, attenuation coefficients are assigned to political, economic, and social events, and a timeliness attenuation factor f is calculated based on these coefficients. t ; Using events as nodes and establishing edges to connect them, an event association graph is constructed. Based on the event correlation graph, the correlation enhancement factor f is calculated according to the in-degree and out-degree statistics of events and the correlation gain coefficient. r ;as well as Based on primary dimension weights, secondary event type weights, and timeliness decay factor f t and correlation enhancement factor f r Perform dynamic calibration and calculate the final impact weight of the event; Based on the calculation results for each event, an event value weight table is generated, including event ID, primary dimension, secondary category, tertiary category, primary dimension weight, secondary event type weight, and timeliness decay factor f. t and correlation enhancement factor f r Ultimately, this affects the weights, which are then stored in the weight database.

6. The method for evaluating the operation of state-owned enterprises based on full event tracing as described in claim 5, characterized in that, Based on the impact cycle of events involving state-owned enterprises, attenuation coefficients are assigned to political, economic, and social events, and a timeliness attenuation factor f is calculated based on these coefficients. t ,include: Set the attenuation coefficient to a range of 0.01-0.05; configure the attenuation coefficient for political events to be 0.02, for economic events to be 0.03, and for social events to be 0.

04. The assessment time difference is determined by calculating the difference between the current assessment date and the event occurrence date, denoted as Δt; and Calculate the time-dependent attenuation factor f based on the attenuation coefficient. t : f t =e -λ×Δt ; Where λ represents the attenuation coefficient for each type of event.

7. The method for evaluating the operation of state-owned enterprises based on full event tracing as described in claim 5, characterized in that, The process of constructing an event association graph by using events as nodes and establishing edge connections based on event associations includes: Using events as nodes, if two events satisfy semantic association and temporal association: semantic association means keyword overlap rate ≥ 30%, and temporal association means time interval ≤ 90 days, then establish an edge connection, construct an event association graph with nodes and edges, and store it through an adjacency list.

8. The method for evaluating the operation of state-owned enterprises based on full event tracing as described in claim 5, characterized in that, The correlation enhancement factor f is calculated based on the event correlation graph, according to the in-degree and out-degree statistics of events and the correlation gain coefficient. r ,include: Based on the event relationship graph, the in-degree and out-degree are calculated. The in-degree refers to the number of previous events referenced by the current event, and the out-degree refers to the number of subsequent events referenced by the current event. The correlation enhancement factor f is calculated based on the correlation gain coefficient. r : f r =1+β×(in-degree+out-degree).

9. The method for evaluating the operation of state-owned enterprises based on full event tracing as described in claim 1, characterized in that, In step 3, based on the standardized three-dimensional event dataset and the dynamic event value weight table, all standardized event data are classified into event subsets according to the political, economic, and social dimensions, and initial scores for each dimension are generated. The event impact factor diffusion algorithm is then used to iteratively calculate the comprehensive scores for each dimension. Finally, a nonlinear weighted fusion is used to generate the three-dimensional evaluation results for state-owned enterprises, including the following processes: Step 3.1: Read standardized event data from the standardized three-dimensional event dataset. Based on the first-level dimension information of the event classification result field, divide the events into event subsets of three dimensions: political, economic, and social, and store them in three temporary data tables respectively. Step 3.2: For all events within the event subset of each dimension, sum their final weights to obtain the initial score for each dimension; Step 3.3: Based on the constructed event association graph, supplement the association strength weight w. ij : w ij =(semantic similarity × temporal relevance) / max(semantic similarity × temporal relevance); Semantic similarity is calculated based on the cosine similarity of event keyword vectors. Temporal relevance is calculated as 1 - |Δt / T|, where T is the maximum time interval threshold, set to 365 days. The association strength weight is w. ij The value range is 0-1; Step 3.4: Combine the initial scores from the three dimensions into a vector S0, and then apply the configured damping factor α and the iteration termination threshold as a convergence condition. Diffusion iteration calculation is performed, and the diffusion iteration calculation formula is as follows: S k =α×P×S k-1 +(1-α)××S0; Among them, S k S k-1 Let represent the score vectors after the kth and (k-1)th iterations, respectively; P is the transition matrix, obtained by normalizing the adjacency matrix A of dimension n×n; the adjacency matrix A is constructed based on the event association graph, and for the element A in the i-th row and j-th column of the adjacency matrix A... ij If event i is associated with event j, then A ij =w ij Otherwise A ij =0; Step 3.5: Based on the final scores S of each dimension obtained after convergence. p S e S s The corresponding first-level dimension weights are used to calculate the comprehensive score through weighted summation.

10. The method for evaluating the operation of state-owned enterprises based on full event tracing as described in claim 9, characterized in that, In step 4, anomaly criteria are determined based on assessment data from state-owned enterprises of similar size and industry. The generated three-dimensional assessment results are compared with these criteria to determine dimensional and overall anomalies. A reverse tracing analysis is then performed based on the event correlation diagram determined by the event value transmission model. Finally, a complete event chain is constructed based on the anomalies, consisting of causal events → intermediate events → result events, including: Step 4.1: Collect evaluation data from state-owned enterprises of the same industry and size, calculate the average score μ and standard deviation σ for each dimension, and set the industry anomaly standard as μ-σ: Step 4.2: Traverse the comprehensive scoring results of the 3D evaluation, perform anomaly screening, mark the anomaly type, and store the results in the anomaly result table: If the score of any dimension is lower than the abnormality standard and the difference exceeds 10 points, the dimension is judged to be abnormal. Those whose overall score ranks in the bottom 10% of the industry or whose overall score is more than 15 points below the industry's abnormal standard are judged as having an overall abnormality. Step 4.3: For dimensional anomalies, select the top 10% of events by weight in that dimension as the starting point for tracing the source; for comprehensive anomalies, select the top 5% of events by weight in all three dimensions as the starting point for tracing the source. Step 4.4: Start from the source node and perform a reverse depth-first search based on the reverse edges of the event association graph. That is, if event j references event i, then the reverse edge is j→i. Recursively traverse the upstream events and record the occurrence time, key content, and association relationship with downstream events for each event. Step 4.5: Sort the upstream events found by time sequence and causal relationship to construct a complete event chain of cause event → intermediate event → result event.

11. The method for evaluating the operation of state-owned enterprises based on full event tracing according to any one of claims 1-10, characterized in that, The method further includes the following steps: For each abnormal result, a source tracing report is generated, which includes the abnormality type, abnormality score, industry benchmark, complete event chain, and timeline of events.

12. The method for evaluating the operation of state-owned enterprises based on full event tracing as described in any one of claims 1-10, characterized in that, The method further includes the following steps: Using the ECharts visualization library, a 3D radar chart is generated with political, economic, and social dimensions as the three axes and a comprehensive score as the numerical value on the axis for visualization.

13. A computer system, characterized in that, include: One or more processors; A memory that stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the process of performing the method of any one of claims 1-12.

14. A computer-readable medium for storing software, characterized in that, The software includes instructions executable by one or more computers, which, when executed by the one or more computers, perform the process of the method as described in any one of claims 1-12.

Images