Project risk event identification method and device, electronic equipment and storage medium

Abstract

This application relates to a method, apparatus, electronic device, and storage medium for identifying project risk events. The method includes: acquiring multi-source unstructured text of a project; preprocessing the unstructured text to obtain standardized text fragments; inputting each standardized text fragment into a deep semantic model, where the deep semantic model identifies whether the standardized text fragment is a candidate risk text; extracting structured risk event objects from the candidate risk text; associating each structured risk event object with nodes in a pre-set project knowledge graph to obtain the association and impact information of each structured risk event object in the project; and classifying the risk of each structured risk event object; and generating visualized risk information based on the risk classification results and association and impact information. This method achieves automatic and accurate identification and global analysis of project risk events, significantly improving identification accuracy, cross-task association capabilities, standardization level, and visualization level.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and in particular to a method, apparatus, electronic device and storage medium for identifying project risk events. Background Technology

[0002] In large-scale project management, project teams generate a large amount of unstructured text daily (such as meeting minutes, daily reports, emails, and project documents), which may contain hidden risk signals such as task delays, delivery conflicts, and client objections. Traditional risk management relies on manual analysis, which suffers from problems such as low efficiency, delayed risk discovery, difficulty in linking risks across teams, tasks, and phases, and a lack of standardized classification.

[0003] Existing technologies mainly fall into two categories: rule-based matching methods, which scan text using keywords or regular expressions, but cannot understand contextual semantics, leading to frequent omissions and false positives; and traditional machine learning methods, which use classifiers to identify risk patterns, but rely on large amounts of labeled data, have limited generalization ability, and struggle to handle complex semantics across sentences and documents. Furthermore, existing methods generally lack the ability to perform unified correlation analysis on multi-source heterogeneous text, failing to achieve cross-team, cross-task, and cross-stage integration and structured expression of risk information, resulting in fragmented risk identification results that are difficult to support overall project decision-making. Summary of the Invention

[0004] This application provides a method, apparatus, electronic device, and storage medium for identifying project risk events, in order to solve the technical problems of low accuracy, high rate of missed and false judgments in risk identification in the prior art, and difficulty in achieving risk correlation analysis across teams, tasks, and stages.

[0005] Firstly, this application provides a method for identifying project risk events, the method comprising: Obtain multi-source unstructured text of the project, preprocess each unstructured text separately to obtain multiple standardized text fragments; Each of the standardized text fragments is input into a pre-trained deep semantic model, which then identifies whether the standardized text fragment is a candidate risk text. Extract structured risk event objects from the identified candidate risk texts; Each of the structured risk event objects is associated with a node in a preset project knowledge graph to obtain the association and impact information of each structured risk event object in the project, and the risk level of each structured risk event object is determined. Based on the risk rating results of each structured risk event object and the associated impact information, visualized risk information for the project is generated.

[0006] In one possible implementation, the deep semantic model identifies whether the standardized text fragment is a candidate risky text, including: The deep semantic model extracts the semantic features of the standardized text fragment and outputs the first probability that the standardized text fragment belongs to risky text based on the semantic features; The deep semantic model performs keyword matching on the standardized text fragments based on a preset risk keyword dictionary, and calculates the keyword matching score as the second probability that the standardized text fragments belong to risky texts. The first probability and the second probability are weighted and fused to obtain the final risk score of the standardized text fragment; The standardized text fragment is determined as a candidate risk text based on the final risk score.

[0007] In one possible implementation, the weights of the weighted fusion are adjusted according to the text length of the standardized text fragment: when the text length is greater than a preset threshold, the weight of the first probability increases; when the text length is less than or equal to the preset threshold, the weight of the second probability increases.

[0008] In one possible implementation, extracting structured risk event objects from the identified candidate risk texts includes: Named entity recognition is performed on the identified candidate risk texts to obtain the entities therein; Relation extraction is performed on the identified candidate risk texts to obtain the semantic relationships between entities; The identified entities and their semantic relationships are combined to construct a structured risk event object corresponding to the candidate risk text.

[0009] In one possible implementation, associating each of the structured risk event objects with nodes in a preset project knowledge graph to obtain the association impact information of each structured risk event object in the project includes: Extract the entities from the structured risk event object; The extracted entities are matched with nodes in the project knowledge graph to obtain the matching nodes of the entities in the project knowledge graph; The information corresponding to other nodes in the project knowledge graph that are related to the matching node is determined as the associated impact information of the structured risk event object in the project.

[0010] In one possible implementation, the risk rating of each of the structured risk event objects includes: The risk type of the structured risk event object is determined according to the preset risk classification rules; Obtain multidimensional risk assessment indicators for the structured risk event object; Calculate the risk severity score based on the risk type and the multidimensional risk assessment indicators; Based on the risk severity score, the structured risk event objects are classified into corresponding risk levels.

[0011] In one possible implementation, the method further includes: Based on the generated visualized risk information, perform at least one of the following operations: Push early warning messages, locate and display risk nodes in the project knowledge graph, mark risk tasks in the project Gantt chart, and generate risk analysis reports.

[0012] Secondly, this application provides a project risk event identification device, the device comprising: The multi-source data acquisition module is used to acquire multi-source unstructured text of the project, and preprocess each unstructured text to obtain multiple standardized text fragments. The text semantic analysis module is used to input each of the standardized text fragments into a pre-trained deep semantic model, and the deep semantic model identifies whether the standardized text fragment is a candidate risk text. The risk event extraction module is used to extract structured risk event objects from the identified candidate risk texts; The knowledge graph alignment module is used to associate each of the structured risk event objects with nodes in a preset project knowledge graph to obtain the association and impact information of each structured risk event object in the project. The risk rating module is used to rate the risk of each of the structured risk event objects. The output and visualization module is used to generate visualized risk information for the project based on the risk rating results of each structured risk event object and the associated impact information.

[0013] In one possible implementation, the deep semantic model identifies whether the standardized text fragment is a candidate risky text, including: The deep semantic model extracts the semantic features of the standardized text fragment and outputs the first probability that the standardized text fragment belongs to risky text based on the semantic features; The deep semantic model performs keyword matching on the standardized text fragments based on a preset risk keyword dictionary, and calculates the keyword matching score as the second probability that the standardized text fragments belong to risky texts. The first probability and the second probability are weighted and fused to obtain the final risk score of the standardized text fragment; The standardized text fragment is determined as a candidate risk text based on the final risk score.

[0014] In one possible implementation, the weights of the weighted fusion are adjusted according to the text length of the standardized text fragment: when the text length is greater than a preset threshold, the weight of the first probability increases; when the text length is less than or equal to the preset threshold, the weight of the second probability increases.

[0015] In one possible implementation, the risk event extraction module includes: An entity-relationship identification unit is used to perform named entity recognition on the identified candidate risk text to obtain the entities therein; Relation extraction is performed on the identified candidate risk texts to obtain the semantic relationships between entities; The entity-relationship combination unit is used to combine the identified entities and the semantic relationships between entities to construct the structured risk event object corresponding to the candidate risk text.

[0016] In one possible implementation, the knowledge graph alignment module includes: An entity-graph matching unit is used to extract entities from the structured risk event object; and to match the extracted entities with nodes in the project knowledge graph to obtain matching nodes of the entity in the project knowledge graph. The information association unit is used to determine the information corresponding to other nodes in the project knowledge graph that have an association relationship with the matching node as the association impact information of the structured risk event object in the project.

[0017] In one possible implementation, the risk rating module includes: The risk type identification unit is used to determine the risk type of the structured risk event object according to a preset risk classification rule; The assessment index acquisition unit is used to acquire multi-dimensional risk assessment indicators for the structured risk event object. The calculation unit is used to calculate the risk severity score based on the risk type and the multidimensional risk assessment index. The risk classification unit is used to classify the structured risk event object into the corresponding risk level based on the risk severity score.

[0018] In one possible implementation, the device further includes: The operation module is configured to perform at least one of the following operations based on the generated visualized risk information: Push early warning messages, locate and display risk nodes in the project knowledge graph, mark risk tasks in the project Gantt chart, and generate risk analysis reports.

[0019] Thirdly, this application provides an electronic device, including: a processor and a memory, wherein the processor is configured to execute a project risk event identification program stored in the memory to implement the project risk event identification method described in any one of the first aspects.

[0020] Fourthly, this application provides a storage medium storing one or more programs that can be executed by one or more processors to implement the project risk event identification method described in any one aspect.

[0021] Compared with existing technologies, the technical solution provided in this application has the following advantages: First, the method provided in this application obtains standardized text fragments by acquiring and preprocessing multi-source unstructured text from the project. Compared to manually reading scattered documents such as meeting minutes, daily reports, and emails one by one, this process achieves automatic access and standardization of massive heterogeneous data, eliminating analytical obstacles caused by format differences, laying a data foundation for subsequent automated analysis, and thus significantly improving the comprehensiveness and efficiency of risk information collection. Second, by inputting the standardized text fragments into a pre-trained deep semantic model, the model automatically identifies candidate risk texts. Unlike existing rule matching methods that rely on predefined keywords or regular expressions, the deep semantic model can understand contextual semantics and capture implicit risk expressions across sentences and paragraphs, effectively detecting risks even if the risk description does not contain preset keywords. Therefore, this solution significantly reduces the probability of missed and false positives, improves the recall and accuracy of risk identification, and avoids the time-consuming and subjective biases of manual reading, achieving real-time and automatic initial screening of risk signals. Furthermore, this solution extracts structured risk event objects from candidate risk texts. Through structured extraction, natural language descriptions are converted into standardized objects containing risk elements. This solves the problems of fragmented risk information and difficulty in quantitative analysis in existing technologies, providing a structured data foundation for subsequent association and classification. Then, this solution associates structured risk event objects with nodes in the project knowledge graph to obtain the association and impact information of each risk event in the overall project and classifies its risk. The project knowledge graph models the dependencies and affiliations between tasks, personnel, and milestones. By mapping isolated risk events to graph nodes, the upstream tasks, downstream tasks, critical paths, and relevant responsible persons that the risk may affect can be automatically deduced. This overcomes the technical shortcomings of traditional methods that cannot perform risk association analysis across teams, tasks, and phases, achieving global visibility and quantitative assessment of risk impact. Simultaneously, risk classification provides unified classification and severity standards, solving the problems of lacking standardized classification methods and difficulty in quantitative comparison of risk management in existing technologies, providing an objective basis for decision-making. Finally, this solution generates visualized risk information for the project based on risk rating results and related impact information, enabling project managers to quickly identify high-risk items, understand risk transmission paths, and formulate countermeasures. This significantly shortens risk response time and improves the real-time nature, operability, and decision support capabilities of project risk management.

[0022] In summary, the technical solution of this application enables automatic and accurate identification and global analysis of project risk events. Compared with existing technologies, it has significant improvements in identification accuracy, cross-team, cross-task and cross-stage correlation capabilities, standardization, and visualization level. Attached Figure Description

[0023] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0026] Figure 1 A flowchart illustrating an embodiment of a project risk event identification method provided in this application; Figure 2 A flowchart illustrating an embodiment of another project risk event identification method provided in this application; Figure 3 A block diagram illustrating an embodiment of a project risk event identification device provided in this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] The following disclosure provides numerous different embodiments or examples for implementing various structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0029] To address the technical problems of low risk identification accuracy, high false negative and false positive rates in existing technologies, and the difficulty in achieving risk correlation analysis across teams, tasks, and stages, this application provides a project risk event identification method, device, electronic device, and storage medium that can achieve automatic and accurate identification and global analysis of project risk events. Compared with existing technologies, it has significant improvements in identification accuracy, cross-team, cross-task, and cross-stage correlation capabilities, standardization, and visualization level.

[0030] Figure 1 This is a flowchart illustrating an embodiment of a project risk event identification method provided in this application. Figure 1 As shown, the method includes the following steps: Step 101: Obtain the multi-source unstructured text of the project, preprocess each unstructured text separately to obtain multiple standardized text fragments.

[0031] In project management, a large amount of unstructured text is typically generated, including meeting minutes, daily reports, project documents, emails, etc. This unstructured text often contains signals of potential project risks, such as task delays, delivery conflicts, client objections, requirement changes, or quality issues.

[0032] To identify potential risk signals in the project, step 101 involves collecting various types of unstructured text (referred to as multi-source unstructured text) generated during project management, including but not limited to meeting minutes, daily and weekly reports, project documents, emails, and instant messages. For example, the executing entity in this embodiment obtains meeting minutes through an office system interface, project documents through a file management system (such as SharePoint or Git repositories), and emails periodically or event-driven via email protocols (IMAP / SMTP). After obtaining the multi-source unstructured text of the project, each unstructured text is preprocessed to obtain corresponding standardized text fragments for use by the deep semantic model in subsequent steps.

[0033] For example, the preprocessing process for unstructured text is as follows: First, convert unstructured text in different formats (such as Word, PDF, Excel, HTML, TXT) to plain text format. Specifically, parsing libraries such as Apache POI and PDFBox can be used for format conversion. For example, for unstructured text in PDF or Word formats, extract the main text; for Excel spreadsheets, extract cell text and convert it into continuous text according to row and column structure; for HTML emails, remove HTML tags, retaining only the main text and necessary attachments.

[0034] Text cleaning is then performed to remove text noise and formatting redundancy, ensuring the effectiveness of subsequent semantic analysis. Cleaning operations include: deleting blank lines and meaningless symbols (such as consecutive spaces and tabs); removing interfering information such as headers, footers, page numbers, and table border symbols; uniformly processing or marking non-Chinese or non-English characters; and retaining key business terms and project-related entities to ensure the integrity of risk signals. For example, regular expressions are used to match and replace noisy characters in the text, while various special symbols are uniformly mapped or directly removed. During the cleaning process, only Chinese, English, numbers, and key symbols related to the project are retained, thereby improving the text's standardization and the effectiveness of subsequent analysis.

[0035] Next, the cleaned text is segmented into sentences and paragraphs according to rules such as periods, line breaks, and bullet points to provide context fragments of reasonable length for the subsequent semantic model, and indexes are created for long paragraphs to ensure that the semantic model can handle context fragments of reasonable length.

[0036] Finally, all text is uniformly converted to UTF-8 encoding, and the processed standardized text fragments are stored in a cache or database for real-time use by subsequent modules. Before storage or caching, special characters are escaped to avoid semantic analysis errors and ensure cross-platform compatibility.

[0037] In step 101, the multi-source unstructured text of the project is obtained, and each unstructured text is preprocessed to obtain multiple standardized text fragments, which are the front-end foundation of this invention and provide standardized, reliable, and real-time input data for subsequent semantic analysis, keyword triggering, risk event extraction, and knowledge graph alignment.

[0038] Step 102: Input each standardized text fragment into the pre-trained deep semantic model, and let the deep semantic model identify whether the standardized text fragment is a candidate risk text.

[0039] Step 102 performs preliminary risk semantic identification. Specifically, each standardized text fragment obtained from preprocessing is used as input and fed into a pre-trained deep semantic model. This model can perform context-level semantic modeling of the text, understand the inherent logical relationships between words and sentences, and thus capture potential risk signals hidden in the text, such as expressions of task delays, resource conflicts, and changes in requirements. Unlike simple keyword matching, the deep semantic model has the ability to identify synonyms, cross-sentence dependencies, and implicit expressions, and can effectively distinguish risk descriptions from normal narratives. The model outputs a risk confidence score for each standardized text fragment, representing the probability that the fragment belongs to a risk event. When the confidence score exceeds a preset threshold, the text fragment is identified as a candidate risk text for further extraction and analysis; otherwise, it is marked as non-risk text and will not proceed to the next processing step.

[0040] This step enables rapid initial screening of massive amounts of project text, significantly reducing the amount of data required for subsequent refined analysis while ensuring recall rate, and improving the real-time processing efficiency of the overall system.

[0041] Step 103: Extract structured risk event objects from the identified candidate risk texts.

[0042] Step 103 involves fine-grained extraction of structured information from the selected candidate risk texts to form standardized structured risk event objects. For example, this object includes at least the risk type (e.g., delay tendency, delivery conflict, customer objection), associated task (task name or module), responsible party (personnel or team), time information (planned time / delay time), and risk description text, thereby achieving automatic conversion from unstructured text to structured data.

[0043] In one embodiment, an exemplary implementation of extracting structured risk event objects from identified candidate risk texts includes: performing named entity recognition on the identified candidate risk texts to obtain entities therein; extracting relationships from the identified candidate risk texts to obtain semantic relationships between entities; and combining the identified entities and the semantic relationships between entities to construct structured risk event objects corresponding to the candidate risk texts.

[0044] This embodiment combines named entity recognition and relation extraction, enabling the system to automatically capture key elements and their interrelationships in risk events, thereby constructing standardized risk event objects.

[0045] Specifically, named entity recognition employs a sequence labeling model using BERT (Bidirectional Encoder Representations from Transformers) + BiLSTM (Bidirectional Long Short-Term Memory) + CRF (Conditional Random Fields). Following the BIO labeling system, candidate risk texts are labeled word-by-word, extracting at least four predefined entity categories: project task (TASK), person (PERSON), time (TIME), and risk type (RISK_TYPE). For example, for the text "Function A may be delayed in delivery," the model can extract "Function A" as the task entity and "delayed delivery" as the risk type entity.

[0046] Relation extraction is implemented based on BERT's relation classification model. First, all possible entity pairs are constructed from the text. Each entity pair, along with the original text, is input into the model to determine if a predefined semantic relationship exists between them. In one example, four relation types are predefined for a project risk scenario: RESPONSIBLE_FOR (person responsible for task), indicating the attribution of responsibility between a person and a task; DELAY_OF (risk affects task), indicating that a certain type of risk has a negative impact on a task; BELONGS_TO (task belongs to project), indicating that a task belongs to a project; and CAUSE_OF (cause of risk), indicating the cause or source of a risk event. Using the relation classification model, the system can output a corresponding relation label and confidence score for each entity pair.

[0047] After completing entity identification and relationship extraction, the system categorizes the identified entities by type and explicitly links the relationships between them, combining them into structured risk event objects. These objects fully include fields such as risk type, associated task, responsible party, time information, and risk description text, providing a standardized data foundation for subsequent knowledge graph alignment and risk analysis.

[0048] This embodiment effectively overcomes the shortcomings of traditional methods in extracting complex semantic relationships and inability to perform cross-entity association analysis, and significantly improves the structuring level and computability of risk information.

[0049] In addition, in some alternative solutions, other sequence labeling models (such as BiLSTM-CRF, CNN-CRF, or pointer network-based methods) can be used to achieve entity recognition, and the relation extraction part can also be completed using rule-based methods, dependency parsing-based methods, or joint extraction models (such as end-to-end entity relation joint extraction models), thereby achieving the structured representation of risk events. This application does not limit this approach.

[0050] Step 104: Associate each structured risk event object with a node in the preset project knowledge graph to obtain the association and impact information of each structured risk event object in the project, and classify the risk of each structured risk event object.

[0051] Step 104 comprises two core tasks: knowledge graph association and risk rating. The knowledge graph association task aims to map structured risk event objects to the project knowledge graph, thereby obtaining information on the risk's global impact, including affected tasks, personnel, and milestones. The risk rating task involves standardizing and quantifying risk events, outputting comparable risk levels.

[0052] The two tasks are independent yet complementary, working together to ensure that risk events are not only semantically understood, but also given a specific position and weight in the overall project, providing a quantitative basis for project management decisions.

[0053] In one embodiment, an exemplary implementation of associating each structured risk event object with nodes in a preset project knowledge graph to obtain the association impact information of each structured risk event object in the project includes: extracting entities from the structured risk event objects; matching the extracted entities with nodes in the project knowledge graph to obtain matching nodes of the entities in the project knowledge graph; and determining the information corresponding to other nodes in the project knowledge graph that have an association relationship with the matching nodes as the association impact information of the structured risk event objects in the project.

[0054] In this embodiment, task entities, personnel entities, and time entities are first extracted from the structured risk event object. Then, the extracted entity names are matched against task nodes, personnel nodes, or milestone nodes in a pre-defined project knowledge graph. Matching methods can include exact matching (strings must match exactly), rule mapping (based on aliases or naming rules), or vector similarity matching. After a successful match, other nodes directly or indirectly associated with the matched node are obtained, such as upstream tasks, downstream dependent tasks, responsible teams, key milestones, and related resources. The project element information (task name, responsible person, deadline, dependencies, etc.) corresponding to these associated nodes is used as the "association impact information" of the risk event. For example, when a task is identified as having a delay risk, the knowledge graph can be used to associate it with subsequent tasks blocked by this task, other tasks sharing the same resources, and milestone nodes on the project's critical path, thereby clarifying the potential scope and transmission path of the risk.

[0055] This embodiment achieves the tracking of risk transmission paths and the quantification of impact scope in the overall project by matching risk event entities with project knowledge graph nodes and obtaining related node information. It overcomes the shortcomings of traditional methods, such as isolated risk information and difficulties in cross-team, cross-task, and cross-stage association. At the same time, it supports multiple matching methods, has good scalability and adaptability, and provides project managers with more accurate and interpretable risk assessment basis.

[0056] In one embodiment, an exemplary implementation of risk rating for each structured risk event object includes: determining the risk type of the structured risk event object according to a preset risk classification rule; obtaining multi-dimensional risk assessment indicators for the structured risk event object; calculating the risk severity score based on the risk type and the multi-dimensional risk assessment indicators; and classifying the structured risk event object into the corresponding risk level based on the risk severity score.

[0057] In this embodiment, the risk type is first determined according to a preset risk classification rule. The risk classification rule covers common project risk categories, including but not limited to: delay risk (such as schedule delays, milestone delays), conflict risk (such as resource contention, scheduling conflicts), objection risk (such as customer disapproval, internal disagreements), requirement change risk (such as scope creep, frequent adjustments), quality problem risk (such as high defect rate, test failures), communication problem risk (such as information asymmetry, delayed feedback), dependency risk (such as unavailable external interfaces, upstream blockages), resource problem risk (such as insufficient manpower, budget shortages), compliance or policy risk (such as regulatory restrictions, audit issues), etc.

[0058] For example, the preset risk classification rules are stored in the form of a knowledge base or configuration file. During classification, the system can map the "risk type" entity extracted from the structured risk event object, for example, mapping the entity value "unable to deliver on time" to "delay risk"; if the risk type entity is missing, the classification can be supplemented based on the keywords or semantic similarity of the risk description text.

[0059] Secondly, obtain multi-dimensional assessment indicators for the risk event, including: scope of impact, urgency, probability of occurrence, etc.

[0060] The scope of impact refers to the number of project elements that will be affected if the risk occurs. For example, by using the knowledge graph association results from step 104, the number of directly and indirectly affected tasks, milestones, and personnel or teams can be counted. A scope of impact scoring mapping table can be predefined: 1 point for affecting 1-2 tasks, 2 points for 3-5 tasks, and 3 points for more than 5 tasks; milestones are also graded by quantity.

[0061] The urgency level reflects the time-sensitive nature of risk response. It is calculated based on the difference between the time entity of the risk event (such as the planned delivery date or milestone deadline) and the current system time. The shorter the remaining time, the higher the urgency level. An example scoring rule is: 1 point for ≥7 days remaining, 2 points for 3-7 days remaining, and 3 points for <3 days remaining or already overdue.

[0062] The probability of occurrence refers to the likelihood of a risk occurring. The risk probability output from the deep semantic model in step 102 (a normalized value, ranging from 0 to 1) can be directly reused, or a Bayesian correction can be made by combining the occurrence frequency of similar risks in the past. An example scoring rule is: 1 point for probability < 0.3, 2 points for probability 0.3 to 0.7, and 3 points for probability > 0.7.

[0063] The raw values ​​of each indicator can be uniformly converted into standardized scores of 0-10 or 1-5 using linear interpolation, piecewise functions, or table lookup methods for subsequent calculations.

[0064] Then, based on the risk type and the above multidimensional indicators, the risk severity score is calculated. The calculation method can be weighted summation, table lookup mapping (combining the risk type and each indicator level into a key value, and directly reading the corresponding score from the preset risk score table), or rule-based reasoning (such as "if the urgency level is high and the probability of occurrence is high, then directly assign the highest score").

[0065] The weights of indicators can vary depending on the risk type. For example, for "delay risk," the weight for urgency can be set to 0.5, the weight for impact scope to 0.3, and the weight for probability of occurrence to 0.2; for "quality issues," the weight for impact scope is 0.2, the weight for urgency is 0.3, and the weight for probability of occurrence is 0.5. The weight configuration is preset by the system administrator based on project characteristics and supports dynamic adjustment. By supporting differentiated configuration of indicator weights according to risk type, the system takes into account the characteristic differences of different types of risks, improves the accuracy of the classification results and the project adaptability, and provides clear and reusable quantitative basis for risk ranking, early warning triggering, and decision support.

[0066] Finally, based on the calculated scores, the risk events are classified into preset risk levels.

[0067] The preset risk level thresholds can correspond to the score range. For example, if the score range is 0-10, then 0-3 points are set as low risk, 4-6 points as medium risk, and 7-10 points as high risk. A five-level classification can also be used: extremely low (0-2), low (3-4), medium (5-6), high (7-8), and extremely high (9-10). The thresholds can be adjusted through a configuration file according to the actual needs of the project.

[0068] The final output risk level and risk type together constitute a complete rating result, which is stored in a structured risk event object for subsequent risk sorting, early warning triggering (such as immediately pushing emails or instant messages for high-risk events), and visualization (such as different colors marking).

[0069] This embodiment achieves a standardized transformation of risk events from subjective descriptions to objectively comparable levels by combining preset classification rules with multi-dimensional index quantification, overcoming the shortcomings of traditional methods that lack unified classification standards and are difficult to quantify and manage.

[0070] Furthermore, step 104, through the collaborative processing of graph association and risk rating, realizes the transformation from scattered risk information to globally quantifiable risk assets, laying the data foundation for the visualization output and decision support in step 105.

[0071] Step 105: Generate visualized risk information for the project based on the risk rating results and associated impact information of each structured risk event object.

[0072] Step 105 transforms the aforementioned analysis results into visualized risk information that project managers can intuitively view, completing the closed loop from massive amounts of unstructured text to actionable risk intelligence. For example, the system generates visualized risk information, including a list of risk events, a knowledge graph view, and a Gantt chart view, based on the risk rating (including risk type and severity level) and related impact information (such as affected tasks, personnel, milestones, etc.) for each risk event. This information supports filtering by project, time, responsible person, and other dimensions, providing a data foundation for subsequent risk display and operation.

[0073] In addition, in one embodiment, the method provided by this application further includes: performing at least one of the following operations based on the generated visualized risk information: pushing early warning messages, locating and displaying risk nodes in the project knowledge graph, marking risk tasks in the project Gantt chart, and generating a risk analysis report.

[0074] This embodiment, based on the visualized risk information generated in step 105, further provides executable operations based on this information. For example, the system can automatically or in response to user instructions perform at least one of the following operations based on the generated visualized risk information: ① Push warning messages to preset project management personnel, which can be achieved through email, instant messaging, or system notifications, ensuring that key risks reach the responsible parties immediately; ② Locate and highlight (e.g., highlight, flash, or label icons) nodes related to risk events in the project knowledge graph interface, and simultaneously display their impact path, facilitating managers' intuitive understanding of the risk propagation chain; ③ Mark tasks with risks in the project Gantt chart (e.g., change color, add warning symbols, or draw delay impact lines), making the impact of schedule deviations immediately apparent; ④ Generate a structured risk analysis report, which can summarize risk events and their classification and related impact information by project, time, responsible person, etc., and supports export or archiving.

[0075] This implementation transforms static, visualized information into interactive risk management actions, enhancing the system's proactive early warning capabilities and user-friendliness. Push notifications shorten risk response time; the combined use of knowledge graphs and Gantt charts reduces the cognitive burden of searching for information across different views; and report generation provides standardized documentation for project reviews and audits. These operations can be performed individually or in combination, fully adapting to the diverse needs of different project management scenarios and further improving the closed-loop process from risk identification to risk response.

[0076] The technical solution provided in this application first obtains standardized text fragments by acquiring and preprocessing multi-source unstructured text from projects. Compared to manually reading scattered documents such as meeting minutes, daily reports, and emails one by one, this process achieves automatic access and standardization of massive heterogeneous data, eliminating analytical obstacles caused by format differences and laying a data foundation for subsequent automated analysis, thereby significantly improving the comprehensiveness and efficiency of risk information collection. Second, by inputting the standardized text fragments into a pre-trained deep semantic model, the model automatically identifies candidate risk texts. Unlike existing rule matching methods that rely on predefined keywords or regular expressions, the deep semantic model can understand contextual semantics and capture implicit risk expressions across sentences and paragraphs, effectively detecting risks even if the risk description does not contain preset keywords. Therefore, this solution significantly reduces the probability of missed and false positives, improves the recall and accuracy of risk identification, and avoids the time-consuming and subjective biases of manual reading, achieving real-time and automatic initial screening of risk signals. Furthermore, this solution extracts structured risk event objects from candidate risk texts. Through structured extraction, natural language descriptions are converted into standardized objects containing risk elements. This solves the problems of fragmented risk information and difficulty in quantitative analysis in existing technologies, providing a structured data foundation for subsequent association and classification. Then, this solution associates structured risk event objects with nodes in the project knowledge graph to obtain the association and impact information of each risk event in the overall project and classifies its risk. The project knowledge graph models the dependencies and affiliations between tasks, personnel, and milestones. By mapping isolated risk events to graph nodes, the upstream tasks, downstream tasks, critical paths, and relevant responsible persons that the risk may affect can be automatically deduced. This overcomes the technical shortcomings of traditional methods that cannot perform risk association analysis across teams, tasks, and phases, achieving global visibility and quantitative assessment of risk impact. Simultaneously, risk classification provides unified classification and severity standards, solving the problems of lacking standardized classification methods and difficulty in quantitative comparison of risk management in existing technologies, providing an objective basis for decision-making. Finally, this solution generates visualized risk information for the project based on risk rating results and related impact information, enabling project managers to quickly identify high-risk items, understand risk transmission paths, and formulate countermeasures. This significantly shortens risk response time and improves the real-time nature, operability, and decision support capabilities of project risk management.

[0077] In summary, the technical solution of this application enables automatic and accurate identification and global analysis of project risk events. Compared with existing technologies, it can achieve stronger semantic understanding capabilities, lower false negative rates, and higher structuring levels while ensuring a high identification accuracy rate. It also significantly improves the real-time performance and scalability of risk identification, thereby providing more accurate and quantifiable decision support for project management.

[0078] Figure 2 A flowchart illustrating an embodiment of another project risk event identification method provided in this application. Figure 2 exist Figure 1 Based on the illustrated process, an exemplary implementation of a deep semantic model for identifying whether a standardized text fragment is a candidate risky text is described. For example... Figure 2 As shown, the method includes the following steps: Step 201: The deep semantic model extracts the semantic features of the standardized text fragments and outputs the first probability that the standardized text fragments belong to risky texts based on the semantic features.

[0079] Step 202: The deep semantic model performs keyword matching on standardized text fragments based on a pre-set risk keyword dictionary, and calculates the keyword matching score as the second probability that the standardized text fragment belongs to risk text.

[0080] Step 203: Weight and fuse the first probability and the second probability to obtain the final risk score of the standardized text fragment.

[0081] Step 204: Determine whether the standardized text fragment is a candidate risk text based on the final risk score.

[0082] For ease of understanding, steps 201 to 204 are explained uniformly below: In summary, the core of the risk candidate text identification process, comprising steps 201 to 204, lies in two aspects: First, using a deep semantic model to perform contextual semantic modeling on standardized text fragments, outputting the first probability that the fragment belongs to risk text, thereby capturing implicit and complex risk semantic expressions; second, based on a pre-set risk keyword dictionary, performing explicit keyword matching to calculate a second probability, thereby enhancing sensitivity to common and explicit risk signals. Finally, the two probabilities are weighted and fused to obtain a comprehensive risk score (i.e., the final risk score), and a preset threshold is used to determine whether the fragment is a candidate risk text. This fusion strategy balances the depth of semantic understanding with the timeliness of keyword matching, effectively improving the recall and accuracy of risk identification.

[0083] Specifically, in step 201, the deep semantic model extracts the semantic features of the standardized text fragments and outputs the first probability that the standardized text fragments belong to risky text based on the semantic features.

[0084] This step uses a deep semantic model based on the Transformer architecture (such as BERT, a Transformer model represented by a bidirectional encoder) to encode the input text. First, the standardized text fragments are segmented and embedded, generating a combination of three types of embedding vectors: Token Embedding (word vectors, representing the semantics of each word), Segment Embedding (sentence discrimination vectors, used to distinguish different sentences), and Position Embedding (position encoding vectors, representing the position of words in the sequence). The superposition of these three forms the model input sequence, represented as: X=[CLS] Token1 Token2... TokenN [SEP], where [CLS] is the classification identifier and [SEP] is the sentence separator.

[0085] The input sequence is computed through a multi-layer Transformer Encoder, with a multi-head self-attention mechanism at its core. For an input matrix X, the query matrix Q, key matrix K, and value matrix V are computed separately, and then the semantic dependencies across words and sentences in the text are captured by Attention(Q,K,V)=softmax(QK^T / √dk)V. The multi-head attention mechanism executes multiple sets of the above computations in parallel, enabling the model to simultaneously focus on the associated features of different semantic spaces (such as time, task, responsible party, etc.).

[0086] Finally, the output vector h_cls=BERT(X)[CLS] at the [CLS] position is extracted as the semantic representation of the entire text segment. This vector is input into a fully connected layer and subjected to Softmax normalization, outputting the probability distribution of the segment belonging to each preset risk category (such as delay, conflict, objection, etc.) and non-risk category. The probability corresponding to the risk category is taken as the first probability, denoted as Score_BERT.

[0087] In step 202, the deep semantic model performs keyword matching on the standardized text fragments based on a preset risk keyword dictionary, and calculates the keyword matching score as the second probability that the standardized text fragment belongs to the risk text.

[0088] This step executes the keyword-triggered enhancement mechanism in parallel. A keyword dictionary covering common project risk scenarios is pre-built, organized by risk type, with each category containing a set of keywords and their synonyms. For example, the keyword dictionary includes, but is not limited to, the following categories and entries: - Delayed categories: Delay, postponement, failure to meet deadline, lag in progress, postponement, failure to meet plan, delayed delivery, extended cycle, slower progress, milestone delay; - Conflict-related issues: conflict, insufficient resources, scheduling conflicts, resource contention, personnel conflicts, time conflicts, priority conflicts, dependency conflicts, task overlap, resource shortages; - Objections: objection, disapproval, need for adjustment, disagreement, differing opinions, disagreement, questioning, controversy, feedback issues, need for reassessment; - Requirement Change Category: Requirement changes, requirement adjustments, frequent changes, scope changes, unclear requirements, unstable requirements, new requirements, and deleted requirements; - Quality issues: quality problems, defects, bugs, errors, anomalies, instability, performance issues, non-compliance, rework, and failed tests; - Communication problems: poor communication, information asymmetry, lack of synchronization, misunderstanding, untimely feedback, difficulty in alignment, and high communication costs; - Dependency risk category: Dependency risk, external dependency, third-party issues, interface dependency, supplier delay, upstream blockage, downstream unpreparedness; - Resource-related issues: insufficient manpower, staff turnover, resource shortage, lack of manpower, absence of key personnel, unavailability of resources, insufficient budget; - Schedule risks: schedule risks, delays, postponements, incomplete tasks, slow progress, and slower-than-expected progress; - Compliance / Policy Risks: Compliance risks, policy restrictions, regulatory requirements, non-compliance, audit issues, and legal risks.

[0089] For each standardized text fragment, iterate through all entries in the keyword dictionary, and count the matched keywords and their preset weights (keywords within the same risk category can be assigned the same weight, or differentiated based on importance). The formula for calculating the second probability (denoted as Score_Keyword) is: Score_Keyword = Σ(wi × hit_i) Where wi is the keyword weight, and hit_i is a 0 / 1 indicator variable, indicating whether a keyword was hit.

[0090] If no keyword is matched, then Score_Keyword = 0.

[0091] Optionally, the Score_Keyword can be normalized to the same numerical range as the first probability (e.g., 0~1) for subsequent fusion.

[0092] In step 203, the first probability and the second probability are weighted and fused to obtain the final risk score (denoted as FinalScore) of the standardized text fragment.

[0093] This step performs a weighted linear fusion of the two probabilities. The basic fusion formula is: FinalScore = α × Score_BERT + β × Score_Keyword Here, α and β are weighting coefficients, satisfying α+β=1.

[0094] In one embodiment, to adapt to different text features, the weight coefficients can be dynamically adjusted according to the length of the standardized text fragment. Specifically, a text length threshold is set (e.g., 512 characters or 100 words). When the text length is greater than the threshold, the text contains more complex contextual information, and the semantic model's understanding ability is more critical; therefore, α (semantic model weight) is increased, and β is decreased accordingly. When the text length is less than or equal to the threshold, the text is shorter, and the accuracy and efficiency of keyword matching are higher; therefore, β is increased, and α is decreased accordingly.

[0095] This fusion mechanism retains the deep semantic model's ability to deeply understand long texts and complex sentences, while also leveraging the rapid response advantage of keyword triggers in short texts and explicit signals.

[0096] Finally, in step 204, the standardized text fragment is determined as a candidate risk text based on the final risk score.

[0097] The final risk score calculated in step 203 is compared with a preset risk threshold (denoted as T, for example, T = 0.5, which can be adjusted according to the actual application scenario). If FinalScore ≥ T, the standardized text fragment is determined to be a candidate risk text and enters the subsequent structured extraction and analysis process; otherwise, it is determined to be non-risk text and is not processed further. This step enables rapid filtering of massive text fragments, concentrating limited computing power on potentially risky texts and improving the overall system processing efficiency.

[0098] Through the synergistic effect of steps 201 to 204, this technical solution achieves a good balance between semantic understanding capability and keyword response speed, significantly improving the accuracy and robustness of project risk text recognition.

[0099] Furthermore, regarding semantic analysis models, other pre-trained language models with contextual semantic understanding capabilities can be used as alternatives in practical applications. These include improved models based on the Transformer structure (such as RoBERTa, ERNIE, ALBERT, etc.) or other deep learning language models (such as GPT-like models, Bi-directional Long Short-Term Memory networks (BiLSTM) combined with attention mechanisms, etc.). These models can also achieve deep representation of text semantics and risk semantic recognition. This application does not impose any limitations on these aspects.

[0100] Similarly, regarding keyword triggering mechanisms, in some alternative implementations, methods based on statistical features (such as TF-IDF keyword extraction), semantic matching methods based on word vector similarity, or pattern matching methods based on rule engines (such as regular expressions or template matching) can also be used. All of these methods can achieve the detection and triggering of explicit risk signals, and are equivalent extensions of the keyword triggering mechanism. This application does not impose any limitations on these methods.

[0101] Figure 3 This is a block diagram illustrating an embodiment of a project risk event identification device provided in this application. Figure 3 As shown, the device includes: The multi-source data acquisition module 31 is used to acquire multi-source unstructured text of the project, and preprocess each unstructured text to obtain multiple standardized text fragments. The text semantic analysis module 32 is used to input each of the standardized text fragments into a pre-trained deep semantic model, and the deep semantic model identifies whether the standardized text fragments are candidate risk texts. Risk event extraction module 33 is used to extract structured risk event objects from the identified candidate risk texts; The knowledge graph alignment module 34 is used to associate each of the structured risk event objects with nodes in the preset project knowledge graph to obtain the association and impact information of each of the structured risk event objects in the project. Risk rating module 35 is used to rate the risk of each of the structured risk event objects. The output and visualization module 36 is used to generate visualized risk information for the project based on the risk rating results of each structured risk event object and the associated impact information.

[0102] In one possible implementation, the deep semantic model identifies whether the standardized text fragment is a candidate risky text, including: The deep semantic model extracts the semantic features of the standardized text fragment and outputs the first probability that the standardized text fragment belongs to risky text based on the semantic features; The deep semantic model performs keyword matching on the standardized text fragments based on a preset risk keyword dictionary, and calculates the keyword matching score as the second probability that the standardized text fragments belong to risky texts. The first probability and the second probability are weighted and fused to obtain the final risk score of the standardized text fragment; The standardized text fragment is determined as a candidate risk text based on the final risk score.

[0103] In one possible implementation, the weights of the weighted fusion are adjusted according to the text length of the standardized text fragment: when the text length is greater than a preset threshold, the weight of the first probability increases; when the text length is less than or equal to the preset threshold, the weight of the second probability increases.

[0104] In one possible implementation, the risk event extraction module 33 includes: An entity-relationship identification unit is used to perform named entity recognition on the identified candidate risk text to obtain the entities therein; Relation extraction is performed on the identified candidate risk texts to obtain the semantic relationships between entities; The entity-relationship combination unit is used to combine the identified entities and the semantic relationships between entities to construct the structured risk event object corresponding to the candidate risk text.

[0105] In one possible implementation, the knowledge graph alignment module 34 includes: An entity-graph matching unit is used to extract entities from the structured risk event object; and to match the extracted entities with nodes in the project knowledge graph to obtain matching nodes of the entity in the project knowledge graph. The information association unit is used to determine the information corresponding to other nodes in the project knowledge graph that have an association relationship with the matching node as the association impact information of the structured risk event object in the project.

[0106] In one possible implementation, the risk rating module 35 includes: The risk type identification unit is used to determine the risk type of the structured risk event object according to a preset risk classification rule; The assessment index acquisition unit is used to acquire multi-dimensional risk assessment indicators for the structured risk event object. The calculation unit is used to calculate the risk severity score based on the risk type and the multidimensional risk assessment index. The risk classification unit is used to classify the structured risk event object into the corresponding risk level based on the risk severity score.

[0107] In one possible implementation, the device further includes: The operation module is configured to perform at least one of the following operations based on the generated visualized risk information: Push early warning messages, locate and display risk nodes in the project knowledge graph, mark risk tasks in the project Gantt chart, and generate risk analysis reports.

[0108] like Figure 4As shown in the figure, this application provides an electronic device, including a processor 111, a communication interface 112, a memory 113, and a communication bus 114, wherein the processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114. Memory 113 is used to store computer programs; In one embodiment of this application, when the processor 111 executes the program stored in the memory 113, it implements the project risk event identification method provided in any of the foregoing method embodiments, including: Obtain multi-source unstructured text of the project, preprocess each unstructured text separately to obtain multiple standardized text fragments; Each of the standardized text fragments is input into a pre-trained deep semantic model, which then identifies whether the standardized text fragment is a candidate risk text. Extract structured risk event objects from the identified candidate risk texts; Each of the structured risk event objects is associated with a node in a preset project knowledge graph to obtain the association and impact information of each structured risk event object in the project, and the risk level of each structured risk event object is determined. Based on the risk rating results of each structured risk event object and the associated impact information, visualized risk information for the project is generated.

[0109] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the project risk event identification method provided in any of the foregoing method embodiments.

[0110] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0111] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0112] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0113] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for identifying project risk events, characterized in that, The method includes: Obtain multi-source unstructured text of the project, preprocess each unstructured text separately to obtain multiple standardized text fragments; Each of the standardized text fragments is input into a pre-trained deep semantic model, which then identifies whether the standardized text fragment is a candidate risk text. Extract structured risk event objects from the identified candidate risk texts; Each structured risk event object is associated with a node in a preset project knowledge graph to obtain the association and impact information of each structured risk event object in the project, and the risk level of each structured risk event object is determined based on the association and impact information; the project knowledge graph contains multiple nodes for representing the global structure of the project and the dependencies between the nodes. Based on the risk rating results of each structured risk event object and the associated impact information, visualized risk information for the project is generated; The step of associating each structured risk event object with a node in a preset project knowledge graph to obtain the association impact information of each structured risk event object in the project includes: extracting entities from the structured risk event objects; matching the extracted entities with nodes in the project knowledge graph to obtain matching nodes of the entities in the project knowledge graph; and determining the project element information corresponding to other nodes in the project knowledge graph that have an association relationship with the matching nodes as the association impact information of the structured risk event objects in the project.

2. The method according to claim 1, characterized in that, The deep semantic model identifies whether the standardized text fragment is a candidate risk text, including: The deep semantic model extracts the semantic features of the standardized text fragment and outputs the first probability that the standardized text fragment belongs to risky text based on the semantic features; The deep semantic model performs keyword matching on the standardized text fragments based on a preset risk keyword dictionary, and calculates the keyword matching score as the second probability that the standardized text fragments belong to risky texts. The first probability and the second probability are weighted and fused to obtain the final risk score of the standardized text fragment; The standardized text fragment is determined as a candidate risk text based on the final risk score.

3. The method according to claim 2, characterized in that, The weights of the weighted fusion are adjusted according to the text length of the standardized text fragment: when the text length is greater than a preset threshold, the weight of the first probability increases; when the text length is less than or equal to the preset threshold, the weight of the second probability increases.

4. The method according to claim 1, characterized in that, The step of extracting structured risk event objects from the identified candidate risk texts includes: Named entity recognition is performed on the identified candidate risk texts to obtain the entities therein; Relation extraction is performed on the identified candidate risk texts to obtain the semantic relationships between entities; The identified entities and their semantic relationships are combined to construct a structured risk event object corresponding to the candidate risk text.

5. The method according to claim 1, characterized in that, The risk rating of each of the structured risk event objects includes: The risk type of the structured risk event object is determined according to the preset risk classification rules; Obtain multidimensional risk assessment indicators for the structured risk event object; Calculate the risk severity score based on the risk type and the multidimensional risk assessment indicators; Based on the risk severity score, the structured risk event objects are classified into corresponding risk levels.

6. The method according to claim 1, characterized in that, The method further includes: Based on the generated visualized risk information, perform at least one of the following operations: Push early warning messages, locate and display risk nodes in the project knowledge graph, mark risk tasks in the project Gantt chart, and generate risk analysis reports.

7. A project risk event identification device, characterized in that, The device includes: The multi-source data acquisition module is used to acquire multi-source unstructured text of the project, and preprocess each unstructured text to obtain multiple standardized text fragments. The text semantic analysis module is used to input each of the standardized text fragments into a pre-trained deep semantic model, and the deep semantic model identifies whether the standardized text fragment is a candidate risk text. The risk event extraction module is used to extract structured risk event objects from the identified candidate risk texts; The knowledge graph alignment module is used to associate each of the structured risk event objects with nodes in a preset project knowledge graph to obtain the association and impact information of each structured risk event object in the project; the project knowledge graph contains multiple nodes that represent the global structure of the project and the dependencies between the nodes. The risk rating module is used to rate the risk of each structured risk event object based on the associated impact information. The output and visualization module is used to generate visualized risk information for the project based on the risk rating results of each structured risk event object and the associated impact information. The step of associating each structured risk event object with a node in a preset project knowledge graph to obtain the association impact information of each structured risk event object in the project includes: extracting entities from the structured risk event objects; matching the extracted entities with nodes in the project knowledge graph to obtain matching nodes of the entities in the project knowledge graph; and determining the project element information corresponding to other nodes in the project knowledge graph that have an association relationship with the matching nodes as the association impact information of the structured risk event objects in the project.

8. An electronic device, characterized in that, include: A processor and a memory, the processor being configured to execute a project risk event identification program stored in the memory to implement the project risk event identification method according to any one of claims 1-6.

9. A storage medium, characterized in that, The storage medium stores one or more programs, which can be executed by one or more processors to implement the project risk event identification method according to any one of claims 1-6.

Images