A method and system for risk assessment of engineering investment projects based on inference models

Abstract

This application relates to the field of risk assessment for engineering investment projects, specifically a method and system for risk assessment of engineering investment projects based on an inference model. It includes: acquiring data elements to be assessed from external data sources of the engineering investment project, obtaining their current behavioral performance, and establishing stable reference performance; comparing the current behavioral performance with the stable reference performance to obtain the degree of semantic deviation; when preset deviation conditions are met, generating and sending semantic correction suggestions based on external supporting data; receiving adjustment results, generating or updating semantic mapping rules; performing semantic transformation according to the semantic mapping rules to generate an internal semantic representation; and inputting the internal semantic representation into the inference model to identify and output the risk transmission path. This solves the problem that existing technologies, when processing external data, struggle to understand and dynamically adapt to changes in the deep meaning of the data in real time, leading to semantic bias in the input data of the inference model, which in turn affects the reliability of risk prediction and the estimation of long-term profitability.

Description

Technical Field

[0001] This application relates to the field of risk assessment for engineering investment projects, and more specifically, to a method and system for risk assessment of engineering investment projects based on a reasoning model. Background Technology

[0002] In engineering investment projects, especially those with long construction periods, large investment scales, and continuous influence from external factors such as policies, supply chains, and market environments, the accuracy of risk assessment results directly affects investment decisions, resource allocation, and long-term return assessments. Existing engineering investment project risk assessment systems typically rely on data preprocessing programs to clean, standardize, and format external data before inputting the processed data into inference models for risk identification and prediction. However, traditional preprocessing programs primarily focus on data format, field structure, numerical range, and syntax consistency, making it difficult to identify whether the actual business meaning behind external data fields, labels, or indicators has changed.

[0003] Because engineering investment projects have long cycles, the meaning of external data sources may gradually and implicitly change with policy adjustments, changes in regulatory interpretations, changes in supply chain rules, or updates to industry standards. For example, the labeling of green energy projects may remain unchanged in policy documents, but their carbon emission standards may have been raised; the field of material origin in supply chain reports may remain unchanged, but its definition may have changed from the final processing location to the raw material extraction location. These changes are often not accompanied by obvious changes in data format, causing existing preprocessing procedures to still process new data according to old rules.

[0004] In this scenario, while the data input into the inference model may superficially conform to the format requirements, its true meaning has already been distorted. As large amounts of data accumulate over long project cycles, this distortion will gradually affect the model's feature construction and risk causal relationship judgments. For example, if the meaning of indicators such as approval cycle and supply chain delivery cycle changes, the model may still construct internal features such as approval efficiency and supply stability based on the old meaning, thus forming a risk perception that is inconsistent with reality.

[0005] Furthermore, inference models may output seemingly stable and reliable risk assessment results that are actually biased, causing assessors to overemphasize traditional risk points while ignoring emerging risk transmission paths caused by changes in the meaning of data. Ultimately, investors may make incorrect judgments in contract negotiations, supply chain layout, financing arrangements, and resource allocation, resulting in economic losses. Therefore, existing technologies urgently need an improvement solution. Summary of the Invention

[0006] This application discloses a risk assessment method and system for engineering investment projects based on inference models. It aims to solve the problem that existing risk assessment methods for engineering investment projects are difficult to understand and dynamically adapt to changes in the deeper meaning of data in real time when processing external data, which leads to meaning bias in the input data of the inference model and thus affects the reliability of risk prediction and the estimation of long-term profitability.

[0007] The technical solution of this application is as follows: Firstly, this application discloses a risk assessment method for engineering investment projects based on a reasoning model, including: The data elements to be evaluated are obtained from external data sources of engineering investment projects. External data sources include at least one of policy documents, market reports, supply chain data, technology assessment documents and news texts. The data elements to be evaluated refer to fields, tags, text fragments or indicators extracted from external data sources for use in the risk assessment of engineering investment projects. Obtain the current behavior of the data element to be evaluated. The current behavior refers to the data change characteristics of the data element to be evaluated within the current time window. Establish stable reference performance for the data elements to be evaluated. Stable reference performance refers to the characteristics of change in reference data that the data elements to be evaluated exhibit within a preset stable reference time period, which are used to compare with the current behavioral performance. By comparing the current behavioral performance with the stable reference performance, the degree of deviation of the meaning of the data element to be evaluated can be obtained; When the degree of deviation of meaning meets the preset deviation conditions, meaning correction suggestions are generated and sent based on external supporting data related to the data element to be evaluated. External supporting data refers to data retrieved from external data sources that are used to explain or prove that the meaning of the data element to be evaluated has changed. Meaning correction suggestions refer to candidate explanations and corresponding correction rules used to indicate that the current meaning of the data element to be evaluated has changed relative to the stable reference performance. Receive the adjustment results returned based on the meaning correction suggestions, and generate or update the meaning mapping rules for the data elements to be evaluated based on the adjustment results. The meaning mapping rules are rules used to convert the data elements to be evaluated into internal semantic representations that can be processed by the inference model. Based on the meaning mapping rules, semantic transformation is performed on the data elements to be evaluated to generate internal semantic representations. Internal semantic representations refer to data representations formed before being input into the inference model, which include the semantic category, semantic weight, risk dimension, or applicable conditions of the data elements to be evaluated after meaning mapping. The internal semantic representation is input into the inference model that executes the preset risk inference rules, so that the inference model can construct internal features based on the internal semantic representation and identify the risk transmission path of the engineering investment project; Output the risk transmission path.

[0008] Secondly, this application also discloses a risk assessment system for engineering investment projects based on a reasoning model, including: The acquisition module is used to acquire data elements to be evaluated from external data sources of engineering investment projects. External data sources include at least one of policy documents, market reports, supply chain data, technology assessment documents, and news texts. Data elements to be evaluated refer to fields, tags, text fragments, or indicators extracted from external data sources for use in risk assessment of engineering investment projects. The behavior performance acquisition module is used to acquire the current behavior performance of the data element to be evaluated. The current behavior performance refers to the data change characteristics of the data element to be evaluated within the current time window. The reference establishment module is used to establish stable reference performance for the data element to be evaluated. Stable reference performance refers to the change characteristics of the reference data of the data element to be evaluated within a preset stable reference time period, which is used to compare with the current behavior performance. The meaning deviation quantification module is used to compare the current behavioral performance with the stable reference performance to obtain the degree of meaning deviation of the data element to be evaluated; The suggestion generation module is used to generate and send meaning correction suggestions based on external supporting data related to the data element to be evaluated when the degree of meaning deviation meets the preset deviation conditions. External supporting data refers to data retrieved from external data sources that is used to explain or prove that the meaning of the data element to be evaluated has changed. Meaning correction suggestions refer to candidate explanations and corresponding correction rules used to indicate that the current meaning of the data element to be evaluated has changed relative to the stable reference performance. The rule update module is used to receive the adjustment results returned based on the meaning correction suggestions, and generate or update the meaning mapping rules of the data elements to be evaluated based on the adjustment results. The meaning mapping rules are rules used to convert the data elements to be evaluated into internal semantic representations that can be processed by the inference model. The semantic transformation module is used to perform semantic transformation on the data elements to be evaluated according to the meaning mapping rules, and generate internal semantic representations. Internal semantic representations refer to data representations formed before being input into the inference model, which include the semantic category, semantic weight, risk dimension, or applicable conditions of the data elements to be evaluated after meaning mapping. The reasoning module is used to input the internal semantic representation into the reasoning model that executes the preset risk reasoning rules, so that the reasoning model can construct internal features based on the internal semantic representation and identify the risk transmission path of the engineering investment project; The output module is used to output the risk transmission path.

[0009] Beneficial Effects: The engineering investment project risk assessment method based on inference models disclosed in this application effectively solves the problem in existing technologies where implicit changes in the meaning of external data prevent the preprocessing process from detecting and correcting them, leading to accumulated semantic biases in the input data of the inference model. By monitoring and correcting the meaning of data in real time, this application ensures that the data input to the inference model is semantically consistent with the actual situation in the real world, avoiding the model from constructing internal features and identifying risk transmission paths based on incorrect semantics. Therefore, this application can significantly improve the reliability and accuracy of engineering investment project risk assessment, helping investors make more informed decisions and effectively avoid potential economic losses. Attached Figure Description

[0010] Figure 1 A flowchart illustrating a risk assessment method for engineering investment projects based on a reasoning model, provided in this application.

[0011] Figure 2 A flowchart of a risk assessment system for engineering investment projects based on a reasoning model, provided for this application.

[0012] In the diagram: 1. Acquisition module; 2. Behavioral performance acquisition module; 3. Reference establishment module; 4. Meaning deviation quantification module; 5. Suggestion generation module; 6. Rule update module; 7. Semantic conversion module; 8. Reasoning module; 9. Output module. Detailed Implementation

[0013] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0014] Reference Figure 1 This application proposes a risk assessment method for engineering investment projects based on a reasoning model, comprising: S1000: Obtain the data elements to be evaluated from the external data source of the engineering investment project; External data sources include at least one of policy documents, market reports, supply chain data, technology assessment documents, and news texts. The data elements to be assessed refer to fields, tags, text snippets, or indicators extracted from external data sources for use in risk assessment of engineering investment projects.

[0015] S2000: Obtain the current behavior of the data element to be evaluated; Current behavior refers to the data change characteristics of the data element to be evaluated within the current time window. The current time window can be preset according to the cycle of the engineering investment project, the update frequency of external data sources, or the risk assessment frequency.

[0016] S3000: Establish a stable reference performance for the data elements to be evaluated; Stable reference performance refers to the characteristics of change in reference data that the data element to be evaluated exhibits within a preset stable reference time period, which is used to compare with the current behavior performance. The preset stable reference time period can be pre-set based on the historical data stability of the engineering investment project, the update cycle of external data sources, or the project risk assessment cycle.

[0017] S4000: Compare the current behavioral performance with the stable reference performance to obtain the degree of deviation of the meaning of the data element to be evaluated; S5000: When the degree of deviation of meaning meets the preset deviation conditions, a meaning correction suggestion is generated and sent based on external supporting data related to the data element to be evaluated; External corroborating data refers to data retrieved from external data sources that is used to explain or prove changes in the meaning of the data element to be evaluated. Meaning correction suggestions refer to candidate explanations and corresponding correction rules used to indicate changes in the current meaning of the data element to be evaluated relative to the stable reference performance.

[0018] S6000: Receives the adjustment results returned based on the meaning correction suggestions, and generates or updates the meaning mapping rules for the data elements to be evaluated based on the adjustment results; Meaning mapping rules refer to the rules used to convert data elements to be evaluated into internal semantic representations that can be processed by the inference model. Adjustment results refer to the confirmation, selection, modification, deletion, or weight allocation results returned based on the meaning correction suggestions, which are used to generate or update the meaning mapping rules of the data elements to be evaluated.

[0019] S7000: Based on the meaning mapping rules, perform semantic transformation on the data elements to be evaluated to generate internal semantic representations; Internal semantic representation refers to the data representation formed before being input into the inference model, which includes the semantic category, semantic weight, risk dimension, or applicable conditions of the data element to be evaluated after meaning mapping. Applicable conditions refer to the data conditions, source conditions, time conditions, geographical conditions, or policy conditions used to determine whether the meaning mapping rules are applicable to the data element to be evaluated.

[0020] S8000: Input the internal semantic representation into the inference model that executes the preset risk inference rules, so that the inference model can construct internal features based on the internal semantic representation and identify the risk transmission path of the engineering investment project; S9000: Output risk transmission path.

[0021] To better understand the technical solutions proposed in this application, the key terms involved will be explained below.

[0022] External data sources refer to various external channels that provide original information for risk assessment of engineering investment projects, such as policy documents issued by the government, market reports issued by market research institutions, supply chain data provided by supply chain management platforms, technical assessment documents issued by professional institutions, and news texts published by various news media.

[0023] The data element to be evaluated refers to a specific information unit extracted from the aforementioned external data source and used directly for risk assessment. It can be a structured field (such as a numerical indicator), a semi-structured label, an unstructured text fragment, or a processed composite indicator.

[0024] Current behavior refers to the dynamic data change characteristics of the data element to be evaluated within the current time window of focus, such as the fluctuation trend of its value, the change in the frequency of keywords in the text content, and the increase or decrease of related entities.

[0025] Stable reference performance refers to the data change characteristics of the data element to be evaluated within a pre-defined, considered stable historical period. It serves as a benchmark for comparison with current behavioral performance in order to identify potential changes in meaning.

[0026] Meaning deviation refers to the degree of change in the meaning of the data element to be evaluated, which is quantified by comparing the current behavior with the stable reference behavior. Meaning deviation can be obtained from at least one of numerical distribution deviation, contextual semantic deviation, or source composition deviation.

[0027] External supporting data refers to auxiliary data retrieved from external data sources that can explain or prove a change in the meaning of the data element being evaluated. For example, if the meaning of a policy term changes, relevant policy interpretation documents, expert opinions in technical assessment documents, or expert comments in news texts can serve as external supporting data.

[0028] Meaning correction suggestions refer to a series of candidate interpretations and corresponding correction rules provided for data elements to be evaluated whose meanings have deviated, in order to guide the evaluation end or system to understand and process the new meanings.

[0029] The adjustment result refers to the confirmation, selection, modification, deletion, or weight allocation result returned based on the meaning correction suggestions, which is used to generate or update the meaning mapping rules of the data elements to be evaluated.

[0030] Meaning mapping rules refer to a set of rules that transform the original external representation of a data element to be evaluated into an internal semantic representation that can be processed by the inference model, in order to ensure that the meaning of the data is accurate and consistent before it enters the inference model.

[0031] The applicable conditions refer to the data conditions, source conditions, time conditions, geographical conditions, or policy conditions used to determine whether the meaning mapping rule is applicable to the data element to be evaluated.

[0032] Internal semantic representation refers to the structured data representation formed after the data elements to be evaluated are mapped with meaning before being input into the inference model. It contains information such as semantic category, semantic weight, risk dimension or applicable conditions, so that the inference model can understand its deeper meaning.

[0033] Pre-defined risk reasoning rules refer to a set of rules pre-configured in the reasoning model for identifying risk factors, the relationships between risk factors, and risk transmission paths based on internal semantic representations. Pre-defined risk reasoning rules include at least one of causal reasoning rules, probabilistic reasoning rules, expert rules, or graph reasoning rules.

[0034] Internal features refer to the feature data generated by the inference model based on internal semantic representation and used for risk inference calculation, including at least one of risk category features, risk intensity features, risk source features, risk association features, or risk impact features.

[0035] A risk transmission path refers to a directed chain of events from one risk factor to another. This directed chain includes the risk origin, at least one intermediate risk node, and the risk outcome, and is used to represent the risk transmission mechanism in engineering investment projects.

[0036] A reasoning model is a computational model that executes pre-defined risk reasoning rules. It can construct internal features based on internal semantic representations and identify the risk transmission path of engineering investment projects.

[0037] The core of the reasoning model-based risk assessment method for engineering investment projects proposed in this application lies in the dynamic perception, correction, and application of changes in the meaning of external data elements.

[0038] To acquire data elements to be evaluated from external data sources for engineering investment projects, a series of data collectors can be configured to periodically or in real-time scrape raw data from external data sources such as policy release platforms, market analysis agency websites, supply chain data platforms, technology assessment document libraries, and news media platforms. These data collectors can employ different collection strategies depending on the data source type (e.g., web pages, API interfaces, databases). For example, for policy documents, policy release sections on government websites can be scraped periodically; for market reports, data services from professional market research institutions can be subscribed to; and for news texts, APIs from mainstream news aggregation platforms can be accessed. After acquiring the raw data, preliminary cleaning and structuring processing can be performed. For example, through keyword matching and regular expression extraction, fields, tags, text fragments, or indicators related to the risk assessment of specific engineering investment projects can be filtered from massive amounts of data and used as data elements to be evaluated.

[0039] To obtain the current behavioral performance of the data elements to be evaluated, time series analysis can be performed on the acquired data elements. For example, for a numerical indicator (such as the market price of a certain raw material), its statistical characteristics such as average value, volatility, and growth rate within the current time window (such as the most recent week or month) can be calculated. For textual data (such as the frequency of occurrence of a keyword in a policy document), its occurrence count and contextual semantic changes within the current time window can be statistically analyzed. These statistical characteristics and semantic changes together constitute the current behavioral performance of the data element to be evaluated.

[0040] To establish a stable reference performance for the data element to be evaluated, a historical time period can be preset as the stable reference period, such as the past year or two. Within this time period, data analysis similar to obtaining current behavioral performance is performed on the data element to be evaluated to obtain its data change characteristics within this stable reference period. For example, the average and volatility of raw material prices over the past year can be calculated, or the average frequency and typical context of a certain keyword over the past year can be statistically analyzed. The stable reference performance serves as a benchmark to measure whether the current meaning has deviated.

[0041] To compare current behavioral performance with stable reference performance and determine the degree of semantic deviation of the data element being evaluated, various quantification methods can be employed. For example, for numerical indicators, the statistical distance (such as Euclidean distance or Mahalanobis distance) or relative rate of change between the current behavioral performance and the stable reference performance can be calculated. For textual data, word vector models (such as Word2Vec or BERT) can be used to calculate the cosine similarity between the semantic vectors of keywords within the current time window and the semantic vectors of keywords within the stable reference time period, or topic models (such as LDA) can be used to analyze the differences between the current topic distribution and the stable reference topic distribution. When these differences exceed a preset threshold, the meaning of the data element being evaluated can be considered to have deviated. The statistical distance, relative rate of change, cosine similarity, or topic distribution difference between the current behavioral performance and the stable reference performance can serve as the basis for calculating the degree of semantic deviation, and the specific quantification method used can be determined based on the data type of the data element being evaluated.

[0042] First, predefined deviation conditions are defined, such as when the deviation of an indicator exceeds two standard deviations, or when the text semantic similarity is below 0.7. Once these conditions are met, the system automatically triggers the retrieval of external supporting data. For example, if a green energy project deviates from its carbon emission standards, the system will retrieve the latest environmental policy documents, industry standard update notices, or expert interpretation articles from external data sources. Based on this external supporting data, the system can automatically generate multiple candidate semantic interpretations using natural language processing techniques (such as information extraction and summary generation), and provide corresponding correction rules for each interpretation. For example, a correction suggestion might be that the green energy project now needs to meet higher carbon emission standards, and the correction rule is to adjust the carbon emission coefficient in the relevant calculation model. These suggestions can be sent to the evaluation end in a structured form for review. The evaluation end can be a terminal, account, or application programming interface used to submit the adjustment results.

[0043] Once the assessment team receives meaning correction suggestions, it can evaluate these suggestions and select one or more of the most appropriate interpretations and their correction rules, or modify the suggestions themselves. For example, the assessment team might confirm a correction suggestion and assign it a risk influence weight. After receiving these adjustments, the system will generate or update the meaning mapping rules for the data elements to be evaluated based on these results. These rules are used to indicate how to convert a data element to be evaluated into an internal semantic representation that can be processed by the inference model when a specific meaning deviation occurs, such as mapping it to a new semantic category, assigning it a new semantic weight, or associating it with a new risk dimension.

[0044] In terms of semantic transformation of data elements to be evaluated and generation of internal semantic representations based on meaning mapping rules, once the meaning mapping rules are established or updated, the system will use these rules to perform real-time semantic transformation of the data elements to be evaluated. For example, when a new policy document is acquired that contains a keyword with a revised meaning, the system will convert it into an internal semantic representation containing new semantic categories, semantic weights, risk dimensions, or applicable conditions according to the corresponding meaning mapping rules. This internal semantic representation is structured and can be directly understood and processed by the inference model to ensure that the data input to the inference model is accurate and consistent at the semantic level.

[0045] In the process of inputting internal semantic representations into an inference model that executes pre-defined risk inference rules, enabling the inference model to construct internal features and identify risk transmission paths in engineering investment projects based on these internal semantic representations, the inference model, upon receiving the internal semantic representations, constructs internal features based on information such as semantic categories, semantic weights, and risk dimensions. For example, if the internal semantic representations indicate that a rise in the price of a certain raw material is due to new trade policies rather than changes in market supply and demand, the inference model will construct internal features related to policy risk, rather than simple market fluctuation features. Based on these internal features, the inference model executes pre-defined risk inference rules, such as using causal graphs, Bayesian networks, or expert systems, to identify potential risk transmission paths in engineering investment projects. For instance, how new trade policies can transmit rising raw material prices to increased project costs, thereby affecting project profitability.

[0046] Regarding the output of risk transmission paths, the risk transmission paths identified by the inference model can be output in the form of graphs, reports, or early warning information. These outputs are used to explain the source of risk, transmission mechanism, potential impact, and related key risk factors, providing a basis for subsequent adjustments to investment strategies and mitigation of potential risks.

[0047] This application further proposes S5000, including: S5100: When the degree of deviation of meaning meets the preset deviation conditions, multiple candidate semantic interpretations corresponding to the data element to be evaluated are extracted from external supporting data. Candidate semantic interpretations refer to different interpretation results formed based on different external supporting data regarding the current meaning of the data element to be evaluated. S5200: Generate multiple semantic scenario cards based on multiple candidate semantic interpretations. A semantic scenario card is a data object used to structurally carry a candidate semantic interpretation and its corresponding correction rules. A semantic scenario card includes a scenario name, policy basis, risk dimensions and impacts, supporting evidence, and corresponding correction rules. S5300: Generates and sends meaning correction suggestions containing multiple semantic scenario cards. The meaning correction suggestions are used to provide multiple candidate semantic interpretations and their corresponding correction rules to the evaluation end in a structured form. The S6000 includes: S6100: Receive the adjustment results of the meaning correction suggestions. The adjustment results include the risk influence weights assigned to multiple semantic scenario cards. The risk influence weights are quantitative values ​​used to represent the degree of influence of the corresponding semantic scenario card on the risk assessment results of the engineering investment project. S6200: Calculate the aggregate weight for each semantic scenario card based on the risk influence weight. The aggregate weight is the weight value obtained after statistical processing of one or more risk influence weights corresponding to the same semantic scenario card. S6300: Normalize the aggregate weights corresponding to each semantic scenario card to form a semantic scenario influence distribution. The semantic scenario influence distribution refers to the set of weights formed after normalizing the aggregate weights corresponding to multiple semantic scenario cards. S6400: Send the semantic context influence distribution and receive the distribution adjustment results returned for the semantic context influence distribution. The distribution adjustment results include at least one of the following: confirmation, modification, deletion or addition of the weight corresponding to at least one semantic context card in the semantic context influence distribution. Wherein, adding refers to adding a semantic context card and its corresponding weight, and deleting refers to deleting a semantic context card and its corresponding weight. S6500: Based on the distribution adjustment results, the semantic context influence distribution is updated to obtain the calibrated semantic context influence distribution. The calibrated semantic context influence distribution refers to the semantic context influence distribution corrected based on the distribution adjustment results. S6600: Based on multiple semantic scenario cards, the distribution of influence of calibrated semantic scenarios, and the applicable conditions corresponding to each semantic scenario card, generate or update the meaning mapping rules for the data elements to be evaluated. The applicable conditions refer to the data conditions, source conditions, time conditions, regional conditions, or policy conditions used to determine whether the data elements to be evaluated are applicable to the corresponding semantic scenario cards.

[0048] Specifically, when the deviation of the meaning of a data element to be evaluated meets a preset deviation condition, the system extracts multiple candidate semantic interpretations related to that data element from external supporting data. These candidate semantic interpretations represent various different interpretations of the potential change in the current meaning of the data element, each based on corresponding external supporting data. For example, an update to a policy document may cause a change in the meaning of a market indicator, and different market analysis reports may provide different interpretations of this change.

[0049] In one embodiment, the degree of meaning deviation can be obtained by weighting at least one of numerical distribution deviation, contextual semantic deviation, and source composition deviation, for example, D=aD1+bD2+cD3, where D represents the degree of meaning deviation, D1 represents the numerical distribution deviation, D2 represents the contextual semantic deviation, D3 represents the source composition deviation, a, b, and c represent preset weights, and a+b+c=1.

[0050] Furthermore, for each candidate semantic interpretation, the system generates a semantic scenario card. A semantic scenario card is a structured data object that carries a candidate semantic interpretation and its corresponding correction rules. Each semantic scenario card typically includes a scenario name, summarizing the scenario; policy basis, indicating the relevant policy documents or regulations supporting the interpretation; risk dimensions and impacts, describing the potential impact of the scenario on the engineering investment project across different risk dimensions (such as market risk and compliance risk); supporting evidence, providing specific external supporting data snippets or citations; and corresponding correction rules, explaining how to map the data elements to be evaluated to the new semantic representation.

[0051] Therefore, the system generates and sends meaning correction suggestions containing multiple semantic scenario cards. These suggestions present multiple possible meaning change scenarios in a structured manner, facilitating review by the evaluation team and the return of adjustment results.

[0052] Upon receiving the adjustment results regarding the proposed meaning revisions, these results may include risk impact weights assigned to each semantic scenario card. These risk impact weights represent the assessor's level of acceptance of the meaning changes represented by each semantic scenario card and their importance to the risk assessment.

[0053] Based on these risk influence weights, the system calculates the aggregate weight for each semantic scenario card. This aggregate weight is used to combine the risk influence weights corresponding to the same semantic scenario card. Subsequently, the aggregate weights corresponding to each semantic scenario card are normalized to form a semantic scenario influence distribution, which represents the relative importance of different semantic scenarios in the current context.

[0054] For further calibration, the system sends the semantic context influence distribution and receives the distribution adjustment results returned for that distribution. These adjustment results allow the evaluator to confirm, modify, delete, or add weights corresponding to at least one semantic context card in the semantic context influence distribution, thereby adjusting the assessment of the semantic context importance.

[0055] Based on the received distribution adjustment results, the system updates the semantic context influence distribution to obtain the calibrated semantic context influence distribution.

[0056] Finally, the system generates or updates the meaning mapping rules for the data elements to be evaluated based on multiple semantic scenario cards, the distribution of influence of calibrated semantic scenarios, and the applicable conditions corresponding to each semantic scenario card. The applicable conditions are used to determine whether the data element to be evaluated is applicable to the corresponding semantic scenario card. These conditions can include data conditions (such as data type and numerical range), source conditions (such as the data source institution), time conditions (such as the data release time), geographical conditions (such as the applicable geographical area), or policy conditions (such as the effective status of related policies). These applicable conditions are used to improve the accuracy and contextual adaptability of the meaning mapping rules.

[0057] In another embodiment of this application, S7000 is further proposed to include: S7100: Based on the identifier of the data element to be evaluated, query the meaning mapping rule corresponding to the data element to be evaluated; S7200: Reads multiple semantic scenario cards from the meaning mapping rules, calibrates the semantic scenario influence distribution, and the applicable conditions corresponding to each semantic scenario card; S7300: Based on the applicable conditions corresponding to each semantic scenario card, match the data content, source attribute, time attribute, regional attribute, or policy-related attribute of the data element to be evaluated, and determine at least one semantic scenario card that the data element to be evaluated matches. Policy association attribute refers to the association identifier or association category between the data element to be evaluated and external supporting data; data content refers to the field value, tag value, text content or indicator value of the data element to be evaluated; source attribute refers to the data provider, source type or issuing agency of the external data source corresponding to the data element to be evaluated; time attribute refers to the publication time, update time or collection time of the external data source corresponding to the data element to be evaluated; and geographic attribute refers to the country, region or regulatory jurisdiction involved in the external data source corresponding to the data element to be evaluated.

[0058] S7400: Extract the weight corresponding to at least one semantic scenario card from the calibration semantic scenario influence distribution; S7500: Generate a semantic probability distribution vector based on at least one semantic context card and its corresponding weight. The semantic probability distribution vector is a vector used to represent at least one semantic context card that the data element to be evaluated hits and the corresponding weight of each semantic context card. S7600: Combine the data content, semantic probability distribution vector, and risk dimensions and impacts of the data element to be evaluated with at least one semantic scenario card to generate an internal semantic representation; The S8000 includes: S8100: Inputs the internal semantic representation into the inference model that executes the preset risk inference rules; S8200: In the inference model, a corresponding risk assessment branch is configured for each semantic scenario card contained in the internal semantic representation. The risk assessment branch is the risk calculation path corresponding to a semantic scenario card in the inference model. S8300: Each risk assessment branch extracts the corresponding risk factors and correlation strength from the internal semantic representation based on its corresponding semantic scenario card. Risk factors refer to policy, market, supply chain, technology, financing or return factors that affect the risk assessment results of engineering investment projects. Correlation strength refers to the quantitative value of the strength of the influence relationship between two risk factors. S8400: Each risk assessment branch independently calculates the risk assessment result under its corresponding semantic scenario card based on the extracted risk factors and correlation strength. The risk assessment result includes at least one of the following: the probability of occurrence of risk events, the intensity of the impact of risk events, the cost impact value, the time impact value, or the project benefit impact value. S8500: Based on the weights corresponding to each semantic scenario card in the semantic probability distribution vector, the risk assessment results of each risk assessment branch are weighted and fused to generate a comprehensive risk assessment result. S8600: Identify key risk transmission paths based on the results of comprehensive risk assessment. Key risk transmission paths refer to risk transmission paths that, according to the results of comprehensive risk assessment, have an impact on the project's revenue, cost, schedule, compliance status, or supply chain stability that exceeds a preset impact threshold. S8700: Generate implementation guidance recommendations based on key risk transmission paths. Implementation guidance recommendations include at least one of the following: cost review recommendations, schedule adjustment recommendations, compliance review recommendations, supplier replacement recommendations, contract term adjustment recommendations, or contingency plan development recommendations.

[0059] Specifically, based on the identifier of the data element to be evaluated, its pre-established meaning mapping rules in the system can be retrieved. These meaning mapping rules include multiple semantic scenario cards, a calibrated semantic scenario influence distribution, and applicable conditions for each semantic scenario card. The semantic scenario card is used to structurally carry a candidate semantic interpretation and its corresponding correction rules; for example, a card might describe the impact of a specific policy change on the market. The calibrated semantic scenario influence distribution quantifies the relative importance of each semantic scenario card to the meaning of the current data element to be evaluated. The applicable conditions limit the data conditions, source conditions, time conditions, geographical conditions, or policy conditions under which each semantic scenario card is matched.

[0060] During the semantic transformation process, the system matches the data content, source attribute, time attribute, geographical attribute, or policy relevance attribute of the data element to be evaluated based on the applicable conditions corresponding to each semantic scenario card. The policy relevance attribute refers to the association identifier or association category between the data element to be evaluated and external supporting data. For example, a data element may be associated with a policy scenario card because its content mentions a specific policy clause or its source is a government report. Through this matching mechanism, at least one semantic scenario card that the data element to be evaluated matches can be identified. Subsequently, the system extracts the weights corresponding to the at least one matched semantic scenario card from the calibration semantic scenario influence distribution. These weights represent the relative importance of different semantic scenario cards to the meaning of the data element. Based on these matched semantic scenario cards and their corresponding weights, the system generates a semantic probability distribution vector, which quantifies the multiple semantic scenario cards that the data element to be evaluated may match and their respective weights. Finally, the data content of the data element to be evaluated, the semantic probability distribution vector, and the risk dimensions and impacts in the at least one semantic scenario card are combined to generate an internal semantic representation. The internal semantic representation is a data representation formed before being input into the inference model, which includes the semantic category, semantic weight, risk dimension, or applicable conditions of the data elements to be evaluated after meaning mapping.

[0061] During the risk reasoning phase, the aforementioned internal semantic representation is input into the reasoning model, which executes preset risk reasoning rules. Within the reasoning model, a corresponding risk assessment branch is configured for each semantic scenario card contained in the internal semantic representation. The risk assessment branch is the risk calculation path within the reasoning model corresponding to a semantic scenario card. This means that for every possible semantic interpretation of a data element, the reasoning model can independently calculate it through the corresponding risk assessment branch. Each risk assessment branch extracts the corresponding risk factors and correlation strength from the internal semantic representation based on its corresponding semantic scenario card. Risk factors refer to policy, market, supply chain, technology, financing, or revenue factors that affect the risk assessment results of engineering investment projects. Correlation strength refers to the quantitative value of the strength of the influence relationship between two risk factors. For example, a semantic scenario card indicating a decline in market demand might lead a branch to extract a decrease in project revenue as a risk factor and assign it a high correlation strength. Each risk assessment branch independently calculates the risk assessment result under its corresponding semantic scenario card based on the extracted risk factors and correlation strength. Finally, based on the weights corresponding to each semantic scenario card in the semantic probability distribution vector, the risk assessment results of each risk assessment branch are weighted and fused to generate a comprehensive risk assessment result. Through weighted fusion, risk assessment results under different semantic scenarios are integrated to form a comprehensive risk view that reflects the impact of multiple semantic scenarios.

[0062] Based on the comprehensive risk assessment results, the system identifies key risk transmission paths—those that, according to the comprehensive risk assessment results, have an impact on the project's returns, costs, schedule, compliance status, or supply chain stability exceeding a preset threshold. Based on these key risk transmission paths, the system generates implementation guidance suggestions, providing actionable decision support for project management.

[0063] In another embodiment of this application, a method for generating adjustment results of meaning correction suggestions is further proposed, including: S5310: Configure a multi-dimensional evaluation framework for the meaning revision recommendations. The multi-dimensional evaluation framework includes policy compliance, market impact and implementation feasibility. S5320: Receives evaluation values ​​and evaluation basis data from multiple evaluation ends on each evaluation dimension of the multi-dimensional evaluation framework based on the meaning correction suggestions; The evaluation value refers to the quantitative evaluation result submitted by the evaluation end for the corresponding evaluation dimension. The evaluation basis data refers to the data records, text fragments, clause identifiers, indicator changes or event identifiers that are used to support the evaluation value.

[0064] S5330: Aggregate the evaluation values ​​of multiple evaluation ends for the same semantic scenario card in the meaning correction suggestions, and calculate the consensus score and divergence index of each semantic scenario card; Consensus score refers to the quantitative result of the consistency of evaluation values ​​submitted by multiple evaluation ends for the same semantic scenario card across various evaluation dimensions, while divergence index refers to the quantitative result of the difference of evaluation values ​​submitted by multiple evaluation ends for the same semantic scenario card across various evaluation dimensions.

[0065] S5340: When the divergence index exceeds the preset divergence threshold, the consultation processing mode is triggered. The consultation and processing mode refers to the processing mode that locates the evaluation dimensions and corresponding evaluation ends that have disagreements and receives the disagreement processing data when the disagreement index exceeds the preset disagreement threshold.

[0066] S5350: In consultation processing mode, it identifies the assessment dimension with the greatest disagreement and the multiple assessment ends corresponding to the assessment dimension with the greatest disagreement, and provides a disagreement processing interface. The disagreement handling interface refers to the interface used to receive disagreement handling data submitted by the evaluation end for the evaluation dimension with the greatest disagreement.

[0067] S5360: Based on the divergence processing data received from the divergence processing interface, update the weights of each evaluation dimension in the multi-dimensional evaluation framework. Evaluation dimension weight refers to the weight value used to represent the proportion of the corresponding evaluation dimension in the calculation of consensus score.

[0068] S5370: Recalculate the consensus score based on the updated evaluation dimension weights; S5380: Based on the recalculated consensus score, determine the recommendation priority of each semantic scenario card in the meaning correction suggestions; S5390: Based on the recommendation priority, determine the target semantic context card with the highest priority from the multiple semantic context cards included in the meaning correction suggestions, and use the target semantic context card and its corresponding evaluation dimension weight as the adjustment result for generating the meaning mapping rules of the data elements to be evaluated.

[0069] Specifically, to ensure a multi-faceted evaluation of the proposed meaning revisions, the system first configures a multi-dimensional evaluation framework. This framework can include policy compliance, market impact, and implementation feasibility. Policy compliance assesses whether the proposed meaning revisions align with current laws, regulations, industry policies, and regulatory requirements, determining their legality and compliance. Market impact assesses the potential impact of the proposed meaning revisions on market supply and demand, competitive landscape, price fluctuations, and investor sentiment, determining their market impact. Implementation feasibility assesses the operability of the proposed meaning revisions in terms of technology, resources, time, and cost, determining whether the revisions can be implemented in actual engineering investment projects.

[0070] After configuring the multi-dimensional evaluation framework, the system receives evaluation values ​​and evaluation basis data from multiple evaluation endpoints for each evaluation dimension in response to the meaning correction suggestions. Evaluation values ​​can be scores from 0 to 100, or qualitative levels such as high, medium, and low; the evaluation basis data provides data support and traceability for the evaluation values. For example, for policy compliance, the evaluation basis data can be the cited policy document clauses; for market impact, the evaluation basis data can be data charts in a market research report; for implementation feasibility, the evaluation basis data can be a technical assessment report or a resource list.

[0071] To quantify the consistency and discrepancies among multiple evaluation endpoints, the system aggregates the evaluation values ​​of multiple endpoints for the same semantic scenario card in the meaning correction suggestions, and calculates the consensus score and divergence index for each semantic scenario card. The consensus score represents the degree of consistency among multiple evaluation endpoints for the same semantic scenario card across various evaluation dimensions; it can be obtained, for example, by calculating the standard deviation, variance, or specific consistency indicators (such as Kendall's coefficient of harmony) between evaluation values. The divergence index represents the degree of difference among multiple evaluation endpoints for the same semantic scenario card across various evaluation dimensions, and is used to identify semantic scenario cards with significant divergence. When the divergence index exceeds a preset divergence threshold, it indicates a large difference between the multiple evaluation endpoints, and the system triggers a consultation processing mode.

[0072] In the consultation and processing mode, the system identifies the assessment dimension with the greatest disagreement and the multiple assessment ends corresponding to that dimension, and provides a disagreement processing interface. This interface receives disagreement processing data submitted by assessment ends for the assessment dimension with the greatest disagreement. This data may include supplementary evidence, assessment basis data, explanations of assessment logic, or revised assessment values. For example, this interface could be an online collaboration platform, allowing assessment ends to submit supplementary evidence, explain their assessment logic, or engage in online discussions. Based on the disagreement processing data received by the interface, the system updates the assessment dimension weights of each dimension in the multi-dimensional assessment framework. For example, if the disagreement processing data indicates that a certain dimension is more critical in the current context, the weight of that dimension can be increased accordingly.

[0073] After updating the evaluation dimension weights, the system recalculates the consensus score based on the updated weights to ensure the score reflects the importance of the updated evaluation dimensions. Based on the recalculated consensus score, the system determines the recommendation priority of each semantic scenario card in the meaning correction proposal. A higher consensus score generally indicates greater consistency of the semantic scenario card across multiple evaluation endpoints, and thus a higher recommendation priority. Finally, based on the recommendation priority, the system identifies the highest-priority target semantic scenario card from the multiple semantic scenario cards included in the meaning correction proposal, and uses the target semantic scenario card and its corresponding evaluation dimension weights as the adjustment result for generating the meaning mapping rules for the data elements to be evaluated.

[0074] In another embodiment of this application, S5360 further includes: S5361: Continuously monitor changes in the external environment. Changes in the external environment refer to policy and regulatory changes, macroeconomic changes, market supply and demand changes, industry technology changes, or geopolitical events that affect the risk assessment of engineering investment projects. S5362: Extract semantic features of the external environment from changes in the external environment. Semantic features of the external environment refer to keywords, topic categories, event categories, semantic vectors, or risk labels used to represent the main meaning of changes in the external environment. S5363: Maintain the association set between each evaluation dimension and the semantic features of the external environment in the multi-dimensional evaluation framework. The association set refers to the data set used to record the correspondence between each evaluation dimension and different semantic features of the external environment. S5364: Based on the degree of matching or deviation between the semantic features of the external environment and the associated set, generate the environmental adjustment coefficients corresponding to each evaluation dimension. The environmental adjustment coefficients are coefficients used to adjust the weights of the evaluation dimensions based on the impact of changes in the external environment on the importance of the evaluation dimensions. S5365: Update the weights of each evaluation dimension in the multi-dimensional evaluation framework based on the environmental adjustment coefficient and the divergence processing data received by the divergence processing interface. S5370 includes: S5371: Update the internal evaluation criteria of each evaluation dimension based on the impact of changes in the external environment on the evaluation criteria of each evaluation dimension. Evaluation criteria refer to the judgment rules used to determine the evaluation value on the corresponding evaluation dimension. S5372: Based on the updated evaluation dimension weights and the updated evaluation criteria, recalculate the consensus score of each semantic scenario card included in the meaning correction proposal.

[0075] Specifically, continuous monitoring of changes in the external environment refers to the system's ongoing attention to and collection of external information that may affect the risk assessment of engineering investment projects. This information can cover the release and revision of policies and regulations, fluctuations in macroeconomic indicators, evolution of market supply and demand, breakthroughs and iterations in industry technologies, and the occurrence of geopolitical events. The purpose is to ensure that the external information upon which the risk assessment is based can be updated as the external environment changes.

[0076] Extracting semantic features of the external environment from changes in the external environment refers to performing semantic analysis on monitored external environment data to identify and extract key information that represents its main meaning, such as keywords, topic categories, event categories, semantic vectors, or risk labels. These semantic features of the external environment are used to subsequently determine the importance and impact of changes in the external environment on each assessment dimension.

[0077] In practical applications, maintaining the set of associations between each assessment dimension in a multi-dimensional assessment framework and the semantic features of the external environment refers to establishing and dynamically updating a mapping relationship. This mapping relationship represents the correlation or influence path between specific semantic features of the external environment and assessment dimensions such as policy compliance, market impact, and implementation feasibility in the multi-dimensional assessment framework. For example, the semantic features of a carbon neutrality policy may be highly correlated with the policy compliance dimension.

[0078] Furthermore, based on the degree of matching or deviation between the semantic features of the external environment and the associated set, environmental adjustment coefficients are generated for each assessment dimension. This refers to calculating coefficients used to adjust the weights of assessment dimensions by quantifying the degree of conformity or difference between the semantic features of the external environment and the existing relationships in the associated set. For example, when changes in the external environment that are highly correlated with the policy compliance dimension are detected, the environmental adjustment coefficient for that dimension can be increased to reflect its increased importance.

[0079] Therefore, updating the weights of each evaluation dimension in the multi-dimensional evaluation framework based on the environmental adjustment coefficient and the disagreement handling data received by the disagreement handling interface means quantifying the impact of changes in the external environment through the environmental adjustment coefficient and combining it with the feedback from the evaluation end reflected in the disagreement handling data to jointly determine the updated weights of each evaluation dimension. In this way, the evaluation dimension weights can reflect both changes in the external objective environment and incorporate data feedback generated during the disagreement handling process.

[0080] Furthermore, updating the internal assessment standards for each assessment dimension based on the impact of changes in the external environment on the assessment standards of each assessment dimension means updating the judgment rules used to determine the assessment values ​​for the corresponding assessment dimension when changes in the external environment cause the original judgment rules to no longer be suitable for the current risk assessment scenario. For example, new environmental protection policies may lead to more stringent environmental standards under the feasibility dimension.

[0081] Finally, based on the updated evaluation dimension weights and updated evaluation criteria, the consensus score of each semantic scenario card included in the meaning correction suggestions is recalculated. This means that under the new weight and standard system, the consistency of the evaluation values ​​of the semantic scenario cards by each evaluation end is recalculated to obtain a consensus evaluation that is more in line with the current external environment.

[0082] In another embodiment of this application, S5320 further includes: S5321: Receive evaluation values ​​and evaluation basis data submitted by multiple evaluation ends on each evaluation dimension of the multi-dimensional evaluation framework based on the meaning correction suggestions; S5322: Perform a completeness check on the evaluation value and a sufficiency check on the evaluation basis data. The completeness check refers to determining whether the evaluation value is a null value, a default value, or exceeds the preset value range of the corresponding evaluation dimension. The sufficiency check refers to determining whether the evaluation basis data is missing data records, text fragments, clause identifiers, indicator changes, or event identifiers, or whether there are invalid reference identifiers or ambiguous expressions in the evaluation basis data. S5323: When the evaluation value fails the integrity check or the evaluation basis data fails the sufficiency check, the supplementary check mode is triggered; S5324: In the supplementary verification mode, supplementary verification items are generated based on the meaning correction suggestions, the corresponding semantic scenario card, the corresponding evaluation dimension, the evaluation value, and the evaluation basis data. Supplementary verification items refer to verification data items used to instruct the corresponding evaluation end to supplement the evaluation value or evaluation basis data. S5325: Based on the assessment end attributes of the assessment end, send supplementary verification items that match the assessment end attributes to the assessment end. The assessment end attributes refer to attribute tags or historical submission records used to characterize the assessment end in at least one assessment direction among policy, market, supply chain, technology, financing or revenue. S5326: After receiving the supplementary evaluation value and supplementary evaluation basis data returned by the evaluation end based on the supplementary verification item, perform integrity verification and sufficiency verification again until the evaluation value and evaluation basis data meet the integrity verification and sufficiency verification. S5328: Use the evaluation values ​​and evaluation basis data that satisfy the integrity check and sufficiency check as evaluation input data for calculating the consensus score and the divergence index.

[0083] Specifically, after receiving evaluation values ​​and supporting data from multiple evaluation endpoints regarding meaning correction suggestions across various dimensions of the multi-dimensional evaluation framework, the system first performs a quality check on the evaluation values ​​and supporting data. The integrity check checks whether the evaluation value is null, a default value, or exceeds the preset value range for the corresponding evaluation dimension. For example, if an evaluation dimension requires evaluation values ​​to be integers between 0 and 100, any input outside this range or that is not numeric will be considered to have failed the integrity check. The sufficiency check checks whether the supporting data can support the corresponding evaluation value, including determining whether the supporting data lacks necessary data records, text fragments, clause identifiers, indicator changes, or event identifiers, and whether there are invalid references or ambiguous expressions, thereby reducing biases in subsequent analysis caused by insufficient evidence or unclear expressions.

[0084] When an assessment value fails the integrity check or the assessment basis data fails the sufficiency check, the system triggers a supplementary verification mode. In this mode, the system generates one or more supplementary verification items based on the current meaning correction suggestion, the corresponding semantic scenario card, the corresponding assessment dimension, the submitted assessment value, and the assessment basis data. These supplementary verification items indicate the data content that the corresponding assessment end needs to supplement or correct. Further, based on the assessment end's attributes, such as attribute tags or historical submission records related to policy, market, supply chain, technology, financing, or revenue, the system sends supplementary verification items matching the assessment end's attributes, ensuring that the supplementary verification items correspond to the assessment end's assessment direction. After receiving the supplementary assessment value and supplementary assessment basis data submitted by the assessment end, the system performs integrity and sufficiency checks again. This process can be repeated until the assessment value and assessment basis data satisfy both integrity and sufficiency checks. Finally, the assessment value and assessment basis data that satisfy both integrity and sufficiency checks are used as input data for calculating the consensus score and divergence index.

[0085] In another embodiment of this application, it is further proposed that, prior to S5328, the following is also included: S5327-1: Continuously track and collect external reference data corresponding to the data used for evaluation. External reference data refers to policy documents, market reports, supply chain reports, technical standard documents, or news texts that are cited or associated with the data used for evaluation. S5327-2: Perform content parsing on externally referenced data to extract the publishing organization, update frequency, and sentiment of the externally referenced data; S5327-3: Calculate the credibility of the issuing organization based on its authority; S5327-4: Calculate the timeliness of the update frequency based on the update frequency and the collection or publication time of external reference data; S5327-5: Calculate the objectivity of sentiment tendency based on sentiment tendency; S5327-6: Construct a source map between evaluation values, evaluation basis data and external reference data. A source map is a data structure used to represent the reference relationship or association between evaluation values, evaluation basis data and external reference data. S5327-7: By tracing the source map, compare the descriptions of the evaluation values ​​and evaluation basis data with the original descriptions of the externally referenced data to identify the degree of deviation of the evaluation values ​​and evaluation basis data from the original descriptions; S5327-8: Calculate the reliability score of externally cited data based on the credibility of the publishing organization, the timeliness of the update frequency, and the objectivity of the sentiment bias. S5327-9: Based on the degree of deviation from the description and the reliability score, quantify the potential deviation information in the evaluation value and the evaluation basis data. Potential deviation information refers to the quantitative information that the evaluation value or evaluation basis data deviates from the original description of the external reference data. S5327-10: Based on potential deviation information, at least one of the evaluation value or evaluation basis data is corrected, and the corrected evaluation value or corrected evaluation basis data is used as one of the evaluation input data.

[0086] Continuously tracking and collecting externally cited data refers to the system actively or passively acquiring the original information sources cited by the evaluation data. Externally cited data can be understood as the underlying facts or viewpoints upon which the evaluation data relies. These can take various forms, such as government policy documents, market reports from authoritative institutions, supply chain reports from industry associations, technical standard documents developed by international organizations, or news texts published by mainstream media. By continuously tracking and collecting externally cited data, the system can further confirm, after the evaluation data has passed integrity and sufficiency checks, whether the cited sources are genuine, whether they correspond to the evaluation data, and whether they are traceable.

[0087] Content analysis of externally cited data to extract publishing institutions, update frequency, and sentiment refers to using technologies such as natural language processing and text mining to identify the source entity, the frequency of information updates, and the sentiment expressed in the text content, such as positive, negative, or neutral. The credibility of the publishing institution refers to its authority, professionalism, and trustworthiness in a specific field, which can usually be quantitatively assessed through factors such as its historical reputation, industry status, and regulatory qualifications. The timeliness of the update frequency refers to whether the information reflected in the externally cited data is up-to-date, which can be measured by the relationship between the publication time, collection time, and the current time window. The objectivity of sentiment refers to whether the externally cited data maintains neutrality and impartiality in expressing facts or opinions, and whether there is obvious subjective bias or emotional expression. Therefore, the publishing institution, update frequency, and sentiment are respectively converted into the credibility of the publishing institution, the timeliness of the update frequency, and the objectivity of the sentiment, serving as the basic indicators for calculating the reliability score of externally cited data.

[0088] Constructing a source map refers to establishing a data structure to represent the citation, association, or supporting relationships between evaluation values, evaluation basis data, and external reference data. This source map can display the chain of information sources; for example, which evaluation basis data supports a given evaluation value, and which external documents or reports these evaluation basis data cite. Through the source map, the statements in the evaluation values ​​and evaluation basis data can be compared with the original statements in the external reference data, thereby identifying any deviations in expression. The degree of expression deviation indicates whether the evaluation basis data, when paraphrasing or referencing the original information, has undergone expanded interpretation, narrowed interpretation, omitted key conditions, altered semantic emphasis, or selective presentation.

[0089] The reliability score of externally cited data can be calculated by comprehensively considering the credibility of the issuing institution, the timeliness of its updates, and the objectivity of its sentiment. For example, a recently updated, objective, and neutral policy document issued by an authoritative government agency will have a higher reliability score; while a news report published by an anonymous source, with outdated content and strong subjective bias, will have a lower reliability score. The reliability score indicates the credibility of externally cited data as a source of evaluation and is used to subsequently determine whether there is potential bias in the evaluation value and the data on which the evaluation is based, in conjunction with the degree of deviation in expression.

[0090] Quantifying potential bias information refers to the quantitative assessment of potential errors, inaccuracies, or misleading information in the evaluation value or evaluation basis data by combining the degree of deviation in expression with the reliability score of externally referenced data. For example, if the evaluation basis data deviates significantly from the expression of a highly reliable externally referenced data, its potential bias information quantification value will be high. Ultimately, based on this potential bias information, the evaluation value or evaluation basis data can be corrected. Correction may include adjusting the magnitude of the evaluation value, modifying the specific expression of the evaluation basis data, or even marking the parts that need to be reviewed, to ensure that the input data used for risk assessment is as accurate, objective, and reliable as possible. The corrected evaluation value or corrected evaluation basis data serves as one of the evaluation input data and can be used together with other evaluation values ​​or evaluation basis data that meet the integrity and sufficiency checks and do not require correction in subsequent calculations of consensus score and divergence index.

[0091] In another embodiment of this application, S1000 is further proposed to include: S1100: Continuously monitor policy release platforms, market analysis agency websites, supply chain data platforms, technology assessment document libraries, and news media platforms to identify new data source access points and update activities of existing data sources. For the identified new data source access points or update activities of existing data sources, determine the data sources to be collected. Data source access points refer to web page addresses, application programming interfaces, database interfaces, or file download addresses used to obtain external data. S1200: For the identified data source to be collected, select the corresponding collection agent according to the geographic attributes, language type and content theme of the data source. The collection agent is a collection program that is configured to extract data according to the access protocol, page structure, file format or interface field of the corresponding data source. S1300: Through the acquisition agent, it extracts raw external data according to the specific structure and access protocol of the data source to be acquired, and performs language recognition and regional attribution marking on the raw external data; S1400: Based on the results of language recognition, call the corresponding translation module to perform cross-language semantic conversion on the original external data, and standardize the region-specific terms according to the regional affiliation mark to obtain standardized external data. Region-specific terms refer to the different expressions used for the same policy, market, engineering or supply chain concept in different countries, regions, industry organizations or regulatory systems. S1500: Encapsulates standardized external data according to a preset unified format and stores it in the structured data storage area; S1600: Extract fields, tags, text fragments, or indicators related to the risk assessment of engineering investment projects from the structured data storage area as data elements to be assessed.

[0092] Specifically, a data source access point can be understood as an entry point for data acquisition. Examples include a specific page on a government department's official website, an application programming interface (API) provided by an industry association, a database connection string, or a link to a downloadable report file. Continuous monitoring is used to discover new data source access points and update activities of existing data sources. New data source access points correspond to data entry points not previously included in the collection scope, while update activities of existing data sources correspond to data additions, revisions, or version updates occurring in data entry points already included in the collection scope. Through this process, the aforementioned new data source access points and update activities of existing data sources can be uniformly identified as data sources to be collected, facilitating subsequent data extraction by the collection agent.

[0093] A data scraping agent is an automated program or script that performs customized data scraping based on the characteristics of a specific data source. For example, for a website that requires login, the agent is configured to simulate user login behavior; for a data source providing a RESTful API, the agent sends requests according to the API documentation and parses the returned JSON or XML data. Geographic attributes refer to the country or region where the data source is located, language type refers to the language used in the data source content, and content theme refers to the main areas of focus of the data source, such as environmental policies or the energy market. The system selects the appropriate data scraping agent based on geographic attributes, language type, and content theme, ensuring that the agent matches the access method, content structure, and semantic context of the data source to be scraped.

[0094] The purpose of language recognition and geographic attribution tagging is to provide a foundation for subsequent standardization processes. Language recognition can be achieved through natural language processing techniques, such as using language models to determine the dominant language of the text. Geographic attribution tagging can be based on the URL or IP address of the data source, the registered location of the publishing organization, or geographical entities mentioned in the text content. The translation module can be a rule-based translation system, a statistical machine translation system, or a neural network machine translation system, used to convert raw data in non-common languages ​​into a unified working language, such as Chinese or English. The standardization of region-specific terms aims to address the issue of different regions or industries using different terms for the same concept. For example, a PPP project may be called a public-private partnership project in some regions and a concession project in others. Standardization processes establish a terminology lookup table or ontology, mapping these region-specific terms to a unified, standardized conceptual representation, thereby obtaining standardized external data that can be used for subsequent risk assessment at both the language and terminology levels.

[0095] In practical applications, a pre-defined unified format can be JSON, XML, CSV, or a specific database table structure to ensure that all collected data is stored in a consistent manner, facilitating subsequent processing and querying. The structured data storage area can be a relational database, NoSQL database, data lake, or data warehouse, used to store and manage standardized external data. Data elements to be evaluated refer to key information directly used for risk assessment, selected from standardized external data. For example, subsidy policy adjustments extracted from policy documents as tags, raw material price indices extracted from market reports as indicators, or tightening environmental protection policies in a certain region extracted from news texts as text fragments.

[0096] In another embodiment of this application, S5327-2 further includes: S5327-21: Perform multi-domain semantic analysis on the text content in externally referenced data. Multi-domain semantic analysis refers to combining professional dictionaries, semantic rules, and contextual analysis from at least one of the following fields: engineering investment, policy and regulations, supply chain, market finance, and industry technology, to perform semantic recognition on the text content. S5327-22: Through multi-domain semantic analysis, the deep semantics of the text content are obtained. Deep semantics refers to the policy meaning, market meaning, supply chain meaning, or risk meaning that are different from the literal expression of the text. S5327-23: Identify the publishing organization corresponding to externally referenced data based on deep semantic analysis; S5327-24: Determine the specific context of the issuing organization based on its background information. The specific context refers to the textual interpretation background formed by the issuing organization due to its nature, regulatory responsibilities, historical stance, geographical attributes, or industry role. S5327-25: Identify the update frequency of externally referenced data based on deep semantics and specific context; S5327-26: Analyze the sentiment tendency of the text content based on deep semantics and specific context, and quantify the degree of sentiment tendency bias; Based on sentiment tendency, the objectivity of sentiment tendency is calculated, including: S5327-51: Calculate the objectivity of emotional tendency based on the emotional tendency and the degree of bias of the emotional tendency.

[0097] Multi-domain semantic parsing refers to the process of semantic analysis of text content in externally referenced data. This parsing process combines dictionaries, semantic rules, and contextual information from multiple professional fields such as engineering investment, policy and regulations, supply chain, market finance, and industry technology to semantically identify the text content. This allows the system to recognize the meaning of the text within a specific professional field, rather than relying solely on literal expression. Through multi-domain semantic parsing, the deep semantics of the text content can be obtained. This deep semantics differs from the literal expression of the text and can reveal its potential policy implications, market implications, supply chain implications, or risk implications. For example, a seemingly neutral word may have a positive or negative policy orientation in a specific policy context.

[0098] Furthermore, the system identifies the issuing organization corresponding to the externally cited data based on deep semantic analysis. The issuing organization can be the department issuing policy documents, the organization publishing market reports, the provider of supply chain reports, the organization formulating technical standards documents, or the media publishing news texts. After identifying the issuing organization, the system determines the specific context of the issuing organization based on its background information. Specific context refers to the textual interpretation background formed by factors such as the issuing organization's nature, regulatory responsibilities, historical stance, geographical attributes, or industry role. For example, even if the content of a policy document issued by a government agency and a report issued by a market research institution are similar, their interpretation angles and potential impacts may differ depending on the specific context of the issuing organization.

[0099] Based on deep semantics and specific context, the system identifies the update frequency of externally cited data and analyzes the sentiment tendency of the text content, while quantifying the degree of sentiment bias. Update frequency represents the frequency with which externally cited data is updated daily, weekly, monthly, quarterly, annually, or irregularly, used to calculate the timeliness of the update frequency. Sentiment tendency indicates the overall positive, negative, or neutral orientation of the text content; the degree of sentiment bias indicates the strength of this positive, negative, or neutral orientation. By combining deep semantics and specific context to identify update frequency and sentiment tendency, the system can reduce biases caused by judging solely based on surface-level time information or emotional vocabulary.

[0100] Based on this, the system calculates the objectivity of sentiment tendency according to the sentiment orientation and the degree of bias in the sentiment orientation. The objectivity of sentiment tendency is a quantitative result used to indicate whether the externally cited data remains neutral in the text expression and whether there is obvious subjective bias or emotional expression. Generally, the closer the sentiment orientation is to neutral and the lower the degree of bias in the sentiment orientation, the higher the objectivity of the sentiment orientation; the more the sentiment orientation is biased towards positive or negative and the higher the degree of bias in the sentiment orientation, the lower the objectivity of the sentiment orientation. Therefore, the sentiment orientation and the degree of bias in the sentiment orientation obtained in S5327-26 can be used as the basis for calculating the objectivity of sentiment tendency in S5327-51, and further used for calculating the reliability score of externally cited data.

[0101] Reference Figure 2 This application proposes a risk assessment system for engineering investment projects based on a reasoning model, comprising: Module 1 is used to acquire data elements to be evaluated from external data sources of engineering investment projects. External data sources include at least one of policy documents, market reports, supply chain data, technology assessment documents and news texts. Data elements to be evaluated refer to fields, tags, text fragments or indicators extracted from external data sources for use in risk assessment of engineering investment projects. Behavioral performance acquisition module 2 is used to acquire the current behavioral performance of the data element to be evaluated. The current behavioral performance refers to the data change characteristics of the data element to be evaluated within the current time window. Module 3 is used to establish stable reference performance for the data element to be evaluated. Stable reference performance refers to the change characteristics of the data element to be evaluated within a preset stable reference time period, which is used to compare with the current behavior performance. The meaning deviation quantification module 4 is used to compare the current behavioral performance with the stable reference performance to obtain the degree of meaning deviation of the data element to be evaluated; It is suggested that module 5 be used to generate and send meaning correction suggestions based on external supporting data related to the data element to be evaluated when the degree of meaning deviation meets the preset deviation conditions. External supporting data refers to data retrieved from external data sources that is used to explain or prove that the meaning of the data element to be evaluated has changed. Meaning correction suggestions refer to candidate explanations and corresponding correction rules used to indicate that the current meaning of the data element to be evaluated has changed relative to the stable reference performance. Rule update module 6 is used to receive the adjustment results returned based on the meaning correction suggestions, and generate or update the meaning mapping rules of the data elements to be evaluated based on the adjustment results. The meaning mapping rules are rules used to convert the data elements to be evaluated into internal semantic representations that can be processed by the inference model. Semantic conversion module 7 is used to perform semantic conversion on the data elements to be evaluated according to the meaning mapping rules, and generate internal semantic representations. Internal semantic representations refer to data representations formed before being input into the inference model, which contain the semantic category, semantic weight, risk dimension, or applicable conditions of the data elements to be evaluated after meaning mapping. Inference module 8 is used to input the internal semantic representation into the inference model that executes the preset risk inference rules, so that the inference model can construct internal features based on the internal semantic representation and identify the risk transmission path of the engineering investment project; Output module 9 is used to output the risk transmission path.

[0102] The core of the engineering investment project risk assessment system based on reasoning model proposed in this application lies in the dynamic perception, correction, and application of changes in the meaning of external data elements through the collaborative work of various functional modules.

[0103] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A risk assessment method for engineering investment projects based on a reasoning model, characterized in that, include: The data elements to be evaluated are obtained from external data sources of engineering investment projects. The external data sources include at least one of policy documents, market reports, supply chain data, technology assessment documents, and news texts. The data elements to be evaluated refer to fields, tags, text fragments, or indicators extracted from the external data sources for use in the risk assessment of engineering investment projects. Obtain the current behavior of the data element to be evaluated, whereby the current behavior refers to the data change characteristics exhibited by the data element to be evaluated within the current time window; Establish a stable reference performance for the data element to be evaluated. The stable reference performance refers to the change characteristics of the reference data element to be evaluated within a preset stable reference time period, which are used to compare with the current behavior performance. The degree of deviation in meaning of the data element to be evaluated is obtained by comparing the current behavior with the stable reference behavior. When the degree of deviation of the meaning meets the preset deviation conditions, a meaning correction suggestion is generated and sent based on the external supporting data related to the data element to be evaluated. The external supporting data refers to the data retrieved from the external data source that is used to explain or prove that the meaning of the data element to be evaluated has changed. The meaning correction suggestion refers to the candidate explanation and corresponding correction rules used to indicate that the current meaning of the data element to be evaluated has changed relative to the meaning corresponding to the stable reference performance. Receive the adjustment result returned based on the meaning correction suggestion, and generate or update the meaning mapping rule of the data element to be evaluated based on the adjustment result. The meaning mapping rule refers to the rule used to convert the data element to be evaluated into an internal semantic representation that can be processed by the inference model. According to the meaning mapping rule, the data element to be evaluated is semantically transformed to generate an internal semantic representation. The internal semantic representation refers to the data representation formed before being input into the inference model, which includes the semantic category, semantic weight, risk dimension or applicable conditions of the data element to be evaluated after meaning mapping. The internal semantic representation is input into the inference model that executes preset risk inference rules, so that the inference model constructs internal features based on the internal semantic representation and identifies the risk transmission path of the engineering investment project; Output the risk transmission path.

2. The risk assessment method for engineering investment projects based on inference models according to claim 1, characterized in that, When the degree of deviation of meaning meets the preset deviation condition, the meaning correction suggestion is generated and sent based on external supporting data related to the data element to be evaluated, including: When the degree of deviation of the meaning meets the preset deviation condition, multiple candidate semantic interpretations corresponding to the data element to be evaluated are extracted from the external supporting data. The candidate semantic interpretations refer to different interpretation results formed on the current meaning of the data element to be evaluated based on different external supporting data. Multiple semantic scenario cards are generated based on multiple candidate semantic interpretations. Each semantic scenario card is a data object used to structurally carry a candidate semantic interpretation and its corresponding correction rules. The semantic scenario card includes a scenario name, policy basis, risk dimensions and impacts, supporting evidence, and corresponding correction rules. Generate and send meaning correction suggestions containing multiple semantic scenario cards; The step of receiving the adjustment result returned based on the meaning correction suggestion, and generating or updating the meaning mapping rule of the data element to be evaluated based on the adjustment result, includes: Receive adjustment results regarding the proposed meaning correction, the adjustment results including risk influence weights assigned to multiple semantic context cards; Based on the risk influence weight, calculate the aggregate weight corresponding to each semantic scenario card; The aggregate weights corresponding to each semantic context card are normalized to form the semantic context influence distribution; Send the semantic context influence distribution and receive the distribution adjustment results returned for the semantic context influence distribution. The distribution adjustment results include at least one of the following: confirmation, modification, deletion or addition of the weight corresponding to at least one semantic context card in the semantic context influence distribution. Based on the distribution adjustment results, the semantic context influence distribution is updated to obtain the calibrated semantic context influence distribution; Based on the multiple semantic scenario cards, the distribution of the influence of the calibration semantic scenario, and the applicable conditions corresponding to each semantic scenario card, the meaning mapping rules of the data element to be evaluated are generated or updated. The applicable conditions refer to the data conditions, source conditions, time conditions, regional conditions, or policy conditions used to determine whether the data element to be evaluated is applicable to the corresponding semantic scenario card.

3. The risk assessment method for engineering investment projects based on inference models according to claim 2, characterized in that, The step of performing semantic transformation on the data element to be evaluated according to the meaning mapping rule to generate an internal semantic representation includes: Based on the identifier of the data element to be evaluated, query the meaning mapping rule corresponding to the data element to be evaluated; Read multiple semantic scenario cards, the calibrated semantic scenario influence distribution, and the applicable conditions corresponding to each semantic scenario card from the meaning mapping rules; Based on the applicable conditions corresponding to each semantic scenario card, the data content, source attribute, time attribute, regional attribute or policy association attribute of the data element to be evaluated are matched to determine at least one semantic scenario card that the data element to be evaluated matches. The policy association attribute refers to the association identifier or association category between the data element to be evaluated and external supporting data. Extract the weights corresponding to at least one semantic scenario card from the calibration semantic scenario influence distribution; Based on the at least one semantic context card and its corresponding weight, a semantic probability distribution vector is generated. The semantic probability distribution vector is a vector used to represent the at least one semantic context card hit by the data element to be evaluated and the corresponding weight of each semantic context card. The internal semantic representation is generated by combining the data content of the data element to be evaluated, the semantic probability distribution vector, and the risk dimensions and impacts in at least one semantic scenario card. The step of inputting the internal semantic representation into a reasoning model that executes preset risk reasoning rules, so that the reasoning model constructs internal features based on the internal semantic representation and identifies the risk transmission path of the engineering investment project, includes: The internal semantic representation is input into the inference model that executes the preset risk inference rules; In the inference model, a corresponding risk assessment branch is configured for each semantic scenario card contained in the internal semantic representation. The risk assessment branch refers to the risk calculation path corresponding to a semantic scenario card in the inference model. Each of the risk assessment branches extracts the corresponding risk factors and correlation strength from the internal semantic representation based on its corresponding semantic scenario card. The risk factors refer to policy, market, supply chain, technology, financing or revenue factors that affect the risk assessment results of engineering investment projects. The correlation strength refers to the quantitative value of the strength of the influence relationship between two risk factors. Each of the risk assessment branches independently calculates the risk assessment result under its corresponding semantic scenario card based on the extracted risk factors and the correlation strength. Based on the weights corresponding to each semantic scenario card in the semantic probability distribution vector, the risk assessment results of each risk assessment branch are weighted and fused to generate a comprehensive risk assessment result. Based on the comprehensive risk assessment results, key risk transmission paths are identified. These key risk transmission paths refer to those risk transmission paths that, according to the comprehensive risk assessment results, have an impact on the project's revenue, cost, construction period, compliance status, or supply chain stability exceeding a preset impact threshold. Implementation guidance recommendations are generated based on the key risk transmission paths described above.

4. The risk assessment method for engineering investment projects based on inference models according to claim 2, characterized in that, The adjustment results of the proposed meaning corrections include: The proposed modification of the meaning suggests configuring a multi-dimensional evaluation framework, which includes policy compliance, market impact, and implementation feasibility. The system receives evaluation values ​​and evaluation basis data for each evaluation dimension of the multi-dimensional evaluation framework based on the meaning correction suggestions from multiple evaluation ends. The evaluation value refers to the quantitative evaluation result input by the evaluation end for the corresponding evaluation dimension, and the evaluation basis data refers to the data records, text fragments, clause identifiers, indicator changes, or event identifiers corresponding to the evaluation value and used to support the evaluation value. The consensus score and divergence index of each semantic scenario card are calculated by aggregating the evaluation values ​​of the same semantic scenario card from multiple evaluation ends in the meaning correction suggestions. The consensus score is a quantitative result of the consistency of the evaluation values ​​of the same semantic scenario card input by multiple evaluation ends in each evaluation dimension, and the divergence index is a quantitative result of the difference of the evaluation values ​​of the same semantic scenario card input by multiple evaluation ends in each evaluation dimension. When the divergence index exceeds the preset divergence threshold, a consultation processing mode is triggered. In the consultation processing mode, the evaluation dimension with the greatest disagreement and the multiple evaluation ends corresponding to the evaluation dimension with the greatest disagreement are identified, and a disagreement processing interface is provided. The disagreement processing interface refers to the interface for receiving disagreement processing data submitted by the evaluation ends for the evaluation dimension with the greatest disagreement. Based on the divergence processing data received by the divergence processing interface, update the evaluation dimension weights of each evaluation dimension in the multi-dimensional evaluation framework. The consensus score is recalculated based on the updated evaluation dimension weights. Based on the recalculated consensus score, the recommendation priority of each semantic scenario card in the meaning correction proposal is determined; Based on the recommendation priority, the target semantic context card with the highest priority is determined from the meaning correction suggestions, and the target semantic context card and its corresponding evaluation dimension weight are used as the adjustment result for generating the meaning mapping rule of the data element to be evaluated.

5. The risk assessment method for engineering investment projects based on inference models according to claim 4, characterized in that, The step of updating the evaluation dimension weights of each evaluation dimension in the multi-dimensional evaluation framework based on the divergence processing data received by the divergence processing interface includes: Continuously monitor changes in the external environment, which refer to policy and regulatory changes, macroeconomic changes, market supply and demand changes, industry technology changes, or geopolitical events that affect the risk assessment of engineering investment projects. Extract semantic features of the external environment from the changes in the external environment. The semantic features of the external environment refer to keywords, topic categories, event categories, semantic vectors or risk tags used to represent the main meaning of the changes in the external environment. Maintain the association set between each evaluation dimension in the multi-dimensional evaluation framework and the semantic features of the external environment; Based on the degree of matching or deviation between the semantic features of the external environment and the association set, an environmental adjustment coefficient is generated for each evaluation dimension. The environmental adjustment coefficient is a coefficient generated based on the impact of changes in the external environment on the importance of the evaluation dimension and is used to adjust the weight of the evaluation dimension. Based on the environmental adjustment coefficient and the divergence processing data received by the divergence processing interface, update the evaluation dimension weights of each evaluation dimension in the multi-dimensional evaluation framework. The step of recalculating the consensus score based on the updated evaluation dimension weights includes: Based on the impact of the external environment changes on the evaluation criteria of each evaluation dimension, the evaluation criteria within each evaluation dimension are updated. The evaluation criteria refer to the judgment rules used to determine the evaluation value on the corresponding evaluation dimension. Based on the updated evaluation dimension weights and updated evaluation criteria, the consensus score of each semantic scenario card included in the meaning correction proposal is recalculated.

6. The risk assessment method for engineering investment projects based on inference models according to claim 4, characterized in that, The receipt of evaluation values ​​and evaluation basis data from multiple evaluation endpoints regarding the meaning correction suggestions on each evaluation dimension of the multi-dimensional evaluation framework includes: Receive evaluation values ​​and evaluation basis data submitted by multiple evaluation ends regarding the proposed corrections to the meaning, on each evaluation dimension of the multi-dimensional evaluation framework; The completeness of the evaluation value is verified, and the sufficiency of the evaluation basis data is verified. The completeness verification refers to determining whether the evaluation value is empty, a default value, or exceeds the preset value range of the corresponding evaluation dimension. The sufficiency verification refers to determining whether the evaluation basis data is missing data records, text fragments, clause identifiers, indicator changes, or event identifiers, or whether there are invalid reference identifiers or ambiguous expressions in the evaluation basis data. When the evaluation value fails the integrity check or the evaluation basis data fails the sufficiency check, the supplementary check mode is triggered. In the supplementary verification mode, supplementary verification items are generated based on the meaning correction suggestions, corresponding semantic scenario cards, corresponding evaluation dimensions, evaluation values, and evaluation basis data. The supplementary verification items refer to verification data items used to instruct the corresponding evaluation end to supplement evaluation values ​​or evaluation basis data. Based on the evaluation terminal attributes, supplementary verification items matching the evaluation terminal attributes are sent to the evaluation terminal. The evaluation terminal attributes refer to attribute tags or historical submission records used to characterize the evaluation terminal in at least one evaluation direction among policy, market, supply chain, technology, financing or revenue. After receiving the supplementary evaluation value and supplementary evaluation basis data returned by the evaluation terminal based on the supplementary verification item, the integrity verification and the sufficiency verification are performed again until the evaluation value and the evaluation basis data satisfy the integrity verification and the sufficiency verification. The evaluation values ​​and evaluation basis data that satisfy the integrity check and the sufficiency check are used as evaluation input data for calculating the consensus score and the divergence index.

7. The risk assessment method for engineering investment projects based on inference models according to claim 6, characterized in that, Before using the evaluation values ​​and evaluation basis data that satisfy the integrity check and the sufficiency check as evaluation input data for calculating the consensus score and the divergence index, the method further includes: Continuously track and collect external reference data corresponding to the assessment basis data. The external reference data refers to policy documents, market reports, supply chain reports, technical standard documents, or news texts that are referenced or associated with the assessment basis data. The external reference data is parsed to extract the publishing organization, update frequency, and sentiment tendency of the external reference data. Calculate the credibility of the issuing organization based on the issuing organization; The timeliness of the update frequency is calculated based on the update frequency and the collection or publication time of the external reference data. Calculate the objectivity of the emotional tendency based on the stated emotional tendency; Construct a source map between the evaluation value, the evaluation basis data, and the external reference data. The source map is a data structure used to represent the reference relationship or association relationship between the evaluation value, the evaluation basis data, and the external reference data. By comparing the descriptions of the evaluation values ​​and the evaluation basis data with the original descriptions of the external reference data using the source map, the degree of deviation of the evaluation values ​​and the evaluation basis data from the original descriptions can be identified. The reliability score of the externally cited data is calculated based on the credibility of the publishing organization, the timeliness of the update frequency, and the objectivity of the sentiment. Based on the degree of deviation of the statement and the reliability score, the potential deviation information in the evaluation value and the evaluation basis data is quantified. The potential deviation information refers to the quantitative information that the evaluation value or the evaluation basis data deviates from the original statement of the external reference data. Based on the potential deviation information, at least one of the evaluation value or the evaluation basis data is corrected, and the corrected evaluation value or the corrected evaluation basis data is used as one of the evaluation input data.

8. The risk assessment method for engineering investment projects based on inference models according to claim 1, characterized in that, The data elements to be evaluated from the external data source of the engineering investment project include: Continuously monitor policy release platforms, market analysis agency websites, supply chain data platforms, technology assessment document libraries, and news media platforms to identify new data source access points and update activities of existing data sources. For the identified new data source access points or update activities of existing data sources, determine the data sources to be collected. The data source access points refer to web page addresses, application programming interfaces, database interfaces, or file download addresses used to obtain external data. For the identified data source to be collected, a corresponding collection agent is selected based on the geographic attribute, language type, and content theme of the data source to be collected. The collection agent refers to a collection program that is configured to extract data according to the access protocol, page structure, file format, or interface field of the corresponding data source. The acquisition agent extracts raw external data based on the specific structure and access protocol of the data source to be acquired, and performs language recognition and geographic attribution marking on the raw external data. Based on the language recognition result, the corresponding translation module is invoked to perform cross-language semantic conversion on the original external data, and the region-specific terms are standardized according to the regional affiliation mark to obtain standardized external data. The region-specific terms refer to the different expressions used for the same policy, market, engineering or supply chain concept in different countries, regions, industry organizations or regulatory systems. The standardized external data is packaged according to a preset unified format and stored in a structured data storage area; Fields, tags, text fragments, or indicators related to the risk assessment of engineering investment projects are extracted from the structured data storage area and used as the data elements to be assessed.

9. The risk assessment method for engineering investment projects based on inference models according to claim 7, characterized in that, The step of parsing the external reference data to extract the publishing organization, update frequency, and sentiment tendency of the external reference data includes: Multi-domain semantic analysis is performed on the text content in the external reference data. The multi-domain semantic analysis refers to the semantic recognition of the text content by combining professional dictionaries, semantic rules and context analysis from at least one of the fields of engineering investment, policy and regulations, supply chain, market finance and industry technology. Through the multi-domain semantic analysis, the deep semantics of the text content are obtained. The deep semantics refers to the policy meaning, market meaning, supply chain meaning, or risk meaning that are different from the literal expression of the text. The publishing organization corresponding to the external reference data is identified based on the deep semantics. Based on the background information of the issuing organization, the specific context of the issuing organization is determined. The specific context refers to the textual interpretation background formed by the issuing organization due to its organizational nature, regulatory responsibilities, historical stance, regional attributes, or industry role. Based on the deep semantics and the specific context, the update frequency of the externally referenced data is identified; Based on the deep semantics and the specific context, analyze the sentiment tendency of the text content and quantify the degree of bias of the sentiment tendency; The step of calculating the objectivity of the emotional tendency based on the emotional tendency includes: The objectivity of the emotional tendency is calculated based on the emotional tendency and the degree of bias of the emotional tendency.

10. A risk assessment system for engineering investment projects based on a reasoning model, characterized in that, include: The acquisition module is used to acquire data elements to be evaluated from external data sources of engineering investment projects. The external data sources include at least one of policy documents, market reports, supply chain data, technology assessment documents, and news texts. The data elements to be evaluated refer to fields, tags, text fragments, or indicators extracted from the external data sources for use in risk assessment of engineering investment projects. The behavior performance acquisition module is used to acquire the current behavior performance of the data element to be evaluated, wherein the current behavior performance refers to the data change characteristics of the data element to be evaluated within the current time window; The reference establishment module is used to establish a stable reference performance of the data element to be evaluated. The stable reference performance refers to the reference data change characteristics of the data element to be evaluated within a preset stable reference time period, which are used to compare with the current behavior performance. The meaning deviation quantification module is used to compare the current behavioral performance with the stable reference performance to obtain the degree of meaning deviation of the data element to be evaluated; The suggestion generation module is used to generate and send meaning correction suggestions based on external supporting data related to the data element to be evaluated when the degree of deviation of the meaning meets the preset deviation conditions. The external supporting data refers to data retrieved from the external data source that is used to explain or prove that the meaning of the data element to be evaluated has changed. The meaning correction suggestions refer to candidate explanations and corresponding correction rules used to indicate that the current meaning of the data element to be evaluated has changed relative to the meaning corresponding to the stable reference performance. The rule update module is used to receive the adjustment result returned according to the meaning correction suggestion, and generate or update the meaning mapping rule of the data element to be evaluated according to the adjustment result. The meaning mapping rule refers to the rule used to convert the data element to be evaluated into an internal semantic representation that can be processed by the inference model. The semantic transformation module is used to perform semantic transformation on the data element to be evaluated according to the meaning mapping rules and generate an internal semantic representation. The internal semantic representation refers to the data representation formed before being input into the inference model, which includes the semantic category, semantic weight, risk dimension or applicable conditions of the data element to be evaluated after meaning mapping. The reasoning module is used to input the internal semantic representation into the reasoning model that executes preset risk reasoning rules, so that the reasoning model can construct internal features based on the internal semantic representation and identify the risk transmission path of the engineering investment project. The output module is used to output the risk transmission path.

Images