A Method and System for Verifying Environmental Investigation Conclusions Based on Heterogeneous Evidence Fusion

Abstract

This application discloses a method and system for verifying environmental survey conclusions based on heterogeneous evidence fusion. First, a set of environmental survey project documents is acquired. The project documents are then subjected to structured parsing to form paragraph objects and table record objects. Structured extraction is performed on the conclusions to be verified to obtain structured conclusion objects. Based on these structured conclusion objects, vector retrieval and reordering are performed in both the project document set and the domain knowledge base to obtain factual evidence and normative evidence. The recalled evidence is standardized to form unified evidence objects. Local evidence chains are constructed based on the conclusion type and verified. The confidence level of the conclusions is calculated based on the strength of supporting evidence, evidence coverage, logical consistency, and conflict penalty. The verification results are then output. This invention is applicable to the automatic verification of environmental survey reports for groundwater, surface water, soil, and sediment, and can improve the traceability of conclusion basis, the ability to judge the applicability of standards, and the ability to assess the credibility of conclusions.

Description

Technical Field

[0001] This application relates to the field of credibility assessment technology, and in particular to a method and system for verifying environmental investigation conclusions based on heterogeneous evidence fusion. Background Technology

[0002] Environmental investigation reports, especially those concerning groundwater, surface water, soil, and sediment, typically include a project overview, monitoring site layout, sampling details, test results, pollutant identification, causal analysis, risk assessment, and recommendations. The conclusions in these reports are often not based on a single piece of evidence, but rather supported by a combination of information, including monitoring results, statistical tables, standard documents, technical guidelines, and background explanations.

[0003] In the current process of compiling and reviewing environmental investigation reports, manual reading of the main text, appendices, and standard documents is typically relied upon, followed by technical personnel comparing and judging the conclusions. Because the basis for conclusions is often scattered across different chapters, tables, and external standard documents, problems such as unclear basis for conclusions, ambiguous standard compatibility, incomplete logic between conclusions and evidence, and failure to explicitly identify conflicting information easily arise. Existing general-purpose text retrieval or question-answering systems mostly only provide text content semantically similar to the conclusions, making it difficult to form a structured chain of evidence around specific conclusions, and also difficult to provide further credibility assessment results. Relying solely on ordinary retrieval methods or end-to-end generation methods can easily lead to incomplete recall, standard mismatch, duplicate evidence inclusion, and unstable confidence assessments. Therefore, a technical solution is needed that can automatically complete conclusion structuring, dual-source evidence retrieval, evidence chain weaving, conflict detection, and confidence assessment for specific conclusions in environmental investigation reports. Summary of the Invention

[0004] Based on this, the embodiments of this application provide a method and system for verifying environmental investigation conclusions based on heterogeneous evidence fusion, which solves the problems of evidence dispersion, unclear standard application relationships, untraceable basis of conclusions, and difficulty in quantifying credibility in the existing environmental investigation report conclusion verification process.

[0005] Firstly, a method for verifying environmental investigation conclusions based on heterogeneous evidence fusion is provided, the method comprising:

[0006] Step 1: Obtain the environmental investigation report, appendices, monitoring results, and related project data. And convert it into a structured collection of project documents. Build an environmental knowledge base ;

[0007] Step 2: Obtain the conclusion to be verified The conclusion to be verified Segment and extract into one or more structured sub-conclusion objects ;

[0008] Step 3: Based on the structured sub-conclusion object In the structured project document collection and domain knowledge base A dual-source search is performed to obtain a set of candidate evidence. ;

[0009] Step 4: Set up candidate evidence Convert into a standardized set of evidence objects And organize them into a partial chain of evidence based on the type of conclusion. ;

[0010] Step 5: Analyze the local chain of evidence. Perform consistency, applicability, and completeness checks to obtain the verification results. ;

[0011] Step 6: Analyze the chain of evidence. Verification results The confidence level of the conclusion is determined by scoring. ;

[0012] Step 7: Generate conclusion verification results Or explanatory text .

[0013] Optionally, in step 1, the acquisition of environmental survey reports, appendices, monitoring results, and related project data... And convert it into a structured collection of project documents. Build an environmental knowledge base Specifically:

[0014] The project's original document collection is The document includes the main body of the investigation report, monitoring results tables, sampling records, enterprise information, historical monitoring records, and appendices, and is a structured project document collection obtained through OCR tools. Domain knowledge base is This includes domain standards, industry norms, technical guidelines, risk assessment parameters, knowledge of pollution migration mechanisms, and historical case studies. The conclusions to be verified are... The target output is then represented as ,in, Indicates the conclusion Constructed local chain of evidence This indicates the supporting evidence set. This indicates a conflict or weakens the set of evidence. This indicates missing or insufficiently covered information. Indicates the final confidence level of the conclusion.

[0015] Optionally, in step 2, the conclusion to be verified is obtained. The conclusion to be verified Segment and extract into one or more structured sub-conclusion objects Specifically:

[0016] The conclusion is segmented and its fields are extracted using a structured conclusion extraction model to obtain one or more structured sub-conclusion objects. ; where each sub-conclusion object Represented as: .

[0017] Optionally, in step 3, based on the structured sub-conclusion object In the structured project document collection and domain knowledge base A dual-source search is performed to obtain a set of candidate evidence. Specifically:

[0018] Based on the structured sub-conclusion objects, respectively in the structured project document collection and domain knowledge base The process involves performing dual-source evidence retrieval to obtain a set of candidate evidence; the dual-source evidence retrieval includes an initial recall based on a text vector representation model and a fine ranking process based on a retrieval re-ranking model.

[0019] First, a conclusion representation is generated using a vector encoding model: ,in If we represent a text vector model, then the recall results on the project document side are represented as follows: , ;

[0020] The recall results from the knowledge base are represented as follows: , On the project document side, the main focus is on recalling factual evidence related to the project, such as monitoring results, site data, descriptions of pollutants, and regional descriptions. On the knowledge base side, the main focus is on recalling normative evidence such as standard clauses, technical guidelines, terminology explanations, and adaptation rules.

[0021] Subsequently, a reordering model was used. Calculate the relevance score, where The overall recall results after reordering candidate evidence based on scores are as follows: .

[0022] Optionally, in step 4, the candidate evidence set is... Convert into a standardized set of evidence objects And organize them into a partial chain of evidence based on the type of conclusion. Specifically:

[0023] For candidate evidence set Evidence standardization is performed to obtain a standardized set of evidence objects:

[0024]

[0025] Each standardized evidence object It can be represented as:

[0026] ;

[0027] The types of evidence include {monitoring evidence, standard evidence, causal evidence, spatial evidence}.

[0028] For well-structured table records and knowledge base entries, they are directly converted into standardized evidence objects through field mapping; for descriptive text evidence, the evidence type is assigned through an evidence type discrimination model.

[0029] Select the corresponding evidence chain template based on the current conclusion type, and perform evidence filling on the standardized evidence set to construct a local evidence chain:

[0030]

[0031] in, This represents the evidence chain template corresponding to the current conclusion type; to avoid the same fact being counted repeatedly, a normalization key is further constructed:

[0032]

[0033] Based on the normalization key, duplicate evidence is either merged or retained as conflict candidates.

[0034] Optionally, in step 5, the local chain of evidence is... Perform consistency, applicability, and completeness checks to obtain the verification results. Specifically:

[0035] Local chain of evidence Perform integrity, consistency, and adaptability checks, and obtain the check results:

[0036]

[0037] Among them, the "pass" item indicates that the verification conditions have been met, the "conflict" item indicates that there is a contradiction between the conclusion and the evidence, the "missing" item indicates that the key evidence is missing, and the "weak" item indicates that the evidence coverage is insufficient but no clear conflict has yet been formed.

[0038] The verification includes field integrity verification, relationship consistency verification, standard adaptation verification, and range consistency verification. For cases where the supported direction is difficult to determine directly from the rules, an evidence-supported relationship discrimination model can be invoked, and the output will be: .

[0039] Optionally, in step 6, the chain of evidence is... Verification results The confidence level of the conclusion is determined by scoring. Specifically:

[0040] According to the partial chain of evidence Validation results Perform a confidence assessment to obtain the confidence level of the conclusion; construct the following scoring function:

[0041]

[0042] in, This indicates support for the strength of the evidence. Indicates the coverage of key evidence. Indicates conflict punishment. This indicates a lack of punishment. These are weight parameters;

[0043] The strength of supporting evidence can be obtained by summing the weights of the main evidence:

[0044]

[0045] in, Indicates the first The weight of each piece of evidence, For indicator functions;

[0046] The coverage of key evidence can be expressed as:

[0047]

[0048] in, Indicates the number of key evidence slots that have been filled. This indicates the total number of key evidence slots required for the conclusion;

[0049] Ultimately, continuous scores are mapped to discrete confidence levels: When there are strong conflicts, standard mismatches, or missing key evidence, manual review is further triggered.

[0050] Optionally, in step 7, the conclusion verification results are generated. Or explanatory text Specifically:

[0051] Based on the structured conclusion object, local chain of evidence, verification results, and confidence level, generate readable conclusion verification output:

[0052]

[0053] The output includes at least the conclusion, a summary of the main chain of evidence, a description of conflicts and missing information, a confidence level, and whether manual review is recommended.

[0054] Secondly, a system for verifying environmental investigation conclusions based on heterogeneous evidence fusion is provided, the system comprising:

[0055] The document parsing module is used to obtain environmental investigation reports, appendices, monitoring results, and related project information. And convert it into a structured collection of project documents. Build an environmental knowledge base ;

[0056] The conclusion parsing module is used to obtain the conclusions to be verified. The conclusion to be verified Segment and extract into one or more structured sub-conclusion objects ;

[0057] The dual-source retrieval module is used to retrieve information based on the structured sub-conclusion object. In the structured project document collection and domain knowledge base A dual-source search is performed to obtain a set of candidate evidence. ;

[0058] The evidence standardization and evidence weaving module is used to organize candidate evidence sets. Convert into a standardized set of evidence objects And organize them into a partial chain of evidence based on the type of conclusion. ;

[0059] The verification and conflict detection module is used to verify the local chain of evidence. Perform consistency, applicability, and completeness checks to obtain the verification results. ;

[0060] The confidence assessment module is used to assess the chain of evidence. Verification results The confidence level of the conclusion is determined by scoring. ;

[0061] The explanation generation and result output module is used to generate conclusion verification results. Or explanatory text .

[0062] Thirdly, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the methods described in the first aspect above.

[0063] The beneficial effects of the technical solutions provided in this application include at least the following:

[0064] (1) This application transforms the process of verifying the conclusions of environmental investigation reports, which originally relied on manual page-by-page comparison, into an automated process through a complete processing link of “structured analysis, dual-source retrieval, evidence chain weaving, conflict detection, and confidence assessment”, which significantly improves the efficiency of review. At the same time, by supporting a comprehensive scoring mechanism that supports evidence strength, coverage, conflict penalty, and missing penalty, the credibility of the conclusions is quantified into discrete levels and triggers manual review marking, which solves the problem that the credibility of conclusions is difficult to quantify and assess in the prior art.

[0065] (2) Construct a local evidence chain around a single conclusion to be verified, and uniformly map the heterogeneous evidence scattered in the report text, monitoring forms, attachments and external standard documents into standardized evidence objects and organize them into chains according to the conclusion type, so that the supporting basis, conflict basis and missing items of each conclusion can be explicitly traced and located; through relationship consistency verification, standard adaptation verification and evidence support relationship discrimination, it can automatically identify contradictions, standard mismatches and insufficient coverage between conclusions and evidence, avoiding the risk of missing conflict information in traditional manual review.

[0066] (3) By performing dual-source retrieval in the structured project document collection and the environmental domain knowledge base, the project factual evidence such as monitoring results and point data, as well as the normative evidence such as standard clauses and technical guidelines, are retrieved respectively. The table records, knowledge base clauses and descriptive text are uniformly converted into standardized evidence objects containing evidence type, source, content, direction and weight through evidence standardization processing. Then, the evidence is filled into the corresponding evidence chain template according to the conclusion type. This effectively solves the problem of difficulty in uniformly organizing and reusing evidence caused by different sources and formats of heterogeneous information. Attached Figure Description

[0067] To more clearly illustrate the embodiments of this application or the technical solutions in the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0068] Figure 1 Flowchart of the method for verifying and assessing the conclusions of the environmental investigation report provided in this application embodiment;

[0069] Figure 2A flowchart of dual-source evidence retrieval and local evidence chain construction provided for embodiments of this application;

[0070] Figure 3 A system block diagram of an environmental investigation conclusion verification method based on heterogeneous evidence fusion provided in this application embodiment;

[0071] Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0073] In the description of this application, the terms "comprising," "having," and any variations thereof are intended to cover non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those steps or units that are expressly listed, but may also include other steps or units that are not expressly listed but are inherent to these processes, methods, products, or apparatuses, or steps or units added based on further optimizations conceived in this application.

[0074] This invention proposes a method for verifying and assessing the confidence level of conclusions in environmental survey reports based on heterogeneous evidence fusion. Addressing the problems of scattered evidence, complex standard adaptation, difficulty in identifying conflicting information, and difficulty in quantifying the credibility of conclusions in environmental survey reports on groundwater, surface water, soil, and sediment, this method constructs a processing chain of "document parsing—conclusion structuring—dual-source retrieval—evidence chain construction—verification and detection—confidence assessment—result output." This method can utilize both factual evidence from project documents and normative evidence formed by environmental standards, norms, and guidelines, thereby generating traceable and interpretable verification results around individual conclusions. The specific steps are as follows:

[0075] S101: Obtain environmental investigation reports, appendices, monitoring results, and related project information. It is then converted into a structured collection of project documents using the document parsing module. Build an environmental knowledge base .

[0076] Specifically, this involves obtaining the main body of the environmental survey project report, monitoring result tables, sampling record tables, historical monitoring data, enterprise information, and supporting documentation to form a project document collection. Structured project document collections can be obtained through OCR tools. Simultaneously, it acquires knowledge documents such as laws and regulations, national standards, industry technical guidelines, risk assessment parameters, pollution screening value standards, terminology explanations, and standard applicability relationships to form a domain knowledge base. Then obtain the text of the conclusion to be verified. This serves as the object of subsequent verification. The text of the conclusion to be verified can originate from the conclusion section, pollution analysis section, risk assessment section, or a conclusion paragraph specified by the user in the investigation report. The target output of this invention can be represented as... ,in, Indicates the conclusion Constructed local chain of evidence This indicates the supporting evidence set. This indicates a conflict or weakens the set of evidence. This indicates missing or insufficiently covered information. Indicates the final confidence level of the conclusion;

[0077] S102: Obtain the conclusion to be verified The conclusion is segmented and extracted into one or more structured sub-conclusion objects through the conclusion parsing module. .

[0078] Specifically, the conclusion is segmented and fields are extracted using a large-scale structured conclusion extraction model to obtain one or more structured sub-conclusion objects. Each sub-conclusion object It can be represented as: .

[0079] S103: Based on the structured sub-conclusion object, in the structured project document collection and domain knowledge base A dual-source search is performed to obtain a set of candidate evidence. .

[0080] Specifically, based on the structured sub-conclusion objects, in the structured project document collection... and domain knowledge base The process involves performing a two-source evidence retrieval to obtain a candidate evidence set. This two-source evidence retrieval includes initial recall based on a text vector representation model and fine ranking based on a retrieval re-ranking model. First, a conclusion representation is generated using a vector encoding model: ,in If we represent a text vector model, then the document-side recall results for the project can be expressed as: , The recall results from the knowledge base side can be represented as: , Subsequently, a reordering model was used. Calculate the relevance score, where The overall recall results after reordering candidate evidence based on scores are as follows: .

[0081] S104: The candidate evidence set is processed through the evidence standardization and evidence weaving module. Convert into a standardized set of evidence objects And organize them into a partial chain of evidence based on the type of conclusion. .

[0082] Specifically, regarding the candidate evidence set Evidence standardization is performed to obtain a standardized set of evidence objects: Each standardized evidence object It can be represented as: The evidence types are categorized as {monitoring evidence, standard evidence, causal evidence, spatial evidence}. For well-structured table records and knowledge base entries, they are directly converted into standardized evidence objects through field mapping; for descriptive textual evidence, the evidence type is assigned using an evidence type discrimination model. Based on the current conclusion type, the corresponding evidence chain template is selected, and evidence is populated into the standardized evidence set to construct a local evidence chain. .in, This represents the evidence chain template corresponding to the current conclusion type. To avoid the same fact being counted repeatedly, a normalization key is further constructed: Based on the normalization key, duplicate evidence is either merged or retained as conflict candidates.

[0083] S105: Verify the local evidence chain using the verification and conflict detection module. Perform consistency, applicability, and completeness checks to obtain the verification results. .

[0084] Specifically, regarding the local chain of evidence Perform integrity, consistency, and adaptability checks, and obtain the check results: The validation criteria are defined as follows: "Passed" indicates a satisfied validation condition; "Conflicting" indicates a contradiction between the conclusion and the evidence; "Missing" indicates the absence of key evidence; and "Weak" indicates insufficient evidence coverage but no clear conflict yet. The validation includes field integrity verification, relationship consistency verification, standard adaptation verification, and scope consistency verification. For cases where the rules cannot directly determine the direction of support, an evidence support relationship discrimination model can be invoked, outputting: .

[0085] S106: The chain of evidence is evaluated using the confidence assessment module. Verification results The confidence level of the conclusion is determined by scoring. .

[0086] Specifically, based on the local chain of evidence Validation results Perform a confidence assessment to obtain the confidence level of the conclusion. Construct the following scoring function:

[0087]

[0088] in, This indicates support for the strength of the evidence. Indicates the coverage of key evidence. Indicates conflict punishment. This indicates a lack of punishment. These are the weighting parameters; the strength of supporting evidence can be obtained by summing the weights of the main evidence. .in, Indicates the first The weight of each piece of evidence, Let this be an indicator function. The coverage of key evidence can be expressed as: .in, Indicates the number of key evidence slots that have been filled. This represents the total number of key evidence slots required for this type of conclusion. Finally, continuous scores are mapped to discrete confidence levels: When there are strong conflicts, standard mismatches, or missing key evidence, manual review can be further triggered.

[0089] S107: Generate conclusions and verify results through the interpretation, generation, and output modules. Or explanatory text .

[0090] Specifically, based on the structured conclusion object, local evidence chain, verification results, and confidence level, a readable conclusion verification output is generated: The output includes at least the conclusion, a summary of the main chain of evidence, a description of conflicts and missing information, a confidence level, and whether manual review is recommended.

[0091] For ease of understanding of the present invention, Figure 2 The flowchart of conclusion-driven dual-source evidence retrieval and local evidence chain construction in this invention is shown.

[0092] like Figure 2 The diagram illustrates the process of dual-source evidence retrieval and partial evidence chain construction, which specifically includes:

[0093] S201: Generate a retrieval representation based on the structured conclusion object. Specifically, read the object scope, medium, pollution index, relationship, standard basis, and cause explanation fields from the current structured conclusion object, and combine these fields to form a conclusion retrieval representation. For different types of conclusions, the system generates retrieval representations with different emphases: for standard comparison conclusions, the focus is on retaining the medium, pollution indicators, relationships, and standard basis; for phenomenon conclusions, the focus is on retaining the scope of the object, the medium, and state relationships; for causal explanation conclusions, the focus is on retaining the pollution indicators, the scope of the object, and the causal explanation. Through this step, natural language conclusions are converted into a unified representation that can be used for retrieval.

[0094] S202: Perform dual-source evidence recall based on conclusion retrieval representation. Specifically, generate conclusion representation using a vector coding model: .in Representing a text vector model, subsequently in a structured project document collection. and domain knowledge base The vector recall process yields a set of candidate evidence for project facts and a set of candidate evidence for normative knowledge. and Among them, candidate evidence of project facts mainly includes monitoring records, summary results, site descriptions, and regional descriptions, while candidate evidence of normative knowledge mainly includes standard provisions, terminology explanations, applicable relationships, and mechanistic knowledge.

[0095] ,

[0096] The recall results from the knowledge base can be represented as:

[0097] ,

[0098] ,

[0099] On the project document side, the main recall is for retrieving factual evidence of the project, such as monitoring results, site data, descriptions of pollutants, and regional descriptions. On the knowledge base side, the main recall is for normative evidence, such as standard clauses, technical guidelines, terminology explanations, and adaptation rules.

[0100] Subsequently, a reordering model was used. Calculate the relevance score, where The overall recall results after reordering candidate evidence based on scores are as follows:

[0101] .

[0102] S203: Perform field constraint screening and standardized representation on the recalled evidence. Specifically, based on the fields of medium, pollution index, relationship, standard basis, and object scope in the current structured conclusion object, perform field constraint screening on the candidate evidence obtained in step S202, retaining candidate evidence that is consistent with or partially consistent with the current conclusion fields; then, use the retrieval re-ranking model to calculate the relevance score between the candidate evidence and the current structured conclusion object. ,in As candidate evidence, This represents the retrieval re-ranking model; subsequently, the filtered candidate evidence is uniformly mapped to standardized evidence objects. The weight of evidence is calculated as follows: , Indicates the relevance score. Indicates the credibility of the source. Indicates the completeness of the field.

[0103] S204: Construct a local chain of evidence based on the conclusion type. Specifically, select the corresponding evidence chain template according to the conclusion type of the current structured conclusion object, and fill the standardized evidence objects obtained in step S203 into the corresponding slots to form a local chain of evidence: ,in, This represents the evidence chain template corresponding to the current conclusion type. Represents a standardized set of evidence objects To avoid the same fact being counted twice, a normalization bond is further constructed: The duplicate evidence is merged or retained as a conflict candidate based on the normalization key.

[0104] This invention proposes a method for verifying and assessing the confidence of environmental investigation report conclusions based on heterogeneous evidence fusion. By performing structured analysis, dual-source evidence retrieval, local evidence chain construction, and confidence scoring on multi-source heterogeneous information from the main text of the environmental investigation report, monitoring tables, appendices, and domain knowledge base, this method achieves automated verification and credibility assessment of environmental investigation report conclusions while ensuring traceability of the basis for conclusions, verifiability of standard adaptation, and identifiable conflict information. This improves the efficiency of conclusion review and the interpretability of results, meeting the intelligent verification needs of environmental investigation scenarios such as groundwater, surface water, soil, and sediment. This promotes the standardization, structuring, and intelligentization of investigation report analysis and conclusion review in related fields.

[0105] like Figure 3 This application also provides an environmental investigation conclusion verification system based on heterogeneous evidence fusion, which may include:

[0106] The document parsing module is used to obtain environmental investigation reports, appendices, monitoring results, and related project information. And convert it into a structured collection of project documents. Build an environmental knowledge base ;

[0107] The conclusion parsing module is used to obtain the conclusions to be verified. The conclusion to be verified Segment and extract into one or more structured sub-conclusion objects ;

[0108] The dual-source retrieval module is used to retrieve information based on the structured sub-conclusion object. In the structured project document collection and domain knowledge base A dual-source search is performed to obtain a set of candidate evidence. ;

[0109] The evidence standardization and evidence weaving module is used to organize candidate evidence sets. Convert into a standardized set of evidence objects And organize them into a partial chain of evidence based on the type of conclusion. ;

[0110] The verification and conflict detection module is used to verify the local chain of evidence. Perform consistency, applicability, and completeness checks to obtain the verification results. ;

[0111] The confidence assessment module is used to assess the chain of evidence. Verification results The confidence level of the conclusion is determined by scoring. ;

[0112] The explanation generation and result output module is used to generate conclusion verification results. Or explanatory text .

[0113] Specific limitations regarding the environmental investigation conclusion verification system based on heterogeneous evidence fusion can be found in the limitations of the environmental investigation conclusion verification method based on heterogeneous evidence fusion mentioned above, and will not be repeated here. Each module in the aforementioned environmental investigation conclusion verification system based on heterogeneous evidence fusion can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0114] In one embodiment, an electronic device is provided, which may be a computer, and its internal structure diagram may be as follows: Figure 4As shown, the electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device is used for verifying environmental survey conclusions. The network interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for verifying environmental survey conclusions.

[0115] Those skilled in the art will understand that, Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0116] In one embodiment of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described method for verifying environmental investigation conclusions based on heterogeneous evidence fusion.

[0117] In one embodiment of this application, a computer program product is provided, including a computer program / instructions, which, when executed by a processor, implements the steps of the above-described method for verifying environmental investigation conclusions based on heterogeneous evidence fusion.

[0118] The computer-readable storage medium and computer program product provided in this embodiment are similar in implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0119] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods.

[0120] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0121] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for verifying environmental investigation conclusions based on heterogeneous evidence fusion, characterized in that, The method includes: Step 1: Obtain the environmental investigation report, appendices, monitoring results, and related project data. And convert it into a structured collection of project documents. Build an environmental knowledge base ; Step 2: Obtain the conclusion to be verified The conclusion to be verified Segment and extract into one or more structured sub-conclusion objects ; Step 3: Based on the structured sub-conclusion object In the structured project document collection and domain knowledge base A dual-source search is performed to obtain a set of candidate evidence. ; Step 4: Set up candidate evidence Convert into a standardized set of evidence objects And organize them into a partial chain of evidence based on the type of conclusion. ; Step 5: Analyze the local chain of evidence. Perform consistency, applicability, and completeness checks to obtain the verification results. ; Step 6: Analyze the chain of evidence. Verification results The confidence level of the conclusion is determined by scoring. ; Step 7: Generate conclusion verification results Or explanatory text .

2. The method for verifying environmental investigation conclusions based on heterogeneous evidence fusion according to claim 1, characterized in that, In step 1, the environmental survey report, appendices, monitoring results, and related project data are obtained. And convert it into a structured collection of project documents. Build an environmental knowledge base Specifically: The project's original document collection is The document includes the main body of the investigation report, monitoring results tables, sampling records, enterprise information, historical monitoring records, and appendices, and is a structured project document collection obtained through OCR tools. Domain knowledge base is This includes domain standards, industry norms, technical guidelines, risk assessment parameters, knowledge of pollution migration mechanisms, and historical case studies. The conclusions to be verified are... The target output is then represented as ,in, Indicates the conclusion Constructed local chain of evidence This indicates the supporting evidence set. This indicates a conflict or weakens the set of evidence. This indicates missing or insufficiently covered information. Indicates the final confidence level of the conclusion.

3. The method for verifying environmental investigation conclusions based on heterogeneous evidence fusion according to claim 1, characterized in that, In step 2, obtain the conclusion to be verified. The conclusion to be verified Segment and extract into one or more structured sub-conclusion objects Specifically: The conclusion is segmented and its fields are extracted using a structured conclusion extraction model to obtain one or more structured sub-conclusion objects. ; where each sub-conclusion object Represented as: .

4. The method for verifying environmental investigation conclusions based on heterogeneous evidence fusion according to claim 1, characterized in that, In step 3, based on the structured sub-conclusion object In the structured project document collection and domain knowledge base A dual-source search is performed to obtain a set of candidate evidence. Specifically: Based on the structured sub-conclusion objects, respectively in the structured project document collection and domain knowledge base The process involves performing dual-source evidence retrieval to obtain a set of candidate evidence; the dual-source evidence retrieval includes an initial recall based on a text vector representation model and a fine ranking process based on a retrieval re-ranking model. First, a conclusion representation is generated using a vector encoding model: ,in If we represent a text vector model, then the recall results on the project document side are represented as follows: , ; The recall results from the knowledge base are represented as follows: , On the project document side, the main focus is on recalling factual evidence related to the project, such as monitoring results, site data, descriptions of pollutants, and regional descriptions. On the knowledge base side, the main focus is on recalling normative evidence such as standard clauses, technical guidelines, terminology explanations, and adaptation rules. Subsequently, a reordering model was used. Calculate the relevance score, where The overall recall results after reordering candidate evidence based on scores are as follows: .

5. The method for verifying environmental investigation conclusions based on heterogeneous evidence fusion according to claim 1, characterized in that, In step 4, the candidate evidence set is... Convert into a standardized set of evidence objects And organize them into a partial chain of evidence based on the type of conclusion. Specifically: For candidate evidence set Evidence standardization is performed to obtain a standardized set of evidence objects: Each standardized evidence object It can be represented as: ； The types of evidence include {monitoring evidence, standard evidence, causal evidence, spatial evidence}. For well-structured table records and knowledge base entries, they are directly converted into standardized evidence objects through field mapping; for descriptive text evidence, the evidence type is assigned through an evidence type discrimination model. Select the corresponding evidence chain template based on the current conclusion type, and perform evidence filling on the standardized evidence set to construct a local evidence chain: in, This represents the evidence chain template corresponding to the current conclusion type; to avoid the same fact being counted repeatedly, a normalization key is further constructed: Based on the normalization key, duplicate evidence is either merged or retained as conflict candidates.

6. The method for verifying environmental investigation conclusions based on heterogeneous evidence fusion according to claim 1, characterized in that, In step 5, the local chain of evidence is... Perform consistency, applicability, and completeness checks to obtain the verification results. Specifically: Local chain of evidence Perform integrity, consistency, and adaptability checks, and obtain the check results: Among them, the "pass" item indicates that the verification conditions have been met, the "conflict" item indicates that there is a contradiction between the conclusion and the evidence, the "missing" item indicates that the key evidence is missing, and the "weak" item indicates that the evidence coverage is insufficient but no clear conflict has yet been formed. The verification includes field integrity verification, relationship consistency verification, standard adaptation verification, and range consistency verification. For cases where the supported direction is difficult to determine directly from the rules, an evidence-supported relationship discrimination model can be invoked, and the output will be: .

7. The method for verifying environmental investigation conclusions based on heterogeneous evidence fusion according to claim 1, characterized in that, In step 6, the chain of evidence is... Verification results The confidence level of the conclusion is determined by scoring. Specifically: According to the partial chain of evidence Validation results Perform a confidence assessment to obtain the confidence level of the conclusion; construct the following scoring function: in, This indicates support for the strength of the evidence. Indicates the coverage of key evidence. Indicates conflict punishment. This indicates a lack of punishment. These are weight parameters; The strength of supporting evidence can be obtained by summing the weights of the main evidence: in, Indicates the first The weight of each piece of evidence, For indicator functions; The coverage of key evidence can be expressed as: in, Indicates the number of key evidence slots that have been filled. This indicates the total number of key evidence slots required for the conclusion; Ultimately, continuous scores are mapped to discrete confidence levels: When there are strong conflicts, standard mismatches, or missing key evidence, manual review is further triggered.

8. The method for verifying environmental investigation conclusions based on heterogeneous evidence fusion according to claim 1, characterized in that, In step 7, the conclusion verification results are generated. Or explanatory text Specifically: Based on the structured conclusion object, local chain of evidence, verification results, and confidence level, generate readable conclusion verification output: The output includes at least the conclusion, a summary of the main chain of evidence, a description of conflicts and missing information, a confidence level, and whether manual review is recommended.

9. A system for verifying environmental investigation conclusions based on heterogeneous evidence fusion, characterized in that, The system includes: The document parsing module is used to obtain environmental investigation reports, appendices, monitoring results, and related project information. And convert it into a structured collection of project documents. Build an environmental knowledge base ; The conclusion parsing module is used to obtain the conclusions to be verified. The conclusion to be verified Segment and extract into one or more structured sub-conclusion objects ; The dual-source retrieval module is used to retrieve information based on the structured sub-conclusion object. In the structured project document collection and domain knowledge base A dual-source search is performed to obtain a set of candidate evidence. ; The evidence standardization and evidence weaving module is used to organize candidate evidence sets. Convert into a standardized set of evidence objects And organize them into a partial chain of evidence based on the type of conclusion. ; The verification and conflict detection module is used to verify the local chain of evidence. Perform consistency, applicability, and completeness checks to obtain the verification results. ; The confidence assessment module is used to assess the chain of evidence. Verification results The confidence level of the conclusion is determined by scoring. ; The explanation generation and result output module is used to generate conclusion verification results. Or explanatory text .

10. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program that, when executed by the processor, implements the method as described in any one of claims 1 to 8.

Images