Optimal transmission method based on structural tree logic and semantic hybrid constraint

Abstract

The present application relates to the technical field of data analysis, and particularly relates to an optimal transmission method based on structural tree logic and semantic hybrid constraint. The method comprises: performing logic layout analysis on an ESG report to obtain a chapter structure tree; obtaining a logic path of each fact node in the ESG report according to the chapter structure tree, and predicting a logic path of each report comment in the comment information according to the chapter structure tree; taking a probability distribution formed by a plurality of report comments as a source distribution, taking a probability distribution formed by a plurality of fact nodes as a target distribution, performing optimal transmission analysis, and obtaining report comment mapping information, wherein a transmission cost between the fact nodes and the report comments is determined according to semantic distances of the fact nodes and the report comments and logic distances of the fact nodes and the report comments. The present application can realize effective utilization of network comments and ESG reports, and improve the accuracy of evaluation on long-term value of enterprises.

Description

Technical Field

[0001] This invention relates to the technical field of data analysis, specifically to an optimal transmission method based on a hybrid constraint of structure tree logic and semantics. Background Technology

[0002] With the deepening consensus on sustainable development, corporate environmental, social, and governance (ESG) disclosures have evolved from voluntary disclosures to a core standard for measuring a company's long-term value in the public eye. Furthermore, verifying the accuracy of ESG reports is crucial for maintaining capital market order and protecting investor rights. Faced with increasingly stringent regulatory requirements and a complex business ecosystem, ensuring compliant information disclosure has become a key link in building a healthy ESG ecosystem. However, existing verification methods mainly rely on periodic manual audits or basic natural language processing technologies based on shallow feature engineering. These traditional methods reveal numerous limitations when dealing with massive, multi-source, heterogeneous data environments, such as high audit costs, significant verification delays, difficulty in covering the entire dataset, and insufficient understanding of deep semantic logic. In addition, while social media and online comments, as important unstructured data sources reflecting actual corporate performance, contain rich real-time monitoring value, their fragmented, noisy, and non-standardized characteristics create a significant semantic gap with rigorous official ESG reports, further increasing the complexity of cross-modal consistency verification and fact-alignment.

[0003] The current lack of effective use of online comments and ESG reports leads to inaccurate assessments of a company's long-term value. Summary of the Invention

[0004] The purpose of this invention is to provide an optimal transmission method based on a hybrid constraint of structure tree logic and semantics, in order to solve the technical problem of inaccurate assessment of the long-term value of enterprises in existing technologies.

[0005] In a first aspect, one embodiment of the present invention provides an optimal transmission method based on a hybrid constraint of structure tree logic and semantics, the method comprising: Logical format parsing of the ESG report yields a chapter structure tree; The logical path of each fact node in the ESG report is obtained based on the chapter structure tree, and the logical path of each report comment in the comment information is predicted based on the chapter structure tree. Using the probability distribution formed by multiple report comments as the source distribution and the probability distribution formed by multiple fact nodes as the target distribution, optimal transmission analysis is performed to obtain report comment mapping information. The transmission cost between the fact node and the report comment is determined based on their semantic distance and logical distance. The semantic distance is used to indicate the degree of difference between the semantics of the corresponding fact node and the semantics of the corresponding report comment, and the logical distance is used to indicate the degree of deviation between the logical path of the corresponding fact node and the logical path of the corresponding report comment.

[0006] Optionally, the chapter structure tree includes multiple logical layers, each logical layer includes multiple logical nodes, the logical path is used to indicate the logical nodes associated with the corresponding fact node or report comment in each corresponding logical layer, the logical distance is obtained by weighting multiple layer distances jointly corresponding to the corresponding fact node and the corresponding report comment, and the layer distance is used to indicate the difference between the logical nodes associated with the corresponding fact node and the corresponding report comment in the corresponding logical layer.

[0007] Optionally, the calculation weight of the layer distance is negatively correlated with the layer depth of the corresponding logical layer.

[0008] Optionally, the rate of change of the number of the plurality of logical layers and the calculation weight of the layer distance are negatively correlated.

[0009] Optionally, the logical path of the report comments is obtained based on predictions from a large language model, and the document structure tree is the prior background knowledge of the large language model.

[0010] Optionally, after obtaining the report comment mapping information, the method further includes: In the multiple mapping groups included in the report comment mapping information, the transmission cost of each mapping group is obtained, wherein one of the mapping groups includes one fact node and one report comment; The transmission costs of multiple mapping groups are weighted to obtain a report-comment deviation value, wherein the weight of the transmission cost of the mapping group is the matching probability between the fact nodes included in the mapping group and the report-comment. The report-comment deviation value is used to characterize the degree of content difference between the ESG report and the comment information.

[0011] Optionally, after obtaining the report comment mapping information, the method further includes: In the multiple mapping groups included in the report comment mapping information, the transmission cost of each mapping group is obtained, wherein one of the mapping groups includes one fact node and one report comment; The transmission cost of different mapping groups associated with the same fact node is analyzed to determine the node comment deviation value of each fact node, wherein the node comment deviation value is used to characterize the degree of content difference between the corresponding fact node and the comment information.

[0012] Optionally, after analyzing the transmission costs of different mapping groups associated with the same fact nodes to determine the node comment deviation value for each fact node, the method further includes: Fact nodes whose comment deviation values ​​exceed the node comment deviation threshold are highlighted.

[0013] Optionally, the method further includes: In the multiple report comments, the transmission cost of each report comment and each event node is analyzed to obtain the multiple event transmission costs for each report comment; Based on the multiple event transmission costs of each report comment, isolated comments are identified among the multiple report comments, wherein the multiple event transmission costs of the isolated comment are all greater than or equal to a cost threshold, and the isolated comment is prohibited from participating in the optimal transmission analysis.

[0014] Optionally, after identifying isolated comments among the multiple reported comments based on the multiple event transmission costs for each reported comment, the method further includes: The probability distribution of isolated comments among multiple report comments is statistically analyzed to obtain the report missing index.

[0015] Secondly, another embodiment of the present invention provides an optimal transmission system based on a hybrid constraint of structure tree logic and semantics, the system comprising: The report parsing module is used to parse the logical layout of ESG reports and obtain the chapter structure tree. The logical path acquisition module is used to acquire the logical path of each fact node in the ESG report based on the chapter structure tree, and to predict the logical path of each report comment in the comment information based on the chapter structure tree. The report comment alignment module is used to perform optimal transmission analysis using a probability distribution composed of multiple report comments as the source distribution and a probability distribution composed of multiple fact nodes as the target distribution to obtain report comment mapping information. The transmission cost between the fact node and the report comment is determined based on their semantic distance and logical distance. The semantic distance is used to indicate the degree of difference between the semantics of the corresponding fact node and the semantics of the corresponding report comment, and the logical distance is used to indicate the degree of deviation between the logical path of the corresponding fact node and the logical path of the corresponding report comment.

[0016] Thirdly, in another embodiment of the present invention, an electronic device is provided, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method described in the first aspect.

[0017] Fourthly, in another embodiment of the present invention, 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 method described in the first aspect.

[0018] The present invention has the following beneficial effects: By analyzing the logical format of an enterprise's ESG report, we can digitize the topological relationships between different contents included in the report. Based on this, we can obtain the logical path of each fact node in the ESG report and predict the logical path of each online comment in the comment information. This allows us to use the logical architecture of the ESG report to supplement the logical constraints of fact nodes and report comments, ensuring that the global deep logic of the ESG report can participate in the alignment process between fact nodes and report comments. Then, by combining the content semantics of fact nodes and report comments, we can comprehensively determine the mapping relationship between fact nodes and report comments from both deep logic and surface semantics perspectives. This enables the effective use of online comments and ESG reports, thereby improving the accuracy of the long-term value of the enterprise assessed based on this data. Attached Figure Description

[0019] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an optimal transmission method based on a hybrid constraint of structure tree logic and semantics provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of an optimal transmission system based on a hybrid constraint of structure tree logic and semantics, provided by an embodiment of the present invention. Figure 3 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an optimal transmission method based on a hybrid constraint of structure tree logic and semantics proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0023] The following description, in conjunction with the accompanying drawings, details the specific scheme of an optimal transmission method based on a hybrid constraint of structure tree logic and semantics provided by this invention.

[0024] In one embodiment, the present invention provides an optimal transmission method based on a hybrid constraint of structure tree logic and semantics, such as... Figure 1 As shown, the method includes: Step S1: Perform logical format parsing on the ESG report to obtain the chapter structure tree.

[0025] Specifically, this invention uses document intelligence technology (such as LayoutLM or PaddleStructure) to perform logical layout parsing on the enterprise's ESG report in order to obtain the chapter structure tree.

[0026] The chapter structure tree can be understood as data stored in a tree structure, representing the hierarchical relationship of the content of the ESG report (similar to the "chapter-section-segment" format), and the chapter structure tree includes multiple logical layers, each of which includes multiple logical nodes.

[0027] It should be understood that the smaller the depth of the logical layer in the chapter structure tree, the higher the importance of the corresponding logical layer in the logical chain of the ESG report.

[0028] Step S2: Obtain the logical path of each fact node in the ESG report based on the chapter structure tree, and predict the logical path of each report comment in the comment information based on the chapter structure tree.

[0029] Specifically, a fact node is the smallest verifiable, quantifiable, and associative unit of disclosed fact extracted from an ESG report, and the logical path is used to indicate the logical nodes associated with the corresponding fact node or report comment at each corresponding logical layer.

[0030] For example, a fact node in an ESG report could be "The nitrogen oxide (NOx) emissions from the North China production base in 2025 were 38.6 tons, a 12% decrease compared to 2024, and the emission concentration met the limit requirements of the Integrated Emission Standard for Air Pollutants (GB16297-1996)", with the corresponding logical path being " / environment / exhaust gas / emissions".

[0031] By using the companies associated with the ESG reports as search keywords, and conducting content searches on various social media platforms, the aforementioned comment information can be obtained. These comment texts are related to the companies associated with the ESG reports.

[0032] In this invention, the logical path of the report comments is obtained based on prediction by a large language model, and the document structure tree is the prior background knowledge of the large language model.

[0033] Specifically, for each report comment, the chapter structure tree obtained in the previous steps is used as prior background knowledge input into the large language model through a preset Prompt template. Then, by utilizing the semantic understanding and common sense reasoning capabilities of the large language model, the model is instructed to automatically determine the logical affiliation of the report comment without manual annotation, thus predicting the logical path of each report comment.

[0034] Step S3: Using the probability distribution formed by the multiple reports and comments as the source distribution and the probability distribution formed by the multiple fact nodes as the target distribution, perform optimal transmission analysis to obtain report and comment mapping information.

[0035] The transmission cost between the fact node and the report comment is determined based on their semantic distance and logical distance. The semantic distance indicates the degree of difference between the semantics of the corresponding fact node and the semantics of the corresponding report comment, and the logical distance indicates the degree of deviation between the logical path of the corresponding fact node and the logical path of the corresponding report comment.

[0036] In this invention, a source distribution is defined. : Represents the probability quality distribution of multiple report comments.

[0037] Source distribution It can be initialized as a uniform distribution, that is: In the above formula, Representatives' comments on the report The probability quality is such that each report comment has equal attention weight in the initial state. This represents the number of comments across multiple reports.

[0038] The logical path for defining report comments is as follows: ,in, Commentary on reports representing predictions from large language models In the The logical nodes corresponding to the logical layer, Represents the number of multiple logical layers. Less than or equal to Positive integers.

[0039] Define target distribution : Represents the probability quality distribution of multiple fact nodes.

[0040] Target distribution The weighted setting can be combined with the text length of each fact node or the layer depth of the logical layer corresponding to its occurrence position, that is: In the above formula, Representing fact nodes The preset probability quality is the same for different fact nodes. Representing fact nodes The corresponding probability quality correction coefficient, the value of which depends on the fact node. The probability quality of a fact node is determined by the length of the text or the depth of the logical layer corresponding to its occurrence position. The longer the text or the deeper the logical layer corresponding to its occurrence position, the greater the probability quality of the fact node. Represents the number of multiple fact nodes.

[0041] The logical path for defining a fact node is: ,in, Representing fact nodes In the The logical nodes corresponding to the logical layer, Represents the number of multiple logical layers. Less than or equal to Positive integers.

[0042] The optimal transmission analysis operation performed in this invention can be understood as: finding a global transmission plan matrix (used to form report comment mapping information) such that the source distribution... Transport to target distribution Minimize the total cost (calculated based on transmission cost).

[0043] The aforementioned report comment mapping information includes multiple mapping groups, and each mapping group includes one fact node and one report comment.

[0044] Specifically, the objective function corresponding to the optimal transmission analysis can be expressed as: In the above formula, This represents the set of all feasible transmission plans that satisfy the marginal distribution constraints, i.e. and ; Indicates the first in the corresponding transmission plan The report comment matched the first The probability weights of each fact node. This indicates that the corresponding transmission plan will include the first... The report comments are aligned to the first The semantic and logical resistance (i.e., transmission cost) of each fact node.

[0045] Furthermore, the transmission cost can be expressed as: In the above formula, This indicates the adjustment of hyperparameters to balance the impact of semantic distance and logical distance on transmission cost. This represents a normalized activation function (such as the Tanh function or the Sigmoid function) used to map the logistic distance to... The interval was adjusted to align with the semantic distance unit. Indicates the first Article 1 Report Comments and No. The semantic distance between fact nodes, Indicates the first Article 1 Report Comments and No. The logical distance between each fact node.

[0046] In this invention, cosine similarity is specifically used to quantify the semantic distance between report comments and fact nodes, i.e. It can be represented as: In the above formula, Indicates the first The report comments are encoded into high-dimensional embedding vectors by a pre-trained language model. Indicates the first The high-dimensional embedding vectors are obtained by encoding each fact node using a pre-trained language model.

[0047] In this invention, the logical distance is obtained by weighting multiple layer distances corresponding to the corresponding fact node and the corresponding report comment. These layer distances are used to indicate the differences between the logical nodes associated with the corresponding fact node and the corresponding report comment at the corresponding logical layer. Therefore, It can be represented as: In the above formula, The maximum hierarchical depth of the chapter structure tree, that is, the number of multiple logical layers (e.g., H=3 when the content hierarchy is three levels: chapter-section-segment, and H=4 when it is four levels: chapter-section-segment-item). and : Representing report comments respectively Logical path and fact nodes The logical path in the first Logical nodes corresponding to the logical layer; Represents an indicator function, when At (i.e., at the time) If a logical bifurcation occurs at the logical layer, the function value is 1 (i.e., the layer distance is 1), indicating a path penalty; otherwise, the function value is 0 (i.e., the layer distance is 0). Representing the The computation weight corresponding to the logic layer (e.g., setting the same computation weight for multiple logic layers).

[0048] The above setup parses the logical layout of an enterprise's ESG report to digitize the topological relationships between different contents included in the report. Based on this, it obtains the logical path of each fact node in the ESG report and predicts the logical path of each online comment in the commentary information. This leverages the logical architecture of the ESG report to supplement the logical constraints between fact nodes and report comments, ensuring that the global deep logic of the ESG report participates in the alignment process. Then, by combining the semantics of the fact nodes and report comments, it comprehensively determines the mapping relationship between fact nodes and report comments from both deep logic and surface semantics perspectives. This enables the effective utilization of online comments and the ESG report, thereby improving the accuracy of the long-term enterprise value assessment obtained from this data.

[0049] In some implementations, the calculation weight of the layer distance is negatively correlated with the layer depth of the corresponding logical layer.

[0050] In this implementation, by implementing the above-mentioned limitations, a strategy of severely punishing top-level logical errors and lightly penalizing bottom-level detail errors is adopted to improve the accuracy and reliability of the calculated logical distance.

[0051] Specifically, it can be set ,definition For the weighted benchmark coefficient, This is a hierarchy attenuation coefficient, used to amplify the impact of upper-level logic errors on the numerical value of logic distance, and reduce the impact of lower-level logic errors on the numerical value of logic distance, thereby supporting the implementation of the above strategy.

[0052] Furthermore, the rate of change of the number of the multiple logical layers and the calculation weight of the layer distance are negatively correlated; that is, the more multiple logical layers there are, the smaller the rate of change of the calculation weight of the layer distance may be.

[0053] By implementing the above limitations, we can achieve flexible adaptation between the calculation weight of layer distance and ESG reports with different content layer layouts, and ensure the effective implementation of the above strategies.

[0054] Specifically, in setting In this case, you can set it to work when H≤3 (short document / shallow level). The value ranges from 0.8 to 1.0. The value is set to 0.6-0.8 to enhance the penalty weight of the top layer (k=1, 2) and avoid cross-chapter errors; while setting it to... When the document level is >3 (long document / deep level), The value ranges from 0.6 to 0.8. The value should be between 0.3 and 0.5 to appropriately reduce the bottom layer ( A penalty weight of ≥ 3) is used to avoid over-constraining the details of the differences.

[0055] Furthermore, since the computational complexity of directly solving the aforementioned linear programming problem (the objective function corresponding to the optimal transport analysis) is too high, this invention introduces entropy regularization technology to transform the original problem into an approximate problem that can be quickly solved by the Sinkhorn-Knopp algorithm, thereby improving the output efficiency of report comment mapping information.

[0056] The objective function after introducing regularization can be expressed as: In the above formula, For Shannon entropy, The regularization coefficient is used to control the smoothness of the matching results and avoid overfitting to noisy data. The solution is the transmission plan. The iterative calculation formula.

[0057] According to the Lagrange multiplier method, the optimal solution has the following form: In the above formula, This refers to the Gibbs core, specifically, the transmission cost. The affinity matrix is ​​derived from (matrix form).

[0058] It can be represented as: Based on the above settings, to reduce transmission costs If it is too large (e.g., due to serious semantic conflicts or path errors), then let The value approaches 0, thus automatically blocking incorrect transmission paths and avoiding misalignment caused by isolated matching.

[0059] The scaling factor in the optimal solution formula is a core parameter for the Sinkhorn-Knopp algorithm's iterative solution. Its core meaning is to dynamically adjust the probability weights of the source and target distributions to ensure that, while satisfying marginal constraints, the globally optimal mapping relationship that minimizes the total transmission cost is found. Reflects the first The importance of a report comment in the global matching reflects the overall matching affinity between that report comment and the fact nodes in the target domain; while Reflects the first The capacity of a fact node in global matching, and correspondingly, its size reflects the overall matching affinity (a decimal value between 0 and 1) between the node and the report comments of the source domain.

[0060] Specifically, the Sinkhorn iterative formula is used to alternately update the formula until convergence: Among them, the constraints are: , .

[0061] Iteration termination condition: and .

[0062] Optionally, after obtaining the report comment mapping information, the method further includes: In the multiple mapping groups included in the report comment mapping information, the transmission cost of each mapping group is obtained, wherein one of the mapping groups includes one fact node and one report comment; The transmission costs of multiple mapping groups are weighted to obtain a report-comment deviation value, wherein the weight of the transmission cost of the mapping group is the matching probability between the fact nodes included in the mapping group and the report-comment. The report-comment deviation value is used to characterize the degree of content difference between the ESG report and the comment information.

[0063] The larger the report comment deviation value, the greater the difference between the ESG report and the comment information, which in turn indicates a more significant discrepancy between the ESG report and actual public feedback. This means that the credibility of the ESG report is lower, and the probability that the company corresponding to the ESG report needs to be closely investigated is higher.

[0064] Specifically, report commentary deviation value It can be represented as: in, For the optimal transmission plan (i.e., report comment mapping information), the th Article 1 Report Comments and No. The matching probability of each fact node. The semantic and logical resistances are represented in the cross-modal cost matrix (i.e., the matrix representation of the transmission costs of the multiple mapping groups).

[0065] In practical applications, a report comment deviation threshold can be set based on experience. When the calculated report comment deviation value is greater than or equal to the report comment deviation threshold, a risk warning message will be output prompting a focused review of the company corresponding to the ESG report.

[0066] Optionally, after obtaining the report comment mapping information, the method further includes: In the multiple mapping groups included in the report comment mapping information, the transmission cost of each mapping group is obtained, wherein one of the mapping groups includes one fact node and one report comment; The transmission cost of different mapping groups associated with the same fact node is analyzed to determine the node comment deviation value of each fact node, wherein the node comment deviation value is used to characterize the degree of content difference between the corresponding fact node and the comment information.

[0067] By outputting the aforementioned node comment deviation values, corresponding comment consistency checks are performed on each fact node in the ESG report. Combined with comment information, fine-grained review of each fact node in the ESG report is achieved, facilitating the review of the ESG report and providing auditors with a traceable and interpretable chain of evidence.

[0068] Specifically, node comment deviation value It can be represented as: In the above formula, the numerator is the weighted sum of the transmission costs of different mapping groups associated with the corresponding fact node, and the denominator is the sum of the matching probabilities corresponding to the fact node. Weight normalization is used to eliminate the interference of the difference in the number of comments on risk assessment.

[0069] Furthermore, after analyzing the transmission costs of different mapping groups associated with the same fact nodes to determine the node comment deviation value for each fact node, the method further includes: Fact nodes whose comment deviation values ​​exceed the node comment deviation threshold are highlighted.

[0070] Define node comment deviation threshold (This threshold can be dynamically adjusted according to industry regulatory requirements and report type. For example, the threshold τ for environmental reports can be set to 0.7, and the threshold for social responsibility reports can be set to...) It can be set to 0.65), if > If a fact node is deemed to have a "potential greenwashing" or "missing disclosure" risk, the system will identify it as such. The former refers to a semantic conflict between the compliance claims in the report (e.g., "emissions meet standards") and numerous negative comments (e.g., "factory exhaust fumes are pungent"), while the latter refers to a situation where the issues raised in the comments (e.g., "solid waste is disposed of arbitrarily") lack a corresponding disclosure node in the report. The system will highlight this node in the visualization interface, allowing auditors to view the corresponding report comment set, path matching details, and other relevant information by clicking on the highlighted fact node. The calculation process achieves an audit closed loop of risk identification, evidence tracing, and conclusion acceptance.

[0071] In the application, a visual evidence heatmap can be generated based on the report comment deviation value and the comment deviation values ​​of multiple nodes: using the "chapter-section-segment" of the chapter structure tree as the spatial dimension and the node comment deviation value as the intensity dimension (using a four-color gradient of red, orange, yellow and green, with red representing high risk and green representing low risk), it intuitively presents the risk distribution of each fact node in the ESG report, helping auditors quickly focus on high-risk areas and improve verification efficiency.

[0072] Optionally, the method further includes: In the multiple report comments, the transmission cost of each report comment and each event node is analyzed to obtain the multiple event transmission costs for each report comment; Based on the multiple event transmission costs of each report comment, isolated comments are identified among the multiple report comments, wherein the multiple event transmission costs of the isolated comment are all greater than or equal to a cost threshold, and the isolated comment is prohibited from participating in the optimal transmission analysis.

[0073] In the above settings, filtering isolated comments can suppress interference from noisy comments that are irrelevant to the ESG report content, reduce the amount of data to be processed during optimal transport analysis, and improve the output efficiency of report comment mapping information.

[0074] In addition, given the fragmented and noisy nature of online comments on social media platforms, introducing asymmetric constraints during transmission can enhance the solution's resistance to interference in complex social environments and improve the accuracy and reliability of the output report's comment mapping information.

[0075] Specifically, when constructing the Unbalanced Optimal Transport (UOT) objective function (i.e., the objective function after introducing regularization) to perform the global optimal transport solution, asymmetric optimal transport constraints are introduced simultaneously. The original strict marginal distribution constraints are relaxed to a divergence-based penalty term. Specifically, a KL divergence penalty term is added to the transport objective function. The regularized and reconstructed optimization objective function is as follows: in, The sum of rows in the transmission plan matrix represents the sum of the probabilistic qualities of several report comments that actually participate in the alignment; This represents the probability quality distribution corresponding to multiple report comments; This is the edge constraint relaxation coefficient, used to control the model's tolerance to noise, and also to determine the aforementioned cost threshold.

[0076] The above asymmetric filtering ensures the final generated optimal transmission plan. (i.e., report comment mapping information) will only retain non-isolated web comments with clear semantic references and logical connections.

[0077] Furthermore, after identifying isolated comments among the multiple reported comments based on the multiple event transmission costs of each reported comment, the method further includes: The probability distribution of isolated comments among multiple report comments is statistically analyzed to obtain the report missing index.

[0078] In the above setup, by statistically analyzing the probability quality of several isolated comments identified in the probability quality distribution of multiple report comments, the proportion of the probability quality of the isolated comments in the entire probability quality distribution is determined and used as the report missing index. This is then used to calculate the quality loss of comments mentioned in online comments but not in ESG reports, providing mathematical support for quantifying the hidden risks that companies avoid mentioning in ESG reports.

[0079] In summary, the present invention effectively overcomes the semantic misjudgment of polysemous words and metaphors in traditional vector retrieval models without requiring manual annotation, solves the problem of logical collapse in long document verification, and significantly improves the robustness, accuracy and interpretability of identifying greenwashing risks.

[0080] In one embodiment, the present invention also provides an optimal transmission system 200 based on a hybrid constraint of structure tree logic and semantics, such as... Figure 2 As shown, the optimal transmission system 200 based on a hybrid constraint of structure tree logic and semantics includes: The report parsing module 201 is used to parse the logical layout of the ESG report to obtain the chapter structure tree. The logical path acquisition module 202 is used to acquire the logical path of each fact node in the ESG report according to the chapter structure tree, and to predict the logical path of each report comment in the comment information according to the chapter structure tree. The report comment alignment module 203 is used to perform optimal transmission analysis using a probability distribution composed of multiple report comments as the source distribution and a probability distribution composed of multiple fact nodes as the target distribution to obtain report comment mapping information. The transmission cost between the fact node and the report comment is determined based on their semantic distance and logical distance. The semantic distance is used to indicate the degree of difference between the semantics of the corresponding fact node and the semantics of the corresponding report comment, and the logical distance is used to indicate the degree of deviation between the logical path of the corresponding fact node and the logical path of the corresponding report comment.

[0081] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the optimal transmission system based on hybrid constraints of tree structure logic and semantics provided in the above embodiments and the optimal transmission method based on hybrid constraints of tree structure logic and semantics belong to the same concept. The specific implementation process is detailed in the method embodiments and will not be repeated here.

[0082] This invention also provides an electronic device. Please refer to [link to relevant documentation]. Figure 3 The electronic device may include a processor 301, a memory 302, and a program 3021 stored in the memory 302 and capable of running on the processor 301.

[0083] When program 3021 is executed by processor 301, it can achieve the following: Figure 1 Any steps in the corresponding method embodiments and the achievement of the same beneficial effects will not be repeated here.

[0084] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by hardware related to program instructions, and the program can be stored in a readable medium.

[0085] This invention also provides a readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described functions. Figure 1 Any step in the corresponding method embodiment can achieve the same technical effect, and will not be repeated here to avoid repetition.

[0086] The computer-readable storage medium of this invention can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0087] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0088] The program code contained on the storage medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0089] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or terminal. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0090] This invention also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned steps to achieve the optimal transmission method based on a hybrid constraint of structure tree logic and semantics provided in the above embodiments.

[0091] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0092] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. An optimal transmission method based on a hybrid constraint of structure tree logic and semantics, characterized in that, The method includes: The ESG report is parsed logically to obtain a chapter structure tree, wherein the chapter structure tree is stored in a tree structure and represents the hierarchical relationship of the content of the ESG report. The chapter structure tree includes multiple logical layers, and each logical layer includes multiple logical nodes. The logical path of each fact node in the ESG report is obtained based on the chapter structure tree, and the logical path of each report comment in the comment information is predicted based on the chapter structure tree. The fact node is the smallest verifiable, quantifiable, and associative unit of disclosed facts extracted from the ESG report. The logical path is used to indicate the logical nodes associated with the corresponding fact node or report comment at each corresponding logical layer. The comment information is information obtained by searching content on various social media platforms using the company corresponding to the ESG report as the search keyword. The report comment is the comment text associated with the company corresponding to the ESG report. The logical path of the report comment is predicted based on a large language model. The chapter structure tree is the prior background knowledge of the large language model. Using the probability distribution formed by multiple report comments as the source distribution and the probability distribution formed by multiple fact nodes as the target distribution, optimal transmission analysis is performed to obtain report comment mapping information. The transmission cost between the fact node and the report comment is determined based on their semantic distance and logical distance. The semantic distance indicates the degree of difference between the semantics of the corresponding fact node and the corresponding report comment. The logical distance indicates the degree of deviation between the logical path of the corresponding fact node and the logical path of the corresponding report comment. The logical distance is obtained by weighting multiple layer distances shared by the corresponding fact node and the corresponding report comment. The layer distance indicates the difference between the logical nodes associated with the corresponding fact node and the corresponding report comment at the corresponding logical layer.

2. The optimal transmission method based on hybrid constraints of structure tree logic and semantics as described in claim 1, characterized in that, The calculation weight of the layer distance is negatively correlated with the layer depth of the corresponding logical layer.

3. The optimal transmission method based on hybrid constraints of structure tree logic and semantics as described in claim 2, characterized in that, The rate of change of the number of the multiple logical layers and the calculation weight of the layer distance are negatively correlated.

4. The optimal transmission method based on hybrid constraints of structure tree logic and semantics as described in claim 1, characterized in that, After obtaining the report comment mapping information, the method further includes: In the multiple mapping groups included in the report comment mapping information, the transmission cost of each mapping group is obtained, wherein one of the mapping groups includes one fact node and one report comment; The transmission costs of multiple mapping groups are weighted to obtain a report-comment deviation value, wherein the weight of the transmission cost of the mapping group is the matching probability between the fact nodes included in the mapping group and the report-comment. The report-comment deviation value is used to characterize the degree of content difference between the ESG report and the comment information.

5. The optimal transmission method based on hybrid constraints of structure tree logic and semantics as described in claim 1, characterized in that, After obtaining the report comment mapping information, the method further includes: In the multiple mapping groups included in the report comment mapping information, the transmission cost of each mapping group is obtained, wherein one of the mapping groups includes one fact node and one report comment; The transmission cost of different mapping groups associated with the same fact node is analyzed to determine the node comment deviation value of each fact node, wherein the node comment deviation value is used to characterize the degree of content difference between the corresponding fact node and the comment information.

6. The optimal transmission method based on hybrid constraints of structure tree logic and semantics as described in claim 5, characterized in that, After analyzing the transmission costs of different mapping groups associated with the same fact nodes to determine the node comment deviation value for each fact node, the method further includes: Fact nodes whose comment deviation values ​​exceed the node comment deviation threshold are highlighted.

7. The optimal transmission method based on hybrid constraints of structure tree logic and semantics as described in claim 1, characterized in that, The method further includes: In the multiple report comments, the transmission cost of each report comment and each event node is analyzed to obtain the multiple event transmission costs for each report comment; Based on the multiple event transmission costs of each report comment, isolated comments are identified among the multiple report comments, wherein the multiple event transmission costs of the isolated comment are all greater than or equal to a cost threshold, and the isolated comment is prohibited from participating in the optimal transmission analysis.

8. The optimal transmission method based on hybrid constraints of structure tree logic and semantics as described in claim 7, characterized in that, After identifying isolated comments among the multiple reported comments based on the multiple event transmission costs for each reported comment, the method further includes: The probability distribution of isolated comments among multiple report comments is statistically analyzed to obtain the report missing index.

Images