Double carbon strategy recommendation system based on knowledge graph

Abstract

The application discloses a double-carbon strategy recommendation system based on a knowledge graph, and particularly relates to the field of double-carbon strategy recommendation, and is used for solving the bias risk problem caused by redundant aggregation and diffusion imbalance in a complex data scene. The bias risk problem is solved by organizing multi-source data into structured fingerprint clusters based on feature hashing and semantic grouping, so as to facilitate subsequent analysis and decision-making. By calculating a consistency aggregation index and a cross-path entropy index, the data internal redundancy and source complexity are revealed, and a score reflecting the bias risk is generated. In the reasoning stage, the bias score is used for weight adjustment, and a multi-path recommendation scheme with similar distribution is used to improve the diversity and scientificity of the recommendation result. Finally, the bias score is dynamically updated through cross-validation of new data and historical trends, so that the accuracy and adaptability of the knowledge graph are improved.

Description

Technical Field

[0001] This invention relates to the field of dual-carbon strategy recommendation, and more specifically, to a dual-carbon strategy recommendation system based on knowledge graphs. Background Technology

[0002] In dual-carbon strategy recommendation systems, a large amount of data from various sources, such as government agencies, industry associations, and enterprises, is often regarded as multi-party endorsement, thereby increasing trust in specific emission reduction paths or technological solutions. However, these data sources may share the same basic data or cite similar conclusions in terms of statistical caliber or information models, leading to highly repetitive and redundant aggregations during dynamic updates of the knowledge graph, and unintentionally amplifying their importance during the inference stage. When redundant aggregations are further solidified in the recommendation logic, it can create a carbon lock-in illusion: the system regards technological routes repeatedly reinforced by information from the same source as the optimal or only feasible solution, making it difficult to identify and absorb subsequent innovative emission reduction pathways. This results in rigid decision-making conclusions and delayed updates, thus becoming out of touch with actual policy needs and the pace of technological iteration. This problem seriously affects the accuracy and foresight of dual-carbon strategy recommendations based on knowledge graphs. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of existing technologies, embodiments of the present invention provide a dual-carbon strategy recommendation system based on knowledge graphs. This system organizes multi-source data into structured fingerprint clusters based on feature hashing and semantic grouping for subsequent analysis and decision-making. By calculating the consistency aggregation index and the cross-path entropy index, it reveals internal data redundancy and source complexity, generating a score reflecting bias risk. During the inference phase, the bias score is used for weight adjustment, and a multi-path recommendation scheme with similar distributions is employed to improve the diversity and scientific rigor of the recommendation results. Finally, the bias score is dynamically updated through cross-validation of new data and historical trends, enhancing the accuracy and adaptability of the knowledge graph, thereby addressing the problems mentioned in the background section.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] The knowledge graph-based dual-carbon strategy recommendation system includes: a fingerprint generation module, a bias assessment module, a weight adjustment module, and a dynamic verification module.

[0006] Fingerprint generation module: After receiving multi-source carbon emission reduction data, it applies the feature hashing method to generate a unique mapping identifier, and groups highly similar data into fingerprint clusters through semantic word segmentation and similarity calculation, and then passes the fingerprint clusters to the bias assessment module.

[0007] Bias assessment module: For each fingerprint cluster, the internal redundancy is calculated based on the consistency of feature distribution, and the complexity of information diffusion is assessed by analyzing the cross-relationship between data sources. The two results are combined to form a bias score, and the bias score of each cluster is passed to the weight adjustment module.

[0008] Weight adjustment module: During the strategy reasoning process, the bias score is called to perform weight reduction processing on fingerprint clusters with high bias scores, adjust their weight contribution in the reasoning results, and generate a multi-path recommendation scheme when the distribution is similar, and pass the weighted fingerprint clusters to the dynamic verification module.

[0009] Dynamic verification module: Introduces new data and historical trends into the downweighted fingerprint clusters for cross-validation, analyzes whether its redundancy characteristics have weakened or whether the diffusion pattern has changed, and dynamically adjusts the bias score based on the verification results.

[0010] In a preferred embodiment, the fingerprint generation module includes the following:

[0011] First, we receive datasets from multiple sources and perform unified feature extraction and unique identifier generation on these datasets; the specific processing flow is as follows:

[0012] Feature extraction and vectorization: For each carbon emission reduction data record, the TF-IDF vectorization method is used to convert the text content into a high-dimensional feature vector;

[0013] Feature hashing generates unique mapping identifiers: The SimHash algorithm is used to hash each TF-IDF vector to generate a fixed-length binary hash value;

[0014] Semantic segmentation and embedding vector generation: For each data record, semantic segmentation is performed using the BERT model, and context-aware embedding vectors are generated;

[0015] By combining the hash value generated by SimHash with the embedding vector generated by BERT, the similarity between data records is calculated using weighted cosine similarity. Data records with similarity scores exceeding the similarity threshold are grouped into the same fingerprint cluster.

[0016] In a preferred embodiment, the bias assessment module includes the following:

[0017] The consistency aggregation index is used to assess the degree of consistency of data features within a fingerprint cluster;

[0018] First, for each fingerprint cluster, extract its high-dimensional feature vector. ,in Indicates the first The set of values ​​for each feature is defined; then, for each feature, its local distribution similarity is calculated, as defined below: ;in, Indicates the number of data records in a fingerprint cluster; and These are features In the Article and No. The value in each data record; It is a feature The distribution scaling factor is defined as follows: ;in It is a feature The range of values ​​for is then determined; subsequently, the global consistency aggregation index is calculated. for: .

[0019] In a preferred embodiment, the bias assessment module further includes the following:

[0020] The cross-path entropy index is used to evaluate the complexity of cross-paths between data sources in a fingerprint cluster;

[0021] First, construct an information diffusion graph for all data sources in the fingerprint cluster. ,in For the data source set, To establish the relationships between sources, calculate the path distribution probability for each node. : Then, calculate the cross-path entropy exponent. : ;in, It is a node The degree of departure; yes and The number of common paths; yes and The total number of paths; using the dynamic entropy distribution method, the nonlinear coupling of the global consistency aggregation index and the cross-path entropy index is used to obtain a comprehensive bias score. .

[0022] In a preferred embodiment, the weight adjustment module includes the following:

[0023] In the strategy recommendation inference, the bias score of the fingerprint cluster is invoked; for bias scores higher than the weighting threshold, the bias score is reduced. The fingerprint clusters are subjected to a weighting reduction operation by constructing dynamic influence factors. Adjust its influence weight in the reasoning process; the dynamic influence factor formula is as follows: ;in, : Bias score of fingerprint clusters; The weighting threshold is dynamically set based on the strategy scenario. : Adjustment factor, which controls the steepness of the weighting curve.

[0024] In a preferred embodiment, the weight adjustment module further includes the following:

[0025] After adjusting the weights, the distribution characteristics of all candidate paths are analyzed. Similarity detection is performed between the reduced-weight fingerprint cluster paths and other paths to determine if there are candidate schemes with similar distributions. The similarity calculation formula is as follows: ;in, and Two candidate paths; and :path and In features The distribution value is calculated; if the similarity is higher than the classification threshold, a multi-path recommendation scheme containing all candidate paths with similar distributions is generated and output, with each scheme accompanied by a path source identifier and a description of the bias distribution characteristics.

[0026] In a preferred embodiment, the dynamic verification module includes the following:

[0027] For each fingerprint cluster that has been downweighted, new datasets are dynamically imported from external data sources, and feature sets of the new data are generated using feature hashing and semantic word segmentation methods. Subsequently, the feature set of existing data in the fingerprint cluster was compared. Perform cross-comparison; define the similarity of newly added data. The calculation formula is: ;in, and : These represent newly added data and the current fingerprint cluster in terms of features, respectively. Distribution values ​​on; Number of features;

[0028] Time series modeling is performed on historical data records of fingerprint clusters to generate feature variation curves. ,in Indicates a time point. For characteristics in time The set of distribution values; by calculating the historical trend index of the time series. The formula is as follows: ;in, The magnitude of change of the eigenvalue at consecutive points in time; Number of historical time points.

[0029] In a preferred embodiment, the dynamic verification module further includes the following:

[0030] when and Add new data similarity and historical trend index Combined application to dynamically adjust the bias score of fingerprint clusters The formula is adjusted as follows: ;in, Current bias score; Adjusted bias score; among which This indicates the addition of a similarity threshold. This indicates the threshold for trend stabilization.

[0031] The technical effects and advantages of the knowledge graph-based dual-carbon strategy recommendation system of this invention are as follows:

[0032] This invention precisely addresses data bias in dual-carbon strategy recommendation by employing a step-by-step feature analysis and dynamic adjustment method, achieving a complete process of data grouping, score generation, weight adjustment, and dynamic verification. Based on feature hashing and semantic grouping, multi-source data is organized into structured fingerprint clusters for subsequent analysis and decision-making. By calculating the consistency aggregation index and cross-path entropy index, internal redundancy and source complexity of the data are revealed, generating scores reflecting bias risk. In the inference stage, bias scores are used for weight adjustment, and a multi-path recommendation scheme with similar distributions is employed to improve the diversity and scientific rigor of the recommendation results. Finally, through cross-validation with new data and historical trends, the bias scores are dynamically updated, enhancing the accuracy and adaptability of the knowledge graph. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the dual-carbon strategy recommendation system based on knowledge graphs according to the present invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Example 1: Figure 1 The present invention provides a dual-carbon strategy recommendation system based on knowledge graphs, comprising: a fingerprint generation module, a bias assessment module, a weight adjustment module, and a dynamic verification module;

[0036] Fingerprint generation module: After receiving multi-source carbon emission reduction data, it applies the feature hashing method to generate a unique mapping identifier, and groups highly similar data into fingerprint clusters through semantic word segmentation and similarity calculation, and then passes the fingerprint clusters to the bias assessment module.

[0037] Bias assessment module: For each fingerprint cluster, the internal redundancy is calculated based on the consistency of feature distribution, and the complexity of information diffusion is assessed by analyzing the cross-relationship between data sources. The two results are combined to form a bias score, and the bias score of each cluster is passed to the weight adjustment module.

[0038] Weight adjustment module: During the strategy reasoning process, the bias score is called to perform weight reduction processing on fingerprint clusters with high bias scores, adjust their weight contribution in the reasoning results, and generate a multi-path recommendation scheme when the distribution is similar, and pass the weighted fingerprint clusters to the dynamic verification module.

[0039] Dynamic verification module: Introduces new data and historical trends into the downweighted fingerprint clusters for cross-validation, analyzes whether its redundancy characteristics have weakened or whether the diffusion pattern has changed, and dynamically adjusts the bias score based on the verification results.

[0040] The fingerprint generation module includes the following:

[0041] The dual-carbon strategy recommendation system first receives datasets from multiple sources, including corporate carbon emission reports, industry research literature, and technical white papers. To ensure efficient data processing and accurate aggregation, these datasets require unified feature extraction and unique identifier generation. The specific processing flow is as follows:

[0042] Feature Extraction and Vectorization: For each carbon emission reduction data record, the TF-IDF vectorization method is used to transform the text content into a high-dimensional feature vector. This method captures key features of the data by calculating word frequency and inverse document frequency, highlighting words that appear frequently in specific documents but are relatively rare in the entire dataset.

[0043] Feature hashing generates unique mapping identifiers: The SimHash algorithm is used to hash each TF-IDF vector, generating a fixed-length binary hash value. SimHash uses locality-sensitive hashing to map high-dimensional vectors to a low-dimensional space, keeping the hash values ​​of similar data close in Hamming distance, thus achieving fast similarity detection.

[0044] Semantic segmentation and embedding vector generation: For each data record, the BERT model is used for semantic segmentation, generating context-aware embedding vectors. The BERT model, based on the Transformer architecture, captures deep semantic relationships between words in sentences through bidirectional training. Its core advantage lies in its ability to understand the meaning of words in different contexts, thereby generating more semantically relevant text embeddings. This process includes the following steps:

[0045] Input processing: The text data is input into the BERT model, which first performs word segmentation, splitting the text into sub-word units.

[0046] Bidirectional encoding: BERT uses a multi-layer Transformer encoder to process contextual information bidirectionally (from left to right and from right to left) and capture the relationship between words in a sentence.

[0047] Embedding generation: Each word generates a high-dimensional embedding vector through BERT, which not only contains the surface meaning of the word, but also incorporates its semantic role in a specific sentence.

[0048] Context aggregation: Aggregate the embedding vectors of all words to generate a comprehensive embedding vector for the entire document, which is used for subsequent similarity calculation.

[0049] By combining the hash value generated by SimHash and the embedding vector generated by BERT, a weighted cosine similarity is used to calculate the similarity between data records. Specifically, the similarity scoring formula is as follows: ;in, and These are the embedding vectors of two data records, For the first The weights of each feature reflect its importance in the TF-IDF vector. The total number of features. Data records with similarity scores exceeding the similarity threshold are grouped into the same fingerprint cluster.

[0050] Each fingerprint cluster is labeled with its data source (e.g., data source type, publishing organization), time tag (e.g., data generation date, update cycle), and upstream / downstream relationships (e.g., reference relationships, dependencies). This structured information is then stored in a knowledge graph using a graph database (e.g., Neo4j) to ensure data traceability and efficient querying.

[0051] Through feature hashing and semantic grouping processing in the fingerprint generation module, the system can efficiently identify and aggregate carbon reduction data from different sources that are highly similar in content, forming structured fingerprint clusters. This process not only eliminates data redundancy but also lays a solid foundation for redundancy assessment and bias risk judgment in subsequent modules, ensuring that the knowledge graph accurately reflects the true distribution and diversity of data during dynamic updates.

[0052] After the fingerprint generation module is completed, a series of structured and highly similar fingerprint clusters have been generated in the knowledge graph, each labeled with clear source and time information. These fingerprint clusters will serve as the basis for subsequent modules to evaluate redundancy and bias risk, ensuring that redundant aggregation can be effectively identified and avoided when making dual-carbon strategy recommendations, thereby improving the diversity and accuracy of recommendation results.

[0053] In the dual-carbon strategy recommendation process, data within a fingerprint cluster may obscure its true distribution characteristics due to highly similar content or complex cross-references between sources, increasing the risk of bias in recommendation inference. The consistency aggregation index focuses on quantifying the distributional consistency of data features within a fingerprint cluster, revealing potential redundancy bias risks through sensitive capture of local distributions. Meanwhile, the cross-path entropy index assesses the complexity of information diffusion between data sources, determining whether diffusion patterns are singular or repetitive by analyzing the interaction path characteristics of sources, ensuring that the contribution of source diversity to eliminating bias is not overlooked. These two indices complement each other: the global consistency aggregation index reveals inherent consistency from the perspective of data content feature distribution, while the cross-path entropy index mines diffusion complexity from the cross-propagation paths of data sources. The combination of these two indices comprehensively quantifies the bias risk of fingerprint clusters, providing a solid quantitative foundation for data weighting and optimization in subsequent recommendation inference, and avoiding carbon lock-in caused by redundant aggregation or diffusion imbalance.

[0054] The bias assessment module includes the following:

[0055] Consistency aggregation indices are used to assess the degree of consistency of data features within a fingerprint cluster. Excessively high consistency suggests that the data may have originated from repeated citations or content rewriting from the same source, thus posing a high risk of redundancy bias. Traditional mean and variance indices are insufficient to capture subtle changes in the distribution of multidimensional features; therefore, complex and reasonable indices are needed to reflect this phenomenon.

[0056] First, for each fingerprint cluster, extract its high-dimensional feature vector. ,in Indicates the first The set of values ​​for each feature is given. Then, the local distribution similarity of each feature is calculated, defined as follows: ;in, Indicates the number of data records in a fingerprint cluster; and These are features In the Article and No. The value in each data record; It is a feature The distribution scaling factor is defined as follows: ;in It is a feature The range of values ​​for is then determined; subsequently, the global consistency aggregation index is calculated. for: .

[0057] The Global Consistency Aggregation Index amplifies the impact of subtle differences in the feature distribution by applying nonlinear decay and squaring to the similarity of local distributions. This avoids the problem that simple linear weighting may ignore subtle changes, making the index more sensitive to capturing the risk of redundant aggregation.

[0058] Cross-path entropy is used to assess the complexity of cross-paths between data sources in a fingerprint cluster. Diversity of data sources can effectively reduce the risk of bias, but the uniformity or repetition of paths can lead to potential diffusion bias. Traditional path counting methods cannot quantify the complex interactions between sources; therefore, entropy theory is needed to introduce nonlinear complexity assessment.

[0059] First, construct an information diffusion graph for all data sources in the fingerprint cluster. ,in For the data source set, This involves identifying relationships between sources (such as co-references, mutual dependencies, etc.). Calculating the path distribution probability for each node. : Then, calculate the cross-path entropy exponent. : ;in, It is a node The degree of departure; yes and The number of common paths; yes and The total number of paths.

[0060] The cross-path entropy index, by combining entropy value and path interaction complexity, can effectively evaluate the diffusion pattern between sources, while taking into account both global entropy distribution and local interaction characteristics.

[0061] By employing the dynamic entropy distribution method, a comprehensive bias score is obtained through the nonlinear coupling of the global consistency aggregation index and the cross-path entropy index. The dynamic entropy distribution method dynamically adjusts the contribution of each index to the bias score, thus more comprehensively reflecting the bias risk characteristics of fingerprint clusters. Specifically, the global consistency aggregation index reflects the consistency of data features; a higher value indicates that the data may come from a single or repeated source. The cross-path entropy index measures the diffusion complexity between data sources; a lower value indicates lower diversity of source interaction paths. The dynamic entropy distribution method couples the impact of increasing the global consistency aggregation index on the score with the complexity change of the cross-path entropy index through an exponential nonlinear decay mechanism. This ensures that when the cross-path entropy index is small, the amplifying effect of the global consistency aggregation index on the score is significant, and vice versa. This nonlinear fusion uses a dynamic adjustment term in the denominator to control the growth rate of the score, ensuring that the assessment of bias risk takes into account both the intrinsic characteristics and external diffusion features of the fingerprint cluster, achieving accurate quantification and differentiated processing of different types of data, and providing a more reliable basis for subsequent recommendation logic. The calculation method is defined as follows: ;in, Indicates the consistency aggregation index; The cross-path entropy exponent is represented by the denominator. Used for dynamic adjustment of overall risk.

[0062] The comprehensive bias score quantifies the overall bias risk of a fingerprint cluster across two dimensions: feature distribution consistency and data source diffusion complexity. A higher value indicates a higher degree of internal feature consistency (i.e., stronger redundancy) and a more homogeneous diffusion pattern between sources, suggesting a greater risk of bias. Conversely, a lower value indicates a more diverse data distribution and more complex interactions between sources, resulting in a lower risk of bias. By using the comprehensive bias score, high-risk fingerprint clusters that may lead to biased inferences can be effectively identified, providing a clear quantitative basis for subsequent weight reduction processing and multi-path strategy generation, thereby improving the scientific rigor and robustness of the recommendation logic.

[0063] By calculating the consistency aggregation index and the cross-path entropy index, and fusing them using a dynamic entropy distribution method, the bias assessment module achieves comprehensive quantification and accurate assessment of fingerprint cluster bias risk. This process not only reveals the redundancy of feature distribution within data records and the complexity of diffusion between sources, but also dynamically adjusts the relative influence of the two through a nonlinear fusion method, ensuring that the score reflects both the inherent consistency of the data content and captures the global characteristics of source diffusion patterns. This dual-dimensional quantification method overcomes the limitations of traditional bias assessments that rely solely on feature consistency or simple path counting, providing a scientific basis for multi-dimensional bias risk control in dynamic and complex data scenarios. The structured storage of the bias score further supports the weighting processing and multi-path recommendation logic of subsequent modules, enabling policy reasoning to avoid carbon lock-in caused by redundant aggregation or diffusion imbalance with higher robustness and accuracy.

[0064] After completing the bias assessment module, each fingerprint cluster has obtained a comprehensive bias score based on the consistency aggregation index and the cross-path entropy index, and this score is stored in a structured manner. The goal of the weight adjustment module is to reduce the weight of fingerprint clusters based on the bias score, while simultaneously generating multi-path recommendation schemes during inference to ensure the accuracy and diversity of strategy recommendations. The specific processing logic is as follows:

[0065] The weight adjustment module includes the following:

[0066] The bias score of the fingerprint cluster is used in the strategy recommendation inference. Bias scores exceeding a weighting threshold are considered. The fingerprint clusters are subjected to a weighting reduction operation by constructing dynamic influence factors. Adjust its influence weight in the reasoning process. The formula for the dynamic influence factor is as follows: ;in, : Bias score of fingerprint clusters; The weighting threshold is dynamically set based on the strategy scenario. : Adjustment factor, which controls the steepness of the weighting curve.

[0067] The dynamic impact factor ranges from (0,1), with smaller values ​​indicating greater weighting. The contribution weight of each fingerprint cluster is updated based on the impact factor during the inference process, ensuring that the interference of highly biased data on the inference results is effectively mitigated.

[0068] After adjusting the weights, the distribution characteristics of all candidate paths are further analyzed. Similarity detection is performed between the reduced-weight fingerprint cluster paths and other paths to determine if there are candidate schemes with similar distributions. The similarity calculation formula is as follows: ;in, and Two candidate paths; and :path and In features The distribution values ​​on.

[0069] If the similarity is higher than the classification threshold, a multi-path recommendation scheme containing all candidate paths with similar distributions is generated and output. Each scheme is accompanied by a path source identifier and a description of the bias distribution characteristics, which facilitates users to conduct visual evaluation and subsequent decision-making.

[0070] The reduced weight values ​​and generated multi-path recommendation schemes are stored in a structured manner. An index is built using the unique identifier of the fingerprint cluster (generated by the fingerprint generation module) to ensure that subsequent modules can directly access the adjusted weights and recommendation schemes. Simultaneously, an output interface is designed for the recommendation schemes, including key information such as path content, source identifier, bias score, and reduced weight impact factor, ensuring the traceability and transparency of the recommendation results.

[0071] The weight adjustment module takes over the bias score generated by the bias assessment module. During the inference process, it uses dynamic influence factors to reduce the weight of highly biased data, ensuring that the weight adjustment dynamically adapts to the bias characteristics of the data. Simultaneously, it generates multi-path recommendation schemes through distribution similarity detection, ensuring the diversity and rationality of strategy recommendations and providing accurate input for subsequent dynamic verification and score updates.

[0072] Through the weighting reduction processing of the weight adjustment module and the multi-path generation logic, the contribution weight of each fingerprint cluster in the inference process is dynamically adjusted, effectively mitigating the impact of highly biased data on policy recommendation. The generation of multi-path recommendation schemes provides more diverse choices for policy decisions in complex scenarios, and by adding bias distribution descriptions and path source identifiers, the transparency and interpretability of the recommendation results are improved. These processing logics ensure the sensitivity of the recommendation process to bias features and support for policy diversity.

[0073] After the weight adjustment module is completed, each fingerprint cluster has been downweighted based on its bias score, and a multi-path recommendation scheme has been generated. The dynamic validation module aims to cross-validate the downweighted fingerprint clusters by introducing new data and analyzing historical trends, dynamically adjusting their bias scores, and synchronously updating the node relationships and priority rankings in the knowledge graph to ensure the continued accuracy and forward-looking nature of the recommendation results. The specific processing logic is as follows:

[0074] The dynamic verification module includes the following:

[0075] For each fingerprint cluster that has been downweighted, new datasets are dynamically imported from external data sources, and feature sets of the new data are generated using feature hashing and semantic word segmentation methods. Subsequently, the feature set of existing data in the fingerprint cluster was compared. Perform cross-comparison. Define the similarity of newly added data. The calculation formula is: ;in, and : These represent newly added data and the current fingerprint cluster in terms of features, respectively. Distribution values ​​on; Number of features.

[0076] Add data similarity The value ranges from [0,1], with higher values ​​indicating a high degree of similarity between the new data and the existing data in terms of feature distribution.

[0077] Time series modeling is performed on historical data records of fingerprint clusters to generate feature variation curves. ,in Indicates a time point. For characteristics in time The set of distribution values. By calculating the historical trend index of the time series. To determine whether the feature distribution has changed significantly, the formula is as follows: ;in, The magnitude of change of the eigenvalue at consecutive points in time; Number of historical time points.

[0078] A higher historical trend index value indicates a more significant change in the characteristic distribution, while a lower value indicates greater stability.

[0079] when and Add new data similarity and historical trend index Combined application to dynamically adjust the bias score of fingerprint clusters The formula is adjusted as follows: ;in, Current bias score; Adjusted bias score; among which This indicates the addition of a similarity threshold. This indicates the threshold for trend stabilization.

[0080] This formula combines the similarity of new data with the changing characteristics of historical distribution. When new data is highly similar to existing data and the historical trend is relatively stable, the bias score will be significantly reduced, reflecting the improvement in data credibility.

[0081] The adjusted bias score updates the fingerprint cluster node attributes in the knowledge graph in real time. By introducing new association rules, the updated knowledge graph is ensured to more accurately reflect the dynamic characteristics of the data.

[0082] The dynamic verification module takes over the weighted data and bias scores generated in the weight adjustment module, and uses new data and historical trends to cross-validate and dynamically adjust the bias characteristics of fingerprint clusters. This process, by comprehensively considering the similarity of new data and historical distribution changes, ensures that the bias scores of fingerprint clusters can reflect the latest characteristics of the data in a timely manner, providing key support for the continuous optimization of the knowledge graph.

[0083] After completing the dynamic verification module, the bias score of the reduced-weight fingerprint cluster has been dynamically adjusted through cross-validation with new data and historical trends. The node attributes and relationships in the knowledge graph are updated in real time, reflecting the latest characteristics of data distribution and changes in credibility. This process ensures the robustness and adaptability of the recommendation logic in dynamic data scenarios, providing robust support for continuously improving the accuracy and scientific rigor of the dual-carbon strategy recommendation.

[0084] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0085] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0086] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "include" or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the inclusion of an element by a statement that defines "..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0087] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A knowledge graph-based dual-carbon strategy recommendation system, characterized in that, include: The module includes a fingerprint generation module, a bias assessment module, a weight adjustment module, and a dynamic verification module. Fingerprint generation module: After receiving multi-source carbon emission reduction data, it applies the feature hashing method to generate a unique mapping identifier, and groups highly similar data into fingerprint clusters through semantic word segmentation and similarity calculation, and then passes the fingerprint clusters to the bias assessment module. Bias assessment module: For each fingerprint cluster, the internal redundancy is calculated based on the consistency of feature distribution, and the complexity of information diffusion is assessed by analyzing the cross-relationship between data sources. The two results are combined to form a bias score, and the bias score of each cluster is passed to the weight adjustment module. The bias assessment module includes the following: The consistency aggregation index is used to assess the degree of consistency of data features within a fingerprint cluster; First, for each fingerprint cluster, extract its high-dimensional feature vector. ,in Indicates the first The set of values ​​for each feature is defined; then, for each feature, its local distribution similarity is calculated, as defined below: ;in, Indicates the number of data records in a fingerprint cluster; and These are features In the Article and No. The value in each data record; It is a feature The distribution scaling factor is defined as follows: ;in It is a feature The range of values ​​for is then determined; subsequently, the global consistency aggregation index is calculated. for: ; The cross-path entropy index is used to evaluate the complexity of cross-paths between data sources in a fingerprint cluster; First, construct an information diffusion graph for all data sources in the fingerprint cluster. ,in For the data source set, To establish the relationships between sources, calculate the path distribution probability for each node. : Then, calculate the cross-path entropy exponent. : ;in, It is a node The degree of departure; yes and The number of common paths; yes and The total number of paths; using the dynamic entropy distribution method, the nonlinear coupling of the global consistency aggregation index and the cross-path entropy index is used to obtain a comprehensive bias score. ; Weight adjustment module: During the strategy reasoning process, the bias score is called to perform weight reduction processing on fingerprint clusters with high bias scores, adjust their weight contribution in the reasoning results, and generate a multi-path recommendation scheme when the distribution is similar, and pass the weighted fingerprint clusters to the dynamic verification module. Dynamic verification module: Introduces new data and historical trends into the downweighted fingerprint clusters for cross-validation, analyzes whether its redundancy characteristics have weakened or whether the diffusion pattern has changed, and dynamically adjusts the bias score based on the verification results.

2. The knowledge graph-based dual-carbon strategy recommendation system according to claim 1, characterized in that: The fingerprint generation module includes the following: First, we receive datasets from multiple sources and perform unified feature extraction and unique identifier generation on these datasets; the specific processing flow is as follows: Feature extraction and vectorization: For each carbon emission reduction data record, the TF-IDF vectorization method is used to convert the text content into a high-dimensional feature vector; Feature hashing generates unique mapping identifiers: The SimHash algorithm is used to hash each TF-IDF vector to generate a fixed-length binary hash value; Semantic segmentation and embedding vector generation: For each data record, semantic segmentation is performed using the BERT model, and context-aware embedding vectors are generated; By combining the hash value generated by SimHash with the embedding vector generated by BERT, the similarity between data records is calculated using weighted cosine similarity. Data records with similarity scores exceeding the similarity threshold are grouped into the same fingerprint cluster.

3. The knowledge graph-based dual-carbon strategy recommendation system according to claim 2, characterized in that: The weight adjustment module includes the following: In the strategy recommendation inference, the bias score of the fingerprint cluster is invoked; for bias scores higher than the weighting threshold, the bias score is reduced. The fingerprint clusters are subjected to a weighting reduction operation by constructing dynamic influence factors. Adjust its influence weight in the reasoning process; the dynamic influence factor formula is as follows: ;in, : Bias score of fingerprint clusters; The weighting threshold is dynamically set based on the strategy scenario. : Adjustment factor, which controls the steepness of the weighting curve.

4. The knowledge graph-based dual-carbon strategy recommendation system according to claim 3, characterized in that: The weight adjustment module also includes the following: After adjusting the weights, the distribution characteristics of all candidate paths are analyzed. Similarity detection is performed between the reduced-weight fingerprint cluster paths and other paths to determine if there are candidate schemes with similar distributions. The similarity calculation formula is as follows: Among them, and Two candidate paths; and :path and In features The distribution value is calculated; if the similarity is higher than the classification threshold, a multi-path recommendation scheme containing all candidate paths with similar distributions is generated and output, with each scheme accompanied by a path source identifier and a description of the bias distribution characteristics.

5. The knowledge graph-based dual-carbon strategy recommendation system according to claim 4, characterized in that: The dynamic verification module includes the following: For each fingerprint cluster that has been downweighted, new datasets are dynamically imported from external data sources, and feature sets of the new data are generated using feature hashing and semantic word segmentation methods. Subsequently, the feature set of existing data in the fingerprint cluster was compared. Perform cross-comparison; define the similarity of newly added data. The calculation formula is: ;in, and : These represent newly added data and the current fingerprint cluster in terms of features, respectively. Distribution values ​​on; Number of features; Time series modeling is performed on historical data records of fingerprint clusters to generate feature variation curves. ,in Indicates a time point. For characteristics in time The set of distribution values; by calculating the historical trend index of the time series. The formula is as follows: ;in, The magnitude of change of the eigenvalue at consecutive points in time; Number of historical time points.

6. The knowledge graph-based dual-carbon strategy recommendation system according to claim 5, characterized in that: The dynamic verification module also includes the following: when and Add new data similarity and historical trend index Combined application to dynamically adjust the bias score of fingerprint clusters The formula is adjusted as follows: ;in, Current bias score; Adjusted bias score; among which This indicates the addition of a similarity threshold. This indicates the threshold for trend stabilization.

Images