Knowledge credibility evaluation system and method

Abstract

The embodiment of the invention provides a knowledge credibility evaluation system and method. The system comprises a knowledge base unit, a knowledge enhancement unit, a data input unit, a knowledge routing unit and a credibility evaluation unit, the knowledge base unit is used for providing knowledge documents and knowledge maps of a power grid, and the knowledge enhancement unit is used for carrying out vector conversion and association processing on the knowledge documents and the knowledge maps. The data input unit is used for acquiring query data and transmitting the query data to the knowledge routing unit, the knowledge routing unit is used for determining at least one recall knowledge semantic vector corresponding to the query data, and the credibility evaluation unit is connected with the knowledge enhancement unit and the knowledge routing unit. And the credibility evaluation module is used for performing knowledge credibility evaluation on the at least one recalled knowledge semantic vector according to the knowledge association relationship, and determining a target credibility score corresponding to the at least one recalled knowledge semantic vector. According to the technical scheme, accurate evaluation of the credibility of the recall knowledge is realized.

Description

Technical Field

[0001] This application relates to the field of power grid technology, and in particular to a system and method for assessing knowledge credibility. Background Technology

[0002] In the power grid sector, the knowledge system encompasses a vast amount of technical standards and operational specifications that are frequently updated. Examples include State Grid standards, industry standards, and internal enterprise technical specifications. These documents typically contain numerous technical terms and complex logical relationships, and different versions may conflict or iteratively overlap, necessitating an assessment of the credibility of knowledge retrieval.

[0003] Existing methods mainly enhance the processing capabilities of large models by using external knowledge bases and employ simple dimensions to evaluate knowledge credibility, reflecting the quality of recalled knowledge.

[0004] However, existing methods suffer from poor accuracy in assessing knowledge credibility. Summary of the Invention

[0005] This application provides a knowledge credibility assessment system and method to improve the accuracy of knowledge credibility assessment.

[0006] In a first aspect, embodiments of this application provide a knowledge credibility assessment system, including: a knowledge base unit, a knowledge enhancement unit, a data input unit, a knowledge routing unit, and a credibility assessment unit;

[0007] The knowledge base unit is used to provide knowledge documents and knowledge graphs for the power grid;

[0008] The knowledge enhancement unit is connected to the knowledge base unit and is used to perform vector transformation and association processing on the knowledge document and the knowledge graph to obtain knowledge semantic vectors and knowledge association relationships.

[0009] The data input unit is connected to the knowledge routing unit and is used to obtain the query data input by the user and transmit the query data to the knowledge routing unit;

[0010] The knowledge routing unit is connected to the knowledge enhancement unit and is used to determine at least one recall knowledge semantic vector corresponding to the query data in the knowledge semantic vector.

[0011] The credibility assessment unit is connected to the knowledge enhancement unit and the knowledge routing unit, and is used to assess the credibility of the at least one recalled knowledge semantic vector according to the knowledge association relationship, and determine the target credibility score corresponding to the at least one recalled knowledge semantic vector.

[0012] In one or more embodiments, the credibility assessment unit includes: a multidimensional assessment unit, a weight adjustment unit, and a recall decay unit;

[0013] The multidimensional evaluation unit is used to determine the basic credibility score corresponding to the at least one recalled knowledge semantic vector based on the preset evaluation dimensions and preset hierarchical weights.

[0014] The weight adjustment unit and the multidimensional evaluation unit are connected and are used to update the basic credibility score corresponding to the at least one recalled knowledge semantic vector according to the knowledge association relationship and the preset adjustment factor to obtain the updated credibility score.

[0015] The recall attenuation unit and the weight adjustment unit are connected and used to correct the updated credibility score based on historical recall data, and determine the target credibility score corresponding to the at least one recall knowledge semantic vector.

[0016] In one or more embodiments, the multidimensional evaluation unit includes: a feature vector construction subunit, a weight management subunit, a comprehensive scoring subunit, and an uncertainty handling subunit;

[0017] The feature vector construction subunit is used to construct a multi-dimensional evaluation vector corresponding to the at least one recalled knowledge semantic vector based on the preset evaluation dimension.

[0018] The weight management subunit is used to obtain the preset hierarchical weights;

[0019] The comprehensive scoring subunit is connected to the feature vector construction subunit and the weight management subunit, and is used to perform weighted summation on at least one multidimensional evaluation vector according to the preset hierarchical weights to obtain the basic credibility score corresponding to the at least one recalled knowledge semantic vector.

[0020] The uncertainty processing subunit is connected to the comprehensive scoring subunit and is used to update the basic credibility score corresponding to the at least one recalled knowledge semantic vector according to the uncertainty penalty factor.

[0021] In one or more embodiments, the weight adjustment unit includes: a knowledge graph construction subunit, a similarity determination subunit, a weight update subunit, and a weight propagation subunit;

[0022] The knowledge graph construction subunit is used to construct a knowledge graph structure based on the knowledge associations. The knowledge associations include at least one knowledge source and the associations between different knowledge sources. The knowledge graph structure includes nodes corresponding to the at least one knowledge source and edges corresponding to the associations between the different knowledge sources.

[0023] The similarity determination subunit is connected to the knowledge graph construction subunit and is used to determine the first similarity value between the first knowledge source and at least one second knowledge source corresponding to the at least one recalled knowledge semantic vector according to the knowledge graph structure, wherein the at least one second knowledge source is associated with the first knowledge source;

[0024] The weight update subunit is connected to the similarity determination subunit and is used to update the basic credibility score corresponding to the at least one recalled knowledge semantic vector according to the first similarity value and the preset adjustment factor to obtain the updated credibility score.

[0025] The weight propagation subunit is connected to the weight update subunit and is used to perform weight adjustment processing on the at least one second knowledge source according to the updated credibility score.

[0026] In one or more embodiments, the recall attenuation unit includes: a time window subunit, a filter rate monitoring subunit, a threshold management subunit, an attenuation strategy subunit, and a recovery balance subunit;

[0027] The time window subunit is used to divide the updated confidence score into windows according to preset time window parameters to obtain at least one time window and corresponding window data.

[0028] The filtration rate monitoring subunit is connected to the time window subunit and is used to determine the filtration rate data corresponding to the at least one time window.

[0029] The threshold management subunit is connected to the filter rate monitoring subunit and is used to determine the adjustment threshold corresponding to the updated confidence score based on the historical recall data and the filter rate data corresponding to the at least one time window.

[0030] The attenuation strategy subunit is connected to the threshold management subunit and is used to correct the updated credibility score according to the adjustment threshold, and determine the target credibility score corresponding to the at least one recalled knowledge semantic vector.

[0031] The recovery balance subunit is used to perform recovery processing on the updated confidence score when the filter rate data corresponding to the at least one time window is less than a preset threshold.

[0032] In one or more embodiments, the system further includes: a source tracing unit and a visualization unit;

[0033] The source tracing unit is connected to the knowledge enhancement unit and the knowledge routing unit, and is used to determine the knowledge source tracing path corresponding to the at least one recalled knowledge semantic vector based on the knowledge association relationship and the at least one recalled knowledge semantic vector;

[0034] The visualization unit is connected to the source tracing unit and is used to visualize the knowledge source tracing path corresponding to the at least one recalled knowledge semantic vector.

[0035] In one or more embodiments, the system further includes: a text similarity unit, a data filtering unit, a recall verification unit, and a user feedback unit;

[0036] The text similarity unit is connected to the knowledge routing unit and the knowledge enhancement unit, and is used to determine the second similarity value between the at least one recalled knowledge semantic vector and the knowledge semantic vector.

[0037] The data filtering unit is connected to the text similarity unit and is used to filter the at least one recalled knowledge semantic vector according to the second similarity value and the preset text similarity threshold to obtain the filtered recalled knowledge semantic vector.

[0038] The recall verification unit is connected to the data filtering unit and is used to verify the filtered recall knowledge semantic vector and determine the verified recall knowledge semantic vector.

[0039] The user feedback unit is connected to the recall verification unit and is used to evaluate the verified recalled knowledge semantic vector to obtain knowledge confidence data.

[0040] The credibility assessment unit is connected to the user feedback unit and is used to adjust the credibility assessment based on the knowledge confidence data.

[0041] In one or more embodiments, the system further includes: a data storage unit and a data query unit;

[0042] The data storage unit is connected to the knowledge base unit and the knowledge enhancement unit, and is used to store the knowledge documents, the knowledge graph, the knowledge semantic vectors, and the knowledge associations.

[0043] The data query unit is connected to the knowledge base unit and the data input unit, and is used to query the query data according to the knowledge graph to determine the associated graph data corresponding to the query data.

[0044] Secondly, embodiments of this application provide a method for evaluating knowledge credibility, applied to a credibility evaluation unit in the knowledge credibility evaluation system described in the first aspect, the method comprising:

[0045] Obtain at least one semantic vector of recall knowledge corresponding to the query data input by the user;

[0046] Based on the preset evaluation dimensions and preset hierarchical weights, the basic credibility score corresponding to the at least one recalled knowledge semantic vector is determined;

[0047] The basic credibility score corresponding to the at least one recalled knowledge semantic vector is updated based on the knowledge association relationship and the preset adjustment factor to obtain the updated credibility score.

[0048] The updated credibility score is corrected based on historical recall data to determine the target credibility score corresponding to the at least one recalled knowledge semantic vector.

[0049] In one or more embodiments, the method further includes:

[0050] Based on the target credibility score corresponding to the at least one recalled knowledge semantic vector, the at least one recalled knowledge semantic vector is sorted to obtain the sorted recalled knowledge semantic vector.

[0051] Determine the third similarity value between at least two recalled knowledge semantic vectors in the sorted recalled knowledge semantic vectors;

[0052] The recall knowledge semantic vectors with a third similarity value greater than a preset threshold are fused together to obtain a fused recall knowledge semantic vector.

[0053] The fused recall knowledge semantic vector is filtered according to a preset recall strategy to determine the target semantic vector.

[0054] Thirdly, embodiments of this application provide a knowledge credibility assessment apparatus, which is applied to the credibility assessment unit in the aforementioned knowledge credibility assessment system. The apparatus includes:

[0055] The acquisition module is used to acquire at least one semantic vector of recall knowledge corresponding to the query data input by the user;

[0056] The determination module is used to determine the basic credibility score corresponding to the at least one recalled knowledge semantic vector based on the preset evaluation dimensions and preset hierarchical weights.

[0057] The first processing module is used to update the basic credibility score corresponding to the at least one recalled knowledge semantic vector according to the knowledge association relationship and the preset adjustment factor, so as to obtain the updated credibility score.

[0058] The second processing module is used to correct the updated credibility score based on historical recall data and determine the target credibility score corresponding to the at least one recalled knowledge semantic vector.

[0059] In one or more embodiments, the second processing module is further configured to:

[0060] Based on the target credibility score corresponding to the at least one recalled knowledge semantic vector, the at least one recalled knowledge semantic vector is sorted to obtain the sorted recalled knowledge semantic vector.

[0061] Determine the third similarity value between at least two recalled knowledge semantic vectors in the sorted recalled knowledge semantic vectors;

[0062] The recall knowledge semantic vectors with a third similarity value greater than a preset threshold are fused together to obtain a fused recall knowledge semantic vector.

[0063] The fused recall knowledge semantic vector is filtered according to a preset recall strategy to determine the target semantic vector.

[0064] This application provides a knowledge credibility assessment system and method. The system includes: a knowledge base unit, a knowledge enhancement unit, a data input unit, a knowledge routing unit, and a credibility assessment unit. The knowledge base unit provides knowledge documents and knowledge graphs related to the power grid. The knowledge enhancement unit, connected to the knowledge base unit, performs vector transformation and association processing on the knowledge documents and knowledge graphs to obtain knowledge semantic vectors and knowledge association relationships. The data input unit, connected to the knowledge routing unit, acquires user-input query data and transmits the query data to the knowledge routing unit. The knowledge routing unit, connected to the knowledge enhancement unit, determines at least one recalled knowledge semantic vector corresponding to the query data from the knowledge semantic vectors. The credibility assessment unit, connected to the knowledge enhancement unit and the knowledge routing unit, assesses the knowledge credibility of at least one recalled knowledge semantic vector based on the knowledge association relationships, and determines a target credibility score corresponding to at least one recalled knowledge semantic vector. The technical solution of this application can achieve accurate evaluation of the credibility of recalled knowledge based on the credibility assessment unit. Attached Figure Description

[0065] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0066] Figure 1 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 1 ;

[0067] Figure 2 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 2 ;

[0068] Figure 3 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 3 ;

[0069] Figure 4 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 4 ;

[0070] Figure 5 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 5 ;

[0071] Figure 6 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 6 ;

[0072] Figure 7 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 7 ;

[0073] Figure 8 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 8 ;

[0074] Figure 9 A flowchart illustrating the knowledge credibility assessment method provided in the embodiments of this application;

[0075] Figure 10 A schematic diagram of the structure of the knowledge credibility assessment device provided in the embodiments of this application.

[0076] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0077] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0078] Before introducing the embodiments of this application, the application background of the embodiments of this application will be explained first:

[0079] In the power grid sector, the knowledge system encompasses a vast amount of technical standards and operational specifications that are frequently updated. Examples include State Grid standards, industry standards, and internal enterprise technical specifications. These documents typically contain numerous technical terms and complex logical relationships, and different versions may conflict or iteratively overlap, necessitating an assessment of the credibility of knowledge retrieval.

[0080] Existing methods mainly enhance the processing capabilities of large models by using external knowledge bases and employ simple dimensions to evaluate knowledge credibility, reflecting the quality of recalled knowledge.

[0081] However, existing methods suffer from poor accuracy in assessing knowledge credibility.

[0082] This application provides a knowledge credibility assessment system aimed at solving the aforementioned technical problems of existing technologies. The technical concept of this application is as follows: Existing methods mainly enhance the processing capabilities of large models through external knowledge bases and use simple dimensions for knowledge credibility assessment, resulting in poor accuracy. To address the issue of single data dimensions, this application considers multi-dimensional analysis of the recalled data and assigns different levels of weight according to the importance of different dimensions. Then, through multi-dimensional weighted fusion, the credibility assessment score of the recalled data is determined. Therefore, this application sets up a credibility assessment unit, and within this unit, it includes a multi-dimensional assessment unit, a weight adjustment unit, and a recall decay unit. Based on multi-dimensional characteristics such as the authority, timeliness, and historical performance of the knowledge source, combined with topological graph structure modeling and time window analysis, the knowledge weights are dynamically adjusted to achieve accurate knowledge credibility assessment.

[0083] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0084] Figure 1 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 1 .like Figure 1 As shown, the knowledge credibility assessment system includes: a knowledge base unit, a knowledge enhancement unit, a data input unit, a knowledge routing unit, and a credibility assessment unit;

[0085] The following implementation instructions are provided for this knowledge credibility assessment system:

[0086] 1) Knowledge base unit, used to provide knowledge documents and knowledge graphs of the power grid;

[0087] For example, the knowledge base unit provides basic data support for the knowledge credibility assessment system, integrates knowledge resources in the power grid field, including a standard knowledge document library and a knowledge graph sub-library, and provides structured and callable knowledge documents and knowledge graphs for subsequent assessments.

[0088] In one possible implementation, the standard knowledge document repository is used to store knowledge documents such as industry standards and technical specifications of power grid companies in the power grid field. It mainly uses web crawling technology or text recognition technology to automatically collect and update documents from open source web pages to ensure the timeliness of the documents.

[0089] The knowledge graph sub-base provides a relational graph of specialized terms and concepts in the power grid field, as well as the relationships between different terms / concepts. In this graph, specialized terms / concepts serve as nodes, and the relationships between these terms / concepts are used as the connections between nodes. For example, substation equipment "belongs to" primary equipment, and relay protection "is applied to" transmission lines.

[0090] 2) The knowledge enhancement unit is connected to the knowledge base unit to perform vector transformation and association processing on knowledge documents and knowledge graphs, so as to obtain knowledge semantic vectors and knowledge association relationships;

[0091] For example, the knowledge enhancement unit is connected to the knowledge base unit to transform the original knowledge in the knowledge base (including knowledge documents and knowledge graphs) into machine-understandable knowledge semantic vectors and establish the relationship between knowledge.

[0092] In one possible implementation, the knowledge augmentation unit uses a pre-trained BERT-based language model to segment and encode the text of the knowledge document, transform the knowledge document into a high-dimensional document vector, and transform the professional concepts in the knowledge graph into concept vectors of the same dimension.

[0093] A "concept-document" association table is generated by calculating the semantic similarity between concept vectors and document vectors. The semantic similarity calculation uses the cosine similarity formula, as shown below:

[0094]

[0095] in, Represent a dimension( Concept vectors (with the same output dimension as BERT-base), Represents document vectors (and) (The dimensions are consistent) Represents the dot product of vectors. Representing vectors L2 norm, Representing vectors The L2 norm.

[0096] Furthermore, the knowledge enhancement unit obtains standard IDs and referenced document IDs through entity representation learning, constructs the relationship between every two nodes in the knowledge graph, and then performs reasoning on the concept vectors. Finally, a computational graph based on the knowledge enhancement model is constructed to encode and compute the input data, sorts the semantic similarity between each document vector and each concept vector, and outputs the association between the most similar document IDs and referenced document IDs.

[0097] 3) The data input unit is connected to the knowledge routing unit and is used to obtain the query data input by the user and transmit the query data to the knowledge routing unit;

[0098] For example, the data input unit is the interaction interface between the knowledge credibility assessment system and the user. After receiving the user's query request and completing the format verification and standardization process, it transmits the valid query data to the knowledge routing unit.

[0099] In one possible implementation, the data input unit provides multiple input interfaces, including natural language queries, keyword queries, and condition combination queries, which can adapt to the usage habits of different users.

[0100] For example, after a user enters "distribution network protection standard for XX year", the data input unit first verifies the input format, removes invalid characters, and then extracts the core elements "distribution network protection standard" and "XX year". After that, the standardized query data is transmitted to the knowledge routing unit in real time, and the query time and type are recorded for easy traceability later.

[0101] 4) The knowledge routing unit is connected to the knowledge enhancement unit to determine at least one recall knowledge semantic vector corresponding to the query data in the knowledge semantic vector;

[0102] For example, the knowledge routing unit filters at least one recall knowledge semantic vector related to the user's query data from the knowledge semantic vector library based on semantic matching.

[0103] In one possible implementation, the knowledge routing unit uses a pre-trained BERT language model to transform user query data into high-dimensional query vectors, ensuring that the query vectors have the same dimension as the knowledge semantic vectors. Then, it calculates the cosine similarity between the query vectors and the knowledge semantic vectors (including document semantic vectors and concept semantic vectors), sets a similarity threshold of 0.6, and filters out recall knowledge semantic vectors with similarity greater than or equal to the similarity threshold.

[0104] 5) The credibility assessment unit is connected to the knowledge enhancement unit and the knowledge routing unit. It is used to assess the credibility of at least one recalled knowledge semantic vector based on the knowledge association relationship and determine the target credibility score corresponding to at least one recalled knowledge semantic vector.

[0105] For example, the credibility assessment unit combines knowledge association relationships and scores the credibility of recalled knowledge semantic vectors in multiple dimensions according to preset assessment dimensions, quantifies the credibility of knowledge, and determines the target credibility score corresponding to at least one recalled knowledge semantic vector.

[0106] This application provides a knowledge credibility assessment system, which includes: a knowledge base unit, a knowledge enhancement unit, a data input unit, a knowledge routing unit, and a credibility assessment unit. The knowledge base unit provides knowledge documents and knowledge graphs of the power grid. The knowledge enhancement unit, connected to the knowledge base unit, performs vector transformation and association processing on the knowledge documents and knowledge graphs to obtain knowledge semantic vectors and knowledge association relationships. The data input unit, connected to the knowledge routing unit, acquires user-input query data and transmits the query data to the knowledge routing unit. The knowledge routing unit, connected to the knowledge enhancement unit, determines at least one recalled knowledge semantic vector corresponding to the query data from the knowledge semantic vectors. The credibility assessment unit, connected to the knowledge enhancement unit and the knowledge routing unit, assesses the knowledge credibility of at least one recalled knowledge semantic vector based on the knowledge association relationships, and determines a target credibility score corresponding to at least one recalled knowledge semantic vector. By applying the technical solution of this application embodiment, the knowledge enhancement unit converts the power grid's knowledge documents and knowledge graphs into semantic vectors, the knowledge routing unit performs rapid retrieval of user query data, and the credibility assessment unit then accurately assesses the credibility of the recalled knowledge.

[0107] Based on the above embodiments, Figure 2 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 2 .like Figure 2 As shown, the credibility assessment unit includes: a multi-dimensional assessment unit, a weight adjustment unit, and a recall decay unit;

[0108] The following implementation instructions are provided for this knowledge credibility assessment system:

[0109] 1) A multi-dimensional evaluation unit, used to determine the basic credibility score corresponding to at least one recalled knowledge semantic vector based on preset evaluation dimensions and preset hierarchical weights;

[0110] For example, the multidimensional evaluation unit divides the recalled knowledge semantic vector into different dimensions according to the preset evaluation dimensions, combines the preset hierarchical weights, quantifies the credibility level of each dimension, and obtains the basic credibility score corresponding to at least one recalled knowledge semantic vector by weighted summation of the divided vectors of different dimensions.

[0111] In one possible implementation, the preset evaluation dimensions include the authority of the knowledge source, the timeliness of the version, the accuracy of historical recall, the completeness of the content, and the relevance of the citations, and the preset hierarchical weights include the basic weights of the preset evaluation dimensions.

[0112] 2) The weight adjustment unit and the multi-dimensional evaluation unit are connected to update the basic credibility score corresponding to at least one recalled knowledge semantic vector according to the knowledge association relationship and the preset adjustment factor, so as to obtain the updated credibility score.

[0113] For example, the weight adjustment unit combines knowledge associations and preset adjustment factors to iteratively update the basic credibility score.

[0114] In one possible implementation, knowledge association focuses on "citation association" (such as the number of times the knowledge is cited by the core standard), with preset adjustment factors including citation enhancement factor (0.1-0.3) and structure adaptation factor (0.8-1.2).

[0115] If the basic credibility score of the recalled knowledge semantic vector A is 0.87, and it is cited by 3 core standards, the citation enhancement factor is determined to be 0.2, and the knowledge graph structure to which the recalled knowledge semantic vector A belongs is highly adapted to the query scenario, the structure adaptation factor is determined to be 1.1, then the updated credibility score is 0.87×(1+0.2)×1.1=1.148, and after normalization, it is finally 0.95 (the upper limit of the score is set to 1.0).

[0116] 3) The recall decay unit and the weight adjustment unit are connected to correct the updated credibility score based on historical recall data and determine the target credibility score corresponding to at least one recall knowledge semantic vector.

[0117] For example, the recall attenuation unit refers to historical recall data to calibrate the updated confidence score, so as to avoid the impact of poorly performing knowledge sources on the accuracy of the assessment.

[0118] In one possible implementation,

[0119] Historical recall data is analyzed in detail for the past three months for the "recall non-adoption rate" (i.e. the proportion of recalled but not selected by users). A preset decay coefficient is set and correlated with the non-adoption rate. For example, a non-adoption rate of 0-20% corresponds to no decay, 20%-50% corresponds to a decay of 0.1-0.3, and greater than 50% corresponds to a decay of 0.3-0.5.

[0120] After the recall knowledge semantic vector A is updated, the score is 0.95. Its historical recall non-adoption rate is 35%, corresponding to a decay coefficient of 0.2. Therefore, the target credibility score is 0.95×(1-0.2)=0.76, which not only retains the core credibility value of the knowledge, but also reflects the historical usage feedback through decay correction.

[0121] The knowledge credibility assessment system provided in this application includes a credibility assessment unit comprising a multi-dimensional assessment unit, a weight adjustment unit, and a recall attenuation unit. The multi-dimensional assessment unit is used to determine the basic credibility score corresponding to at least one recalled knowledge semantic vector based on preset assessment dimensions and preset hierarchical weights. The weight adjustment unit is connected to the multi-dimensional assessment unit and is used to update the basic credibility score corresponding to at least one recalled knowledge semantic vector based on knowledge association relationships and preset adjustment factors to obtain an updated credibility score. The recall attenuation unit is connected to the weight adjustment unit and is used to correct the updated credibility score based on historical recall data to determine the target credibility score corresponding to at least one recalled knowledge semantic vector. By applying the technical solution of this application embodiment, the multi-dimensional evaluation unit performs basic credibility scoring on the recalled semantic vectors according to preset evaluation dimensions and hierarchical weights, which can comprehensively reflect the quality of the recalled semantic vectors; the weight adjustment unit updates the basic credibility score through dynamic processing of knowledge association relationships and preset adjustment factors, thereby improving the accuracy of credibility evaluation; the recall decay unit uses historical recall data to correct the updated credibility score, ensuring the real-time performance and reliability of credibility evaluation.

[0122] Based on the above embodiments, Figure 3 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 3 .like Figure 3 As shown, the multidimensional evaluation unit includes: a feature vector construction subunit, a weight management subunit, a comprehensive scoring subunit, and an uncertainty handling subunit;

[0123] The following implementation instructions are provided for this knowledge credibility assessment system:

[0124] 1) Feature vector construction sub-unit, used to construct a multi-dimensional evaluation vector corresponding to at least one recalled knowledge semantic vector based on a preset evaluation dimension;

[0125] For example, based on the preset evaluation dimensions that reflect the core influencing factors of power grid knowledge credibility, the recalled knowledge semantic vector is transformed into a structured multi-dimensional evaluation vector to ensure that the knowledge credibility evaluation covers the key dimensions of knowledge quality.

[0126] In one possible implementation, the preset evaluation dimensions include five categories: knowledge source authority, version timeliness, historical recall accuracy, content completeness, and citation relevance. The multidimensional evaluation vector can then be represented as:

[0127]

[0128] in, This indicates the score for the authority of the knowledge source. This indicates the version's timeliness score. This represents the historical recall accuracy score. The score represents the completeness of the content. This indicates the score for the relevance of the reference.

[0129] For example, a document updated a month ago will have a timeliness score. It could be 0.85, representing knowledge with a complete structure and detailed attributes; its completeness score is... It could be 0.88.

[0130] 2) Weight management sub-unit, used to obtain preset hierarchical weights;

[0131] For example, the weight management subunit highlights the impact of key evaluation indicators by pre-setting the weight ratio of different dimensions, making the final score more in line with the knowledge application scenarios of the power grid industry.

[0132] In one possible implementation, the preset hierarchical weights include first-level basic weights, second-level type adjustment weights, and third-level scenario fine-tuning weights.

[0133] The primary basic weights can be represented as:

[0134]

[0135] in, This indicates the weight of the authority dimension of the knowledge source. This indicates the weight of the version timeliness dimension. This indicates the weight of the historical recall accuracy dimension. The weight representing the content completeness dimension. This represents the weight of the reference relevance dimension, and the sum of the weights of the five dimensions is 1.

[0136] For example, if the first-level basic weight is (0.3, 0.25, 0.2, 0.15, 0.1), for safety production standards, the second-level type adjustment weight adjusts the weight of the knowledge source authority dimension and the version timeliness dimension. If applied to a 10kV power distribution scenario, the third-level fine-tuning weight increases the weight of the relevance dimension, ultimately obtaining a weight vector (0.33, 0.3, 0.2, 0.15, 0.12) adapted to a specific scenario.

[0137] 3) The comprehensive scoring subunit is connected to the feature vector construction subunit and the weight management subunit. It is used to perform weighted summation on at least one multidimensional evaluation vector according to the preset hierarchical weights to obtain the basic credibility score corresponding to at least one recalled knowledge semantic vector.

[0138] For example, by weighting and summing the multidimensional evaluation vectors with hierarchical weights, the multidimensional qualitative evaluation is transformed into a single quantitative score, ensuring that the score not only fully covers all dimensions, but also reflects the importance of different factors through weights.

[0139] In one possible implementation, the basic credibility score The calculation formula is:

[0140]

[0141] in, Indicates the first Weights of each dimension Represents the first in the multidimensional evaluation vector Scores for each dimension.

[0142] 4) The uncertainty processing subunit is connected to the comprehensive scoring subunit and is used to update the basic credibility score corresponding to at least one recalled knowledge semantic vector according to the uncertainty penalty factor.

[0143] For example, the uncertainty processing subunit identifies the degree of dimensional dispersion of the multidimensional evaluation vector, introduces a penalty factor to correct the basic credibility score, and avoids score distortion caused by excessive differences in scores of each dimension.

[0144] In one possible implementation, when the differences in multi-dimensional scores are too large, an uncertainty penalty factor is introduced to adjust the final score. Uncertainty Penalty Factor The calculation uses the standard deviation method:

[0145]

[0146]

[0147]

[0148] in, Representing vectors The standard deviation of each element in the sample. express The average value of each element in the mixture.

[0149] In one possible implementation, the updated base credibility score The calculation formula is:

[0150]

[0151] in, This represents the basic credibility score. The uncertainty penalty factor represents the uncertainty penalty factor. When the consistency of the multi-dimensional scores is high, the uncertainty penalty factor is close to 0, and the original score is basically maintained. When the difference of the multi-dimensional scores is large, the uncertainty penalty factor increases and the final score decreases.

[0152] The knowledge credibility assessment system provided in this application includes a multi-dimensional assessment unit comprising: a feature vector construction subunit, a weight management subunit, a comprehensive scoring subunit, and an uncertainty processing subunit. The feature vector construction subunit is used to construct a multi-dimensional assessment vector corresponding to at least one recalled knowledge semantic vector based on a preset assessment dimension. The weight management subunit is used to obtain preset hierarchical weights. The comprehensive scoring subunit is connected to the feature vector construction subunit and the weight management subunit, and is used to perform a weighted summation of at least one multi-dimensional assessment vector according to the preset hierarchical weights to obtain a basic credibility score corresponding to at least one recalled knowledge semantic vector. The uncertainty processing subunit is connected to the comprehensive scoring subunit, and is used to update the basic credibility score corresponding to at least one recalled knowledge semantic vector according to an uncertainty penalty factor. By applying the technical solution of this application embodiment, the feature vector construction subunit generates multi-dimensional assessment vectors based on preset assessment dimensions, enabling each recalled knowledge semantic vector to have a richer and more comprehensive description during assessment. The multi-dimensional assessment unit, through the comprehensive application of feature vector construction, weight management, comprehensive scoring, and uncertainty processing, effectively improves the accuracy and reliability of information credibility assessment.

[0153] Based on the above embodiments, Figure 4 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 4 .like Figure 4 As shown, the weight adjustment unit includes: a knowledge graph construction subunit, a similarity determination subunit, a weight update subunit, and a weight propagation subunit;

[0154] The following implementation instructions are provided for this knowledge credibility assessment system:

[0155] 1) Knowledge graph construction sub-unit, used to construct the knowledge graph structure based on knowledge relationships;

[0156] Among them, the knowledge association relationship includes at least one knowledge source and the association relationship between different knowledge sources, and the knowledge graph structure includes at least one node corresponding to a knowledge source and the edge corresponding to the association relationship between different knowledge sources;

[0157] For example, the knowledge graph construction subunit abstracts each knowledge source as a node, and the relationships between knowledge sources, such as ownership and reference, are transformed into edges, thus transforming the scattered knowledge relationships into a structured knowledge graph structure, clearly presenting the relationship structure of the knowledge network.

[0158] In one possible implementation, the knowledge graph structure is a directed weighted graph, denoted as G(V, E), where V represents the set of nodes (knowledge sources) and E represents the set of edges (relationships between knowledge sources).

[0159] 2) The similarity determination subunit is connected to the knowledge graph construction subunit to determine the first similarity value between the first knowledge source and at least one second knowledge source corresponding to at least one recall knowledge semantic vector based on the knowledge graph structure;

[0160] Among them, at least one second knowledge source is related to the first knowledge source;

[0161] In one possible implementation, the similarity determination subunit performs multi-level similarity calculation through hierarchical text encoding, domain-adaptive processing, and context enhancement, achieving lexical, semantic, structural, and citation-level similarity calculations, and obtains the first similarity value through weighted fusion.

[0162] 3) The weight update subunit is connected to the similarity determination subunit, which is used to update the basic credibility score corresponding to at least one recalled knowledge semantic vector according to the first similarity value and the preset adjustment factor, so as to obtain the updated credibility score.

[0163] For example, the preset adjustment factors include a similarity adjustment factor, a graph structure adjustment factor, and a time decay adjustment factor;

[0164] The similarity adjustment factor is determined based on the normalization of the first similarity value. The graph structure adjustment factor is determined based on the centrality and authority of the node in the knowledge graph structure. The time decay adjustment factor is determined based on the timeliness of the knowledge. The weight ratio of the three factors is usually set to 4:3:3, which can be fine-tuned according to the actual application scenario.

[0165] In one possible implementation, the basic credibility score is updated according to the weight ratio of a preset adjustment factor, and then normalized to obtain the updated credibility score.

[0166] 4) The weight propagation subunit is connected to the weight update subunit, and is used to perform weight adjustment processing on at least one second knowledge source according to the updated credibility score.

[0167] For example, during weight propagation, a change in the weight of one node will affect the nodes directly connected to it, with the intensity of the impact decreasing as the relationship strength and distance increase. For instance, when the weight of a core standard document increases, the weight of documents that directly cite it will increase accordingly, while the weight of indirectly cited documents will increase less. Therefore, based on the association between at least one second knowledge source and the first knowledge source, the weight of at least one second knowledge source is adjusted according to the updated credibility score.

[0168] The knowledge credibility evaluation system provided in this application includes a weight adjustment unit comprising: a knowledge graph construction subunit, a similarity determination subunit, a weight update subunit, and a weight propagation subunit. The knowledge graph construction subunit is used to construct a knowledge graph structure based on knowledge relationships, wherein the knowledge relationships include at least one knowledge source and relationships between different knowledge sources, and the knowledge graph structure includes nodes corresponding to at least one knowledge source and edges corresponding to relationships between different knowledge sources. The similarity determination subunit, connected to the knowledge graph construction subunit, is used to determine a first similarity value between at least one first knowledge source corresponding to a recalled knowledge semantic vector and at least one second knowledge source, based on the knowledge graph structure, wherein at least one second knowledge source has a relationship with the first knowledge source. The weight update subunit, connected to the similarity determination subunit, is used to update the basic credibility score corresponding to at least one recalled knowledge semantic vector based on the first similarity value and a preset adjustment factor to obtain an updated credibility score. The weight propagation subunit, connected to the weight update subunit, is used to perform weight adjustment processing on at least one second knowledge source based on the updated credibility score. By applying the technical solution of this application embodiment, the knowledge graph construction subunit constructs a knowledge graph structure based on knowledge associations, which helps to better understand and analyze the relationships between knowledge; the similarity determination subunit uses the knowledge graph structure to calculate the similarity values ​​between different knowledge sources, providing more accurate data support for subsequent weight updates, thereby enhancing the rationality and effectiveness of the evaluation results; the weight update subunit updates the basic credibility score based on the similarity value and a preset adjustment factor, and the weight propagation subunit adjusts the weights of relevant knowledge sources based on the updated credibility score, ensuring that the evaluation of knowledge credibility can be propagated throughout the entire knowledge graph structure.

[0169] Based on the above embodiments, Figure 5 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 5 .like Figure 5As shown, the recall attenuation unit includes: a time window subunit, a filter rate monitoring subunit, a threshold management subunit, an attenuation strategy subunit, and a recovery balance subunit;

[0170] The following implementation instructions are provided for this knowledge credibility assessment system:

[0171] 1) Time window sub-unit, used to divide the updated confidence score into windows according to preset time window parameters, to obtain at least one time window and corresponding window data;

[0172] For example, the time window subunit segments the recall records corresponding to the updated confidence score by time according to the preset time window parameters, providing structured data for subsequent filtering rate calculation.

[0173] In one possible implementation, at least one time window includes a micro window, a meso window, and a macro window. The micro window is typically set to 1 to 7 days to capture short-term fluctuations; the meso window is set to 8 to 30 days to identify trend changes; and the macro window is set to more than 30 days to assess long-term performance.

[0174] 2) The filtration rate monitoring subunit is connected to the time window subunit to determine the filtration rate data corresponding to at least one time window;

[0175] For example, the filtering rate data is used to indicate the percentage of knowledge sources that are recalled but not adopted within a certain period of time. The formula for calculating the filtering rate is as follows:

[0176]

[0177] in, Indicates time window Knowledge source corresponding to inner window data Number of times filtered, Indicates time window Knowledge source corresponding to inner window data Total number of recalls, time window It can be any of the micro, meso, or macro windows, and the filtering rate ranges from [0, 1]. The larger the value, the more serious the quality problem of the knowledge source.

[0178] 3) The threshold management subunit is connected to the filter rate monitoring subunit and is used to determine the adjustment threshold corresponding to the updated confidence score based on historical recall data and filter rate data corresponding to at least one time window.

[0179] For example, the threshold management subunit combines historical recall data and current window filtering rate data to set a reasonable adjustment threshold and distinguish whether knowledge needs to be decayed or restored.

[0180] In one possible implementation, in power grid applications, the filtering rate threshold is typically initialized to 0.3, meaning that a weight adjustment will be triggered when the filtering rate of a knowledge source exceeds 30% within a monitoring period. This threshold can be dynamically adjusted according to the system's operating status; for example, it may be appropriately relaxed to 0.35 during peak system load periods.

[0181] 4) The decay strategy subunit is connected to the threshold management subunit and is used to correct the updated credibility score according to the adjusted threshold, and determine the target credibility score corresponding to at least one recalled knowledge semantic vector.

[0182] For example, a decay strategy is determined based on the adjusted threshold, and the target credibility score is reduced through the decay mechanism to avoid low-credibility knowledge being given priority recommendation.

[0183] In one possible implementation, the decay strategy includes linear decay, exponential decay, or step decay.

[0184] Typically, a linear decay strategy is used, as shown below:

[0185]

[0186]

[0187] in, This represents the attenuation coefficient, which is usually set to 0.5. This indicates the current credibility score of the knowledge source. This indicates the target credibility score after correction. This indicates the portion of the filter rate that exceeds the adjustment threshold. This indicates that the threshold has been adjusted and set to 0.3. This represents the filtration rate data.

[0188] For knowledge sources with serious problems, an exponential decay strategy can be used, as shown below:

[0189]

[0190] in, This represents the exponential decay coefficient, which is usually set to 2.0.

[0191] 5) The recovery balance subunit is used to restore the updated confidence score when the filter rate data corresponding to at least one time window is less than the preset threshold.

[0192] For example, the recovery and balancing subunit replenishes the scores of knowledge that improves performance. When the filtering rate is lower than a preset threshold, the credibility score is increased through recovery processing to ensure that high-quality knowledge is not underestimated in the long term.

[0193] In one possible implementation, when a knowledge source performs well in a new monitoring period (filtering rate below 50% of the threshold), its weight can be gradually restored:

[0194]

[0195]

[0196] in, This represents the coefficient of recovery, which is typically set to 0.3. Indicators that measure the degree of recovery This represents the initial basic credibility score of the knowledge source.

[0197] The knowledge credibility assessment system provided in this application includes a recall attenuation unit comprising: a time window subunit, a filter rate monitoring subunit, a threshold management subunit, an attenuation strategy subunit, and a recovery balance subunit. The time window subunit is used to divide the updated credibility score into windows according to preset time window parameters, obtaining at least one time window and corresponding window data. The filter rate monitoring subunit is connected to the time window subunit and is used to determine the filter rate data corresponding to at least one time window. The threshold management subunit is connected to the filter rate monitoring subunit and is used to determine the adjustment threshold corresponding to the updated credibility score based on historical recall data and the filter rate data corresponding to at least one time window. The attenuation strategy subunit is connected to the threshold management subunit and is used to correct the updated credibility score according to the adjustment threshold, determining the target credibility score corresponding to at least one recalled knowledge semantic vector. The recovery balance subunit is used to restore the updated credibility score when the filter rate data corresponding to at least one time window is less than a preset threshold. By applying the technical solutions of the embodiments of this application, and comprehensively utilizing time window division, filter rate monitoring, threshold management, decay strategy, and recovery balance, not only is the accuracy and timeliness of knowledge credibility assessment improved, but the flexibility and adaptability of the system are also enhanced.

[0198] Based on the above embodiments, Figure 6 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 6 .like Figure 6 As shown, the knowledge credibility assessment system also includes: a source tracing unit and a visualization unit;

[0199] The following implementation instructions are provided for this knowledge credibility assessment system:

[0200] 1) The source tracing unit is connected to the knowledge enhancement unit and the knowledge routing unit, and is used to determine the knowledge source tracing path corresponding to at least one recalled knowledge semantic vector based on the knowledge association relationship and at least one recalled knowledge semantic vector;

[0201] For example, the source tracing unit traces the source and flow path of knowledge in reverse according to the node relationships in the knowledge graph structure corresponding to the knowledge association relationship, connecting the complete link from the original release of knowledge to the final recall.

[0202] In one possible implementation, the source tracing unit uses a graph traversal algorithm to start from the target knowledge node corresponding to at least one recalled knowledge semantic vector, trace back to the source node along the reverse relationship link, and record the complete path.

[0203] 2) The visualization unit is connected to the source tracing unit and is used to visualize the source tracing path corresponding to at least one recalled knowledge semantic vector.

[0204] For example, visualization units transform abstract knowledge tracing paths into intuitive and easy-to-understand visual presentations.

[0205] In one possible implementation, a dual-view display of "graph + list" is used: the graph view presents the tracing path in the form of nodes and edges, with different types of nodes marked with different colors and edges marked with relationship types; the list view displays node details in the tracing order, including information such as document number, publication time, and credibility score.

[0206] The knowledge credibility assessment system provided in this application includes a tracing unit and a visualization unit. The tracing unit is connected to a knowledge enhancement unit and a knowledge routing unit, and is used to determine the knowledge tracing path corresponding to at least one recalled knowledge semantic vector based on knowledge association relationships and at least one recalled knowledge semantic vector. The visualization unit is connected to the tracing unit and is used to visualize the knowledge tracing path corresponding to at least one recalled knowledge semantic vector. By applying the technical solution of this application, the tracing unit, by connecting the knowledge enhancement unit and the knowledge routing unit, can determine the knowledge tracing path based on knowledge association relationships and recalled knowledge semantic vectors, allowing users to clearly understand the origin and transmission path of recalled knowledge, thereby improving knowledge credibility. The visualization unit visualizes the knowledge tracing path, presenting complex knowledge relationships in a graphical way. The combination of the tracing unit and the visualization unit not only improves the transparency and credibility of knowledge but also provides users with a more intuitive and easy-to-understand information display method, thereby supporting a more scientific and effective decision-making process.

[0207] Based on the above embodiments, Figure 7 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 7 .like Figure 7As shown, the knowledge credibility evaluation system also includes: a text similarity unit, a data filtering unit, a recall verification unit, and a user feedback unit;

[0208] The following implementation instructions are provided for this knowledge credibility assessment system:

[0209] 1) The text similarity unit is connected to the knowledge routing unit and the knowledge enhancement unit to determine at least one recall knowledge semantic vector and a second similarity value between the knowledge semantic vectors;

[0210] For example, the text similarity unit quantifies the correlation strength between the recalled knowledge semantic vector and the knowledge semantic vector in the knowledge base through similarity calculation.

[0211] 2) The data filtering unit is connected to the text similarity unit and is used to filter at least one recalled knowledge semantic vector based on the second similarity value and the preset text similarity threshold to obtain the filtered recalled knowledge semantic vector.

[0212] For example, the data filtering unit mainly filters duplicate, irrelevant, and outdated content from at least one recalled knowledge semantic vector.

[0213] For duplicate content, a text similarity threshold method is used for detection, and content with a similarity score exceeding 0.85 is considered duplicate; for irrelevant content, filtering is performed by combining query intent and text relevance score; for outdated content, screening is performed by combining knowledge timeliness score.

[0214] 3) The recall verification unit is connected to the data filtering unit to verify the filtered recall knowledge semantic vector and determine the verified recall knowledge semantic vector.

[0215] For example, a knowledge graph reasoning method is used to verify the logical relevance and rationality of the recalled knowledge after screening.

[0216] In one possible implementation, the recall verification unit employs graph neural network technology. Based on the association between "concept nodes and document content" in the aforementioned knowledge graph, it learns the implicit relationships between nodes through multi-layer graph convolution operations to verify the rationality of the filtered recall knowledge semantic vector.

[0217] 4) The user feedback unit is connected to the recall verification unit to evaluate the verified recalled knowledge semantic vector and obtain knowledge confidence data;

[0218] For example, the user feedback unit receives the user's confidence rating of the input keywords and recall results, including a five-level rating mechanism (very trustworthy, somewhat trustworthy, neutral, somewhat trustworthy, and completely untrustworthy), and supports text feedback.

[0219] 5) The credibility assessment unit is connected to the user feedback unit and is used to adjust the credibility assessment based on the knowledge confidence data.

[0220] For example, the credibility assessment unit uses user feedback data (i.e. knowledge confidence data) to dynamically optimize the assessment system, integrates subjective feedback into the objective scoring model, and realizes closed-loop iteration of credibility assessment.

[0221] The knowledge credibility evaluation system provided in this application embodiment further includes: a text similarity unit, a data filtering unit, a recall verification unit, and a user feedback unit. The text similarity unit is connected to the knowledge routing unit and the knowledge enhancement unit, and is used to determine a second similarity value between at least one recalled knowledge semantic vector and the knowledge semantic vector. The data filtering unit is connected to the text similarity unit, and is used to filter at least one recalled knowledge semantic vector according to the second similarity value and a preset text similarity threshold to obtain a filtered recalled knowledge semantic vector. The recall verification unit is connected to the data filtering unit, and is used to verify the filtered recalled knowledge semantic vector to determine a verified recalled knowledge semantic vector. The user feedback unit is connected to the recall verification unit, and is used to evaluate the verified recalled knowledge semantic vector to obtain knowledge confidence data. The credibility evaluation unit is connected to the user feedback unit, and is used to adjust the credibility evaluation based on the knowledge confidence data. By applying the technical solution of this application embodiment, the combination of the text similarity unit, data filtering unit, recall verification unit and user feedback unit not only improves the relevance and reliability of knowledge, but also forms a closed-loop mechanism through user feedback, realizing the dynamic optimization and intelligence of the system, thereby improving the accuracy of knowledge credibility assessment.

[0222] Based on the above embodiments, Figure 8 Schematic diagram of the knowledge credibility assessment system provided in the embodiments of this application Figure 8 .like Figure 8 As shown, the knowledge credibility assessment system also includes: a data storage unit and a data query unit;

[0223] The following implementation instructions are provided for this knowledge credibility assessment system:

[0224] 1) The data storage unit is connected to the knowledge base unit and the knowledge enhancement unit to store knowledge documents, knowledge graphs, knowledge semantic vectors, and knowledge relationships;

[0225] For example, the data storage unit uses the distributed file system HDFS to build the underlying storage architecture, and combines the graph database Neo4j to store the knowledge graph.

[0226] HDFS is used to store unstructured data such as power grid standard documents and knowledge semantic vectors; Neo4j is used to store nodes and knowledge relationships (such as references and ownership relationships) in the knowledge graph.

[0227] 2) The data query unit is connected to the knowledge base unit and the data input unit, and is used to query the query data based on the knowledge graph to determine the corresponding related graph data.

[0228] For example, the data query unit receives user queries transmitted by the data input unit, extracts professional concepts from the query data, retrieves knowledge graphs using the Cypher query language, and determines the associated graph data corresponding to the query data.

[0229] The knowledge credibility assessment system provided in this application includes a data storage unit and a data query unit. The data storage unit is connected to a knowledge base unit and a knowledge enhancement unit, and is used to store knowledge documents, knowledge graphs, knowledge semantic vectors, and knowledge relationships. The data query unit is connected to the knowledge base unit and a data input unit, and is used to query the query data based on the knowledge graph to determine the corresponding association graph data. By applying the technical solution of this application, the connection between the data storage unit and the knowledge base unit and the knowledge enhancement unit enables centralized storage of knowledge documents, knowledge graphs, knowledge semantic vectors, and knowledge relationships. This centralized management helps improve data integration and accessibility, ensuring that users can quickly obtain the information they need. The data query unit, through its linkage with the knowledge base unit and the data input unit, can efficiently retrieve the query data based on the knowledge graph. The integration of the data storage unit and the data query unit can effectively improve the information management and access efficiency of the knowledge credibility assessment system.

[0230] Figure 9 This is a flowchart illustrating the knowledge credibility assessment method provided in the embodiments of this application. Figure 9 As shown, the knowledge credibility assessment method is applied to the credibility assessment unit in the aforementioned knowledge credibility assessment system. The method includes the following steps:

[0231] S910. Obtain at least one recall knowledge semantic vector corresponding to the query data input by the user.

[0232] In this step, in order to evaluate the credibility of the knowledge, we can first obtain at least one recall knowledge semantic vector corresponding to the query data input by the user.

[0233] In one possible implementation, the user-input query data can be obtained through the data input unit of the aforementioned knowledge credibility evaluation system, and at least one recall knowledge semantic vector corresponding to the query data can be determined according to the knowledge enhancement unit.

[0234] S920. Based on the preset evaluation dimensions and preset hierarchical weights, determine the basic credibility score corresponding to at least one recalled knowledge semantic vector.

[0235] In this step, at least one recalled knowledge semantic vector is evaluated in multiple dimensions according to the preset evaluation dimensions. Combined with the preset hierarchical weights, the multi-dimensional evaluation scores are weighted and summed to determine the basic credibility score corresponding to at least one recalled knowledge semantic vector.

[0236] For example, the preset evaluation dimensions include the authority of the knowledge source, the timeliness of the version, the accuracy of historical recall, the completeness of the content, and the relevance of the citation.

[0237] In one possible implementation, it can be achieved through a multi-dimensional evaluation unit in the aforementioned knowledge credibility evaluation system.

[0238] S930. Update the basic credibility score corresponding to at least one recalled knowledge semantic vector according to the knowledge association relationship and the preset adjustment factor to obtain the updated credibility score.

[0239] For example, based on knowledge associations, a knowledge graph structure is constructed, and a first similarity value is determined between a first knowledge source and at least a second knowledge source corresponding to at least one recalled knowledge semantic vector. Then, the basic credibility score corresponding to at least one recalled knowledge semantic vector is updated based on the similarity value and a preset adjustment factor to obtain the updated credibility score.

[0240] In one possible implementation, this can be achieved through the weight adjustment unit in the aforementioned knowledge credibility assessment system.

[0241] S940. Based on historical recall data, the updated credibility score is corrected to determine the target credibility score corresponding to at least one recalled knowledge semantic vector.

[0242] For example, based on historical recall data, the ratio of knowledge sources corresponding to at least one recalled knowledge semantic vector that were recalled but not adopted is determined, and the updated credibility score is corrected according to the adjustment threshold to determine the target credibility score corresponding to at least one recalled knowledge semantic vector.

[0243] In one possible implementation, it can be achieved through the recall decay unit in the aforementioned knowledge credibility assessment system.

[0244] In one possible implementation, the method for evaluating the credibility of knowledge further includes the following:

[0245] Step 1: Sort at least one recalled knowledge semantic vector according to the target credibility score corresponding to at least one recalled knowledge semantic vector to obtain the sorted recalled knowledge semantic vector.

[0246] Step 2: Determine the third similarity value between at least two recall knowledge semantic vectors in the sorted recall knowledge semantic vectors;

[0247] For example, by calculating the third similarity value, the semantic redundancy between the sorted recall results is quantified, highly similar redundant vectors are identified, and a basis is provided for subsequent fusion processing.

[0248] Step 3: The recall knowledge semantic vectors with a third similarity value greater than a preset threshold are fused to obtain the fused recall knowledge semantic vector.

[0249] For example, semantic vectors of recalled knowledge with high semantic repetition can be fused to form a more comprehensive and concise fused vector, avoiding content redundancy.

[0250] Step 4: Based on the preset recall strategy, filter the fused recall knowledge semantic vectors to determine the target semantic vectors.

[0251] For example, the preset recall strategy includes credibility, completeness, and timeliness screening dimensions. According to the preset recall strategy, the target semantic vector is obtained by sequentially screening from the fused recall knowledge semantic vector according to "credibility priority + full coverage of core professional concepts + timeliness screening".

[0252] The knowledge credibility assessment method provided in this application embodiment is applied to the credibility assessment unit in the aforementioned knowledge credibility assessment system. First, it obtains at least one recalled knowledge semantic vector corresponding to the query data input by the user. Then, it determines the basic credibility score corresponding to at least one recalled knowledge semantic vector based on the preset assessment dimension and preset hierarchical weight. Next, it updates the basic credibility score corresponding to at least one recalled knowledge semantic vector based on the knowledge association relationship and preset adjustment factor to obtain the updated credibility score. Finally, it corrects the updated credibility score based on historical recall data to determine the target credibility score corresponding to at least one recalled knowledge semantic vector. In this embodiment, based on the query data input by the user, the knowledge semantic vector is recalled to ensure that the most relevant information can be found based on the user's needs. By using preset evaluation dimensions and hierarchical weights, the credibility of the recalled knowledge can be evaluated from multiple perspectives, avoiding the bias that may be caused by a single indicator. The basic credibility score is updated by combining knowledge association relationships and preset adjustment factors, which can reflect the relative importance and credibility of knowledge in practical applications. The updated credibility score is corrected by using historical recall data, which can effectively reduce the impact of random factors on the evaluation results, improve the stability and reliability of the score, and make the final target credibility score more realistic and credible.

[0253] Based on the above embodiments, the following is a knowledge credibility assessment apparatus provided in the embodiments of this application, which can execute the methods provided in the above method embodiments.

[0254] Figure 10 A schematic diagram of the structure of the knowledge credibility assessment device provided in the embodiments of this application. Figure 10 As shown, the knowledge credibility assessment device 1000 is applied to the credibility assessment unit in the aforementioned knowledge credibility assessment system, and includes:

[0255] The acquisition module 1010 is used to acquire at least one semantic vector of recall knowledge corresponding to the query data input by the user;

[0256] The determination module 1020 is used to determine the basic credibility score corresponding to at least one recalled knowledge semantic vector based on the preset evaluation dimensions and preset hierarchical weights.

[0257] The first processing module 1030 is used to update the basic credibility score corresponding to at least one recalled knowledge semantic vector according to the knowledge association relationship and the preset adjustment factor, so as to obtain the updated credibility score.

[0258] The second processing module 1040 is used to correct the updated credibility score based on historical recall data and determine the target credibility score corresponding to at least one recall knowledge semantic vector.

[0259] In one or more embodiments, the second processing module 1040 is further configured to:

[0260] Based on the target credibility score corresponding to at least one recalled knowledge semantic vector, sort at least one recalled knowledge semantic vector to obtain the sorted recalled knowledge semantic vector.

[0261] Determine the third similarity value between at least two recalled knowledge semantic vectors in the sorted recalled knowledge semantic vectors;

[0262] The recalled knowledge semantic vectors with a third similarity value greater than a preset threshold are fused together to obtain a fused recalled knowledge semantic vector.

[0263] The target semantic vector is determined by filtering the fused recall knowledge semantic vectors according to the preset recall strategy.

[0264] The knowledge credibility assessment device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0265] This application also provides a computer-readable storage medium storing computer instructions that, when executed on a computer, cause the computer to perform the aforementioned related steps.

[0266] The aforementioned computer-readable storage medium can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0267] Optionally, a readable storage medium can be coupled to the processor, enabling the processor to read information from and write information to the readable storage medium. Alternatively, the readable storage medium can be an integral part of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0268] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium, and the above steps can be implemented when the at least one processor executes the computer program.

[0269] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A knowledge credibility assessment system, characterized in that, include: The system includes a knowledge base unit, a knowledge enhancement unit, a data input unit, a knowledge routing unit, and a credibility assessment unit. The knowledge base unit is used to provide knowledge documents and knowledge graphs for the power grid; The knowledge enhancement unit is connected to the knowledge base unit and is used to perform vector transformation and association processing on the knowledge document and the knowledge graph to obtain knowledge semantic vectors and knowledge association relationships. The data input unit is connected to the knowledge routing unit and is used to obtain the query data input by the user and transmit the query data to the knowledge routing unit; The knowledge routing unit is connected to the knowledge enhancement unit and is used to determine at least one recall knowledge semantic vector corresponding to the query data in the knowledge semantic vector. The credibility assessment unit is connected to the knowledge enhancement unit and the knowledge routing unit, and is used to assess the credibility of the at least one recalled knowledge semantic vector according to the knowledge association relationship, and determine the target credibility score corresponding to the at least one recalled knowledge semantic vector.

2. The system according to claim 1, characterized in that, The credibility assessment unit includes: a multi-dimensional assessment unit, a weight adjustment unit, and a recall decay unit; The multidimensional evaluation unit is used to determine the basic credibility score corresponding to the at least one recalled knowledge semantic vector based on the preset evaluation dimensions and preset hierarchical weights. The weight adjustment unit and the multidimensional evaluation unit are connected and are used to update the basic credibility score corresponding to the at least one recalled knowledge semantic vector according to the knowledge association relationship and the preset adjustment factor to obtain the updated credibility score. The recall attenuation unit and the weight adjustment unit are connected and used to correct the updated credibility score based on historical recall data, and determine the target credibility score corresponding to the at least one recall knowledge semantic vector.

3. The system according to claim 2, characterized in that, The multidimensional evaluation unit includes: a feature vector construction subunit, a weight management subunit, a comprehensive scoring subunit, and an uncertainty handling subunit; The feature vector construction subunit is used to construct a multi-dimensional evaluation vector corresponding to the at least one recalled knowledge semantic vector based on the preset evaluation dimension. The weight management subunit is used to obtain the preset hierarchical weights; The comprehensive scoring subunit is connected to the feature vector construction subunit and the weight management subunit, and is used to perform weighted summation on at least one multidimensional evaluation vector according to the preset hierarchical weights to obtain the basic credibility score corresponding to the at least one recalled knowledge semantic vector. The uncertainty processing subunit is connected to the comprehensive scoring subunit and is used to update the basic credibility score corresponding to the at least one recalled knowledge semantic vector according to the uncertainty penalty factor.

4. The system according to claim 2, characterized in that, The weight adjustment unit includes: a knowledge graph construction subunit, a similarity determination subunit, a weight update subunit, and a weight propagation subunit; The knowledge graph construction subunit is used to construct a knowledge graph structure based on the knowledge associations. The knowledge associations include at least one knowledge source and the associations between different knowledge sources. The knowledge graph structure includes nodes corresponding to the at least one knowledge source and edges corresponding to the associations between the different knowledge sources. The similarity determination subunit is connected to the knowledge graph construction subunit and is used to determine the first similarity value between the first knowledge source and at least one second knowledge source corresponding to the at least one recalled knowledge semantic vector according to the knowledge graph structure, wherein the at least one second knowledge source is associated with the first knowledge source; The weight update subunit is connected to the similarity determination subunit and is used to update the basic credibility score corresponding to the at least one recalled knowledge semantic vector according to the first similarity value and the preset adjustment factor to obtain the updated credibility score. The weight propagation subunit is connected to the weight update subunit and is used to perform weight adjustment processing on the at least one second knowledge source according to the updated credibility score.

5. The system according to claim 2, characterized in that, The recall attenuation unit includes: a time window subunit, a filter rate monitoring subunit, a threshold management subunit, an attenuation strategy subunit, and a recovery balance subunit; The time window subunit is used to divide the updated confidence score into windows according to preset time window parameters to obtain at least one time window and corresponding window data. The filtration rate monitoring subunit is connected to the time window subunit and is used to determine the filtration rate data corresponding to the at least one time window. The threshold management subunit is connected to the filter rate monitoring subunit and is used to determine the adjustment threshold corresponding to the updated confidence score based on the historical recall data and the filter rate data corresponding to the at least one time window. The attenuation strategy subunit is connected to the threshold management subunit and is used to correct the updated credibility score according to the adjustment threshold, and determine the target credibility score corresponding to the at least one recalled knowledge semantic vector. The recovery balance subunit is used to perform recovery processing on the updated credibility score when the filter rate data corresponding to the at least one time window is less than a preset threshold.

6. The system according to any one of claims 1-5, characterized in that, The system also includes: a source tracing unit and a visualization unit; The source tracing unit is connected to the knowledge enhancement unit and the knowledge routing unit, and is used to determine the knowledge source tracing path corresponding to the at least one recalled knowledge semantic vector based on the knowledge association relationship and the at least one recalled knowledge semantic vector; The visualization unit is connected to the source tracing unit and is used to visualize the knowledge source tracing path corresponding to the at least one recalled knowledge semantic vector.

7. The system according to any one of claims 1-5, characterized in that, The system also includes: a text similarity unit, a data filtering unit, a recall verification unit, and a user feedback unit; The text similarity unit is connected to the knowledge routing unit and the knowledge enhancement unit, and is used to determine the second similarity value between the at least one recalled knowledge semantic vector and the knowledge semantic vector. The data filtering unit is connected to the text similarity unit and is used to filter the at least one recalled knowledge semantic vector according to the second similarity value and the preset text similarity threshold to obtain the filtered recalled knowledge semantic vector. The recall verification unit is connected to the data filtering unit and is used to verify the filtered recall knowledge semantic vector and determine the verified recall knowledge semantic vector. The user feedback unit is connected to the recall verification unit and is used to evaluate the verified recalled knowledge semantic vector to obtain knowledge confidence data. The credibility assessment unit is connected to the user feedback unit and is used to adjust the credibility assessment based on the knowledge confidence data.

8. The system according to any one of claims 1-5, characterized in that, The system also includes: a data storage unit and a data query unit; The data storage unit is connected to the knowledge base unit and the knowledge enhancement unit, and is used to store the knowledge documents, the knowledge graph, the knowledge semantic vectors, and the knowledge associations. The data query unit is connected to the knowledge base unit and the data input unit, and is used to query the query data according to the knowledge graph to determine the associated graph data corresponding to the query data.

9. A method for assessing the credibility of knowledge, characterized in that, A credibility assessment unit applied in a knowledge credibility assessment system according to any one of claims 1-8, the method comprising: Obtain at least one semantic vector of recall knowledge corresponding to the query data input by the user; Based on the preset evaluation dimensions and preset hierarchical weights, the basic credibility score corresponding to the at least one recalled knowledge semantic vector is determined; The basic credibility score corresponding to the at least one recalled knowledge semantic vector is updated based on the knowledge association relationship and the preset adjustment factor to obtain the updated credibility score. The updated credibility score is corrected based on historical recall data to determine the target credibility score corresponding to the at least one recalled knowledge semantic vector.

10. The method according to claim 9, characterized in that, The method further includes: Based on the target credibility score corresponding to the at least one recalled knowledge semantic vector, the at least one recalled knowledge semantic vector is sorted to obtain the sorted recalled knowledge semantic vector. Determine the third similarity value between at least two recalled knowledge semantic vectors in the sorted recalled knowledge semantic vectors; The recall knowledge semantic vectors with a third similarity value greater than a preset threshold are fused together to obtain a fused recall knowledge semantic vector. The fused recall knowledge semantic vector is filtered according to a preset recall strategy to determine the target semantic vector.

Images