Urban knowledge correctness evaluation method and system based on multi-level abnormal signals

Abstract

The invention relates to the technical field of detection and identification of small targets in images in the power industry, and provides a city knowledge correctness evaluation method and system based on multi-level abnormal signals. The method comprises the steps that firstly, entity subgraphs are generated based on depth-first search, the association strength of entity pairs is calculated through the resource transfer amount, and entity-level credibility evaluation is constructed; secondly, innovatively providing a hierarchical path modeling method, dynamically allocating path weights in combination with an adaptive attention mechanism, and realizing credibility evaluation of a relationship level; and finally, the consistency of the entity type and the graph structure is introduced into an evaluation system as a core abnormal signal for the first time, semantic consistency is quantified through a relation hypergraph and a type hypergraph, and credibility evaluation of a type-graph structure level is generated. By integrating the entity association strength, the path credibility and the type-structure credibility, the model can generate the final credibility of the triple, and reliable theoretical and method support is provided for correctness evaluation of the knowledge graph.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge graphs, and in particular, to a method and system for evaluating the correctness of urban knowledge based on multi-level abnormal signals. Background Art

[0002] The statements in this part merely provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] The comprehensive digital transformation of cities has generated a large amount of data related to cities. These data cover multiple aspects such as transportation, environment, energy, and public services, providing rich information that helps to more comprehensively understand the urban operation status, optimize urban management, and promote the construction of smart cities. According to the data-information-knowledge hierarchy, these massive urban data can be transformed into information and then form knowledge about cities. With the continuous deepening of the construction of smart cities, the dependence of urban governance on knowledge is increasing. However, in a large and complex information environment, how to obtain reliable urban knowledge that can meet the requirements and achieve the goals has become a key problem that urgently needs to be solved in the field of urban governance.

[0004] In the research on knowledge trustworthiness evaluation, many existing methods mainly rely on the embedding representation of entities and relationships. For example, CKRL generates credibility scores for triples by mining the internal structure information of the knowledge graph and uses these scores to learn more robust representations. However, knowledge graphs in the real world often have complex path structures. There may be multiple paths between two entities, and each path is composed of multiple relationships. Although existing methods attempt to evaluate the rationality of triples through path information, they often ignore the hierarchical structure information contained in the paths. In addition, relying solely on the topological structure of the knowledge graph for evaluation often makes it difficult to comprehensively and accurately measure the trustworthiness of knowledge. Therefore, how to effectively utilize the hierarchical information in the paths and combine context or other external information has become an important research direction for improving the effect of knowledge trustworthiness evaluation.

[0005] Entity types are usually closely related to the graph structure, that is, entities with similar semantics are often connected to each other through specific relationships. For example, entities with entity types {location, campus, station, area} are usually connected to located in or belong to, while entities with types {educator, scientist} are more likely to be connected to relationships such as graduated. This indicates that if the type of an entity is inconsistent with the type of its adjacent relationship, the corresponding triple may have a low credibility. Taking the triple {Tsinghua Garden, graduated, Tsinghua University} as an example, Tsinghua Garden is a location and Tsinghua University is a university. By comparing the learned semantics with the relationship graduated, it can be easily inferred that this triple is incorrect because there cannot be a graduation relationship between a location and a school. Based on this, many existing knowledge-driven learning tasks or applications, such as question answering systems, recommendation systems, information retrieval, etc., often assume that the information in the knowledge graph is completely correct, so it is easy to introduce potential errors and biases, thus affecting the reliability of urban knowledge, the application effect, and the scientific nature of decision-making. Summary of the Invention

[0006] To solve the technical problems existing in the above-mentioned background technology, the present invention provides a method and system for evaluating the correctness of urban knowledge based on multi-level abnormal signals. By constructing a multi-level abnormal signal system, the present invention can evaluate triples from multiple dimensions at the entity level, relationship level, and type-graph structure level, significantly improving the ability to identify the credibility of triples. This multi-level abnormal signal system enables the abnormal signals at each level to be independently evaluated and effectively integrated with the information at other levels, thereby achieving a more accurate credibility evaluation.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] The first aspect of the present invention provides a method for evaluating the correctness of urban knowledge based on multi-level abnormal signals.

[0009] A method for evaluating the correctness of urban knowledge based on multi-level abnormal signals includes:

[0010] Based on the obtained urban knowledge graph, generate an entity subgraph through depth-first search, and calculate the association strength of entity pairs through the resource transfer amount to obtain the entity relationship credibility score;

[0011] Based on the urban knowledge graph, evaluate the importance of path information, generate a query vector; calculate the weights of the additive scores between the query vector and the path vectors, combine the path vectors, and calculate the path representation vector; concatenate the path representation vector with the relationships to obtain a comprehensive vector; based on the comprehensive vector and the total number of all paths, obtain the path credibility score;

[0012] Regard the relationships in the triples sharing the same entity as interconnected nodes, and connect them to construct a relationship hypergraph; reconstruct the triples in the urban knowledge graph from the perspective of entity types to obtain a type hypergraph; quantify the semantic consistency through the relationship hypergraph and the type hypergraph to generate a type-structure credibility score;

[0013] Based on the entity relationship credibility score, the path credibility score, and the type-structure credibility score, obtain a comprehensive credibility score for evaluating the correctness of urban knowledge.

[0014] Furthermore, the method for calculating the association strength of entity pairs through the resource transfer amount to obtain the entity relationship credibility score includes: constructing a feature vector based on the resource amount from the head entity to the tail entity, the in-degree of the head node, the out-degree of the head node, the in-degree of the tail node, the out-degree of the tail node, and the number of relationships connecting the head entity and the tail entity; based on the feature vector, using a logistic regression model to obtain the entity relationship credibility score.

[0015] Furthermore, the method for evaluating the importance of path information and generating a query vector based on the urban knowledge graph includes: calculating the path semantic similarity based on the semantic similarity between the head entity and the tail entity in each triple of the path; calculating the average path distance based on the sum of the head entity vector and the relationship vector minus the tail entity vector in each triple of the path; obtaining the path transition probability based on the dot product between the head entity vector and the relationship vector plus the dot product between the relationship vector and the tail entity vector in each triple of the path; generating a query vector based on the path semantic similarity, the average path distance, and the path transition probability.

[0016] The present invention enhances the semantic information of entities through entity types and identifies potential anomalies by evaluating the consistency degree between types and the graph structure, thereby providing a basis for knowledge trustworthiness evaluation.

[0017] Furthermore, the method for obtaining the path credibility score based on the comprehensive vector and the total number of all paths includes: inputting the comprehensive vector into a two-layer perceptron network and combining it with the logarithmic transformation value of the total number of all paths to obtain the path credibility score.

[0018] Furthermore, the method for regarding the relationships in the triples sharing the same entity as interconnected nodes and connecting them to construct a relationship hypergraph includes:

[0019] Using the translation hypothesis of the TransE embedding model, extract the embedding representations of entities and relationships from the urban knowledge graph to obtain the embedding representation of the first anchor triple;

[0020] Adopt a multi-head attention mechanism to aggregate the embedding representations of neighbor triples related to the embedding representation of the first anchor triple, obtain attention scores, and perform normalization to obtain the attention coefficients for each pair of triples;

[0021] Based on the attention coefficients of each pair of triples, weighted aggregate the output results of the multi-head attention mechanism, and after processing by a non-linear activation function, obtain the updated embedding representation of the first anchor triple to construct a relational hypergraph.

[0022] Furthermore, the method of reconstructing the triples in the urban knowledge graph from the perspective of entity types to obtain a type hypergraph includes:

[0023] Based on the triples in the urban knowledge graph, construct a type matrix for the head entity and a type matrix for the tail entity;

[0024] By projecting the head entity embedding into the head entity type space, calculate the similarity metric between the projection and the head entity type matrix to obtain the head entity type embedding;

[0025] By projecting the tail entity embedding into the tail entity type space, calculate the similarity metric between the projection and the tail entity type matrix to obtain the tail entity type embedding;

[0026] Concatenate the head entity type embedding, the relationship embedding, and the tail entity type embedding to obtain the embedding representation of the second anchor triple;

[0027] Based on the embedding representation of the second anchor triple, aggregate and update the neighbor information to obtain the type embedding triple to construct a type hypergraph.

[0028] The present invention constructs a multi-level abnormal signal system by integrating the internal structure information of triples and external entity type information, and realizes fine-grained reliable evaluation of urban knowledge.

[0029] The second aspect of the present invention provides an urban knowledge correctness evaluation system based on multi-level abnormal signals.

[0030] An urban knowledge correctness evaluation system based on multi-level abnormal signals includes:

[0031] An entity relationship credibility evaluation module, which is configured to: based on the obtained urban knowledge graph, generate an entity subgraph through depth-first search, and calculate the association strength of entity pairs through the resource transfer amount to obtain an entity relationship credibility score;

[0032] A path credibility evaluation module, which is configured to: evaluate the importance of path information based on the urban knowledge graph, generate a query vector; calculate the weight of the additive score between the query vector and the path vector, combine the path vector, and calculate a path representation vector; splice the path representation vector with the relationship to obtain a comprehensive vector; obtain a path credibility score based on the comprehensive vector and the total number of all paths;

[0033] A type-structure credibility evaluation module, which is configured to: regard the relationships in the triples sharing the same entity as interconnected nodes and connect them to construct a relationship hypergraph; reconstruct the triples in the urban knowledge graph from the perspective of entity types to obtain a type hypergraph; quantify semantic consistency through the relationship hypergraph and the type hypergraph to generate a type-structure credibility score;

[0034] A comprehensive credibility evaluation module, which is configured to: obtain a comprehensive credibility score based on the entity relationship credibility score, the path credibility score, and the type-structure credibility score, and use it to evaluate the correctness of urban knowledge.

[0035] The third aspect of the present invention provides a computer device, which includes:

[0036] A processor, adapted to execute a computer program;

[0037] A computer-readable storage medium, in which a computer program is stored. When the computer program is executed by the processor, the steps in the method for evaluating the correctness of urban knowledge based on multi-level abnormal signals described in the first aspect above are implemented.

[0038] The fourth aspect of the present invention provides a computer-readable storage medium, which stores a computer program. The computer program is adapted to be loaded and executed by a processor to implement the steps in the method for evaluating the correctness of urban knowledge based on multi-level abnormal signals described in the first aspect above.

[0039] The fifth aspect of the present invention provides a computer program product or a computer program.

[0040] The present invention provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to implement the steps in the method for evaluating the correctness of urban knowledge based on multi-level abnormal signals described in the first aspect above.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] The existing knowledge trustworthy evaluation method mainly relies on the topological structure of the knowledge graph for analysis, which often leads to the evaluation results being difficult to cover multi-dimensional complex scenarios. For example, the existing methods ignore the hierarchical information in the path structure, which may result in inaccurate credibility evaluation of some complex triples. By constructing a multi-level abnormal signal system, the present invention can evaluate triples from multiple dimensions at the entity level, relationship level, and type-graph structure level, significantly improving the ability to identify the credibility of triples. This multi-level abnormal signal system enables the abnormal signals at each level to be independently evaluated and effectively integrated with the information at other levels, thereby achieving more accurate credibility evaluation.

[0043] Traditional methods fail to effectively utilize the hierarchical structure information in the path, and the path usually involves multiple relationships, and there are hierarchical semantic differences between the relationships. By adopting hierarchical path modeling and adaptive attention mechanism, the present invention can dynamically adjust the weights of the paths, thereby better capturing the hierarchical relationships in the path information. This approach not only improves the utilization efficiency of the path information but also enhances the robustness of the evaluation model, enabling it to handle complex knowledge graph structures. Especially in the case of dealing with multiple paths, the accuracy of the evaluation results has been significantly improved.

[0044] Existing methods often ignore the matching degree between the entity type and its connection relationship. By taking the consistency between the entity type and the graph structure as the core abnormal signal, the present invention can quantify the matching degree between the entity type and its adjacent relationships, and enhance this consistency evaluation through relationship hypergraph and type hypergraph. Specifically, using graph attention network to aggregate neighbor information and dynamically adjust weights, the evaluation system can make judgments based on the consistency between the entity type and the graph structure, thereby effectively identifying potentially incorrect triples and improving the accuracy of the credibility evaluation of the knowledge graph. Brief Description of the Drawings

[0045] The specification drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0046] Figure 1 is a flowchart of the method for evaluating the correctness of urban knowledge based on multi-level abnormal signals shown in the embodiments of the present invention;

[0047] Figure 2 is a detailed flowchart of the method for evaluating the correctness of urban knowledge based on multi-level abnormal signals shown in the embodiments of the present invention;

[0048] Figure 3 is a structural diagram of the system for evaluating the correctness of urban knowledge based on multi-level abnormal signals shown in the embodiments of the present invention;

[0049] Figure 4 It is the structural diagram of the computer device shown in the embodiments of the present invention. Detailed implementation manners

[0050] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0051] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanations of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0052] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0053] To facilitate the understanding of the technical solution of the present invention, some technical terms related to the present invention will be introduced below.

[0054] Knowledge Graph: A collection of knowledge elements and their connections described in a structured form, which is a visual representation of knowledge and constructs a representation of structured knowledge with entities and relationships at its core. It is usually composed of triples, and each triple is represented in the form of (head entity, relationship, tail entity).

[0055] Trustworthy: The ability to meet the expectations of stakeholders in a verifiable manner.

[0056] Depending on different backgrounds or industries, as well as specific products, services, technologies, data, techniques, and processes used, different trustworthy characteristics are applicable and need to be verified to ensure that the expectations of stakeholders are met.

[0057] Trustworthy characteristics include but are not limited to source reliability, correctness, timeliness, security and privacy, integrity, relevance, consistency, interpretability, accountability, controllability, stability, universality, etc.

[0058] Verification includes measurability and demonstrability through objective evidence, such as verifying the correctness of knowledge through methods such as comparison with actual situations, experimental verification, and invitation of expert reviews, which are measurable or demonstrable.

[0059] The evaluation of the degree of trustworthiness includes the identification and analysis of trustworthy characteristics, as well as the evaluation of the degree of trustworthiness of the analysis results based on relevant criteria, etc.

[0060] Based on this, as Figure 1 shown, this embodiment provides a method for evaluating the correctness of urban knowledge based on multi-level abnormal signals, including:

[0061] Based on the obtained urban knowledge graph, generate entity subgraphs through depth-first search, and calculate the association strength of entity pairs through resource transfer volume to obtain entity relationship credibility scores;

[0062] Based on the urban knowledge graph, evaluate the importance of path information to generate query vectors; calculate the weights of the additive scores between the query vectors and the path vectors, and combine the path vectors to calculate path representation vectors; splice the path representation vectors with the relationships to obtain comprehensive vectors; based on the comprehensive vectors and the total number of all paths, obtain path credibility scores;

[0063] Regard the relationships in the triples sharing the same entity as interconnected nodes and connect them to construct a relationship hypergraph; reconstruct the triples in the urban knowledge graph from the perspective of entity types to obtain a type hypergraph; quantify the semantic consistency through the relationship hypergraph and the type hypergraph to generate type-structure credibility scores;

[0064] Based on the entity relationship credibility scores, path credibility scores, and type-structure credibility scores, obtain comprehensive credibility scores for evaluating the correctness of urban knowledge.

[0065] The present invention proposes a method for evaluating the correctness of urban knowledge based on multi-level abnormal signals. Its detailed process is as Figure 2 shown. By deeply integrating internal structure information and external type information, a multi-level and multi-dimensional abnormal signal system is constructed, and the correctness in the trustworthy characteristics is mainly evaluated, realizing a fine-grained comprehensive evaluation of triple credibility. Specifically, the method includes: First, generate entity subgraphs based on depth-first search, and calculate the association strength of entity pairs through resource transfer volume to construct entity-level credibility evaluation; Second, innovatively propose a hierarchical path modeling method, and dynamically allocate path weights by combining an adaptive attention mechanism to achieve relationship-level credibility evaluation; Finally, for the first time, introduce the consistency between entity types and graph structures as the core abnormal signal into the evaluation system, and quantify the semantic consistency through the relationship hypergraph and the type hypergraph to generate type-graph structure-level credibility evaluation. By integrating entity association strength, path credibility, and type-structure credibility, the model can generate the final credibility of triples, providing reliable theoretical and method support for the correctness evaluation of knowledge graphs. This method not only makes full use of graph structure information but also innovatively uses type hypergraphs, significantly improving the accuracy and robustness of knowledge graph embedding, and providing new theoretical and method support for the trustworthy evaluation of knowledge graphs.

[0066] In some possible embodiments, based on the obtained urban knowledge graph, an entity subgraph is generated through depth-first search, and the association strength of entity pairs is calculated through the resource transfer amount to obtain the entity relationship credibility score. The specific implementation process includes:

[0067] The association strength between entities is an important indicator for measuring entity relationships. It can effectively reflect the potential connections between entity pairs in the knowledge graph. By calculating the association strength between entity pairs (h, t), the possibility of relationships between different entities can be evaluated. When the association strength of a pair of entities is low, the possibility of a relationship between them will also be significantly reduced, thus affecting the credibility of the triple composed of the entity pair. If there are multiple edges (relationships) between node e1 and node e5, this indicates a high association strength between (e1, e5), suggesting that there is likely a certain relationship between them. On the contrary, if node e2 and node e4 cannot be directly connected by a directed edge, this means a weak association strength between (e2, e4), thereby reducing the possibility of a relationship between them. Therefore, the greater the association strength between entity pairs, the higher the possibility of a relationship between them. The core idea of this process is to quantify the relationship strength between entities through resource flow, reflecting the potential connections between entity pairs in the knowledge graph. Specifically, the calculation of association strength mainly includes the following three steps: (1) Construct a directed graph through depth-first search (DFS) to capture the path information between entities; (2) Calculate the resource amounts of the head entity h and the tail entity t, comprehensively considering factors such as their out-degrees, in-degrees, and path weighting coefficients; (3) Based on the result of resource flow, calculate the association strength EA(h, t) of the entity pair (h, t). This method can not only effectively quantify the relationship strength between entities but also provide an important basis for the trustworthy evaluation of the knowledge graph.

[0068] Inspired by the ResourceRank algorithm, the present invention calculates the total amount of resources received by the tail entity t by simulating the resource flow process in the knowledge graph, thereby quantifying the association degree between entity pairs (h, t). The calculation process of the algorithm is divided into the following steps: First, initialize the resource allocation, allocate the initial resource amount to the head entity h, and initialize the resource amounts of other entities to 0; Second, simulate the resource flow process, and calculate the resource amount of each entity by weighted summation of the contributions of the predecessor nodes. Specifically, for each node n, the calculation of its resource amount R(n) considers the contributions of all predecessor nodes m, where R(m) is the resource amount of the predecessor node m, and the set S contains all predecessor nodes of node n, N hrDenotes the number of relationship paths between h and t. The resource outflow of node m is determined by its out-degree OD(m) and is weighted by the number of paths associated with m. Finally, the total resource of node n is obtained by weighted summation of the resources of each predecessor node m in set S. This method can not only effectively capture the potential connections between entities, but also provide an important basis for the trustworthy evaluation of knowledge graphs:

[0069]

[0070]

[0071]

[0072] Each entity is regarded as a node in a graph, and its resource amount is calculated by a formula. In the formula, R(h|t) represents the resource amount from head entity h to tail entity t, R(h) is the resource amount of the head entity, and R(t) is the resource amount of the tail entity. The propagation of resources is carried out through all paths related to the tail entity until the resource distribution reaches a stable state. Assume that during the resource propagation process, each node randomly jumps with probability θ, and at this time the resources are evenly distributed to all N e entities, where N e is the number of entities. The (1 - θ) in the formula controls the proportion of resource flow and balances the influence from each adjacent node. The resource amount of an entity is calculated based on the out-degree of the dependent path, the weighting coefficient, and the resource amounts of the connected nodes.

[0073] Therefore, this embodiment considers six indicators to calculate the association strength between two entities. (1) R(h|t) represents the resource amount from head entity h to tail entity t. (2) The in-degree ID(h) of the head node: the number of edges pointing to head entity h; (3) The out-degree OD(h) of the head node: the number of edges pointing from head entity h to other entities; (4) The in-degree ID(t) of the tail node: the number of edges pointing to tail entity t; (5) The out-degree OD(t) of the tail node: the number of edges pointing from tail entity t to other entities; (6) N ht represents the number of relationship paths connecting the two nodes h and t. These six indicators are constructed into a feature vector V, which can comprehensively describe the relationship between the two entities. To train the model and further evaluate the association strength between entity pairs, a logistic regression model is used to calculate the relationship probability between the two entities h and t. By calculating this concatenated feature vector through the logistic regression model, the relationship probability EA(h,t) is obtained as the credibility between entity pairs, and its calculation formula is:

[0074] EA(h,t) = σ(WV + b) (3)

[0075] Among them, σ is the Sigmoid activation function, W is the weight vector, b is the bias term, and V is the concatenated feature vector. The Sigmoid activation function maps the output value to the range [0, 1], representing the probability that there is a relationship between two entities.

[0076] In some possible embodiments, based on the urban knowledge graph, the importance of path information is evaluated to generate a query vector; the weight of the additive score between the query vector and the path vector is calculated, and combined with the path vector, the path representation vector is calculated; the path representation vector is concatenated with the relationship to obtain a comprehensive vector; based on the comprehensive vector and the total number of all paths, the path credibility score is obtained; the specific implementation process includes:

[0077] In the knowledge graph, the paths between entities usually follow a certain hierarchical structure. These paths are composed of multiple relationships, and the information carried by each path is multi-level. Based on the translation hypothesis, it can be verified whether each triple conforms to the condition, where r represents the relationship between the head entity h and the tail entity t. Secondly, through the hierarchical path information, the multi-step relationship paths from position C to regions A and B can be analyzed. If there are more reliable paths from C to B, and these paths are semantically closer to the corresponding relationship, then the triple "C, located in, B" will have a higher credibility, thus more accurately judging that C is more likely to be located in B rather than A. Therefore, the hierarchical structure of the paths provides valuable fine-grained information for the relationships between entity pairs in the knowledge graph, significantly improving the accuracy of knowledge graph credibility evaluation.

[0078] First, the TransE model is used to utilize the local hierarchical information, that is, the information within each independent path. Specifically, the method of TransE is followed to learn the embedding representations of entities and relationships. The core assumption of TransE is that for a triple (h, r, t), the sum of the vectors of the head entity h and the relationship r should be close to the vector of the tail entity t. The representation of each individual path is obtained through the TransE model. Since there may be multiple paths between the head entity h and the tail entity t, the global information, that is, all paths connecting h and t, is further utilized e1 is the head entity, e l+1 is the tail entity, and l is the path length. Since different paths have different importance in representing the relationship information between the head and tail entity pairs, the importance of each path is evaluated from the following three aspects.

[0079] (1) Path semantic similarity. The semantic similarity between each pair of entities in a path can be measured by calculating the dot product of their vector representations. To evaluate the semantic consistency of the entire path, the similarities of all entity pairs in the path are averaged. Specifically, for each triple in the path, first calculate the dot product between the head entity and the tail entity (i.e., their semantic similarity), then sum up the entity pair similarities of all triples in the path and divide by the path length to obtain the average semantic similarity of the path. This average value reflects the overall semantic tightness of the entity pairs in the path. A higher average similarity usually means that the entities in the path are more relevant and semantically more consistent, thus potentially having a greater positive impact on the effectiveness of the task and the importance of the path.

[0080]

[0081] where l is the path length, e j is the head entity, e j+1 is the tail entity, PS(h,p i ,t) represents the path semantic similarity.

[0082] (2) Average path distance. According to the translation hypothesis of the TransE model, the vector relationships between the head entity, relation, and tail entity in each triple can be used to calculate the distance of the path. Specifically, for each triple in the path, the distance is measured by calculating the sum of the head entity vector and the relation vector minus the tail entity vector. This distance can be represented by the L2 norm of the vector. By summing up the distance values of all triples in the path and dividing by the length of the path, the average distance of the path is obtained.

[0083]

[0084] where l is the path length, e j is the head entity, e j+1 is the tail entity, r j is the relation, AD(h,p i ,t) represents the average distance.

[0085] (3) Path transition probability. During the path transition process, the importance of the path can be evaluated by calculating the transition probability of each triple in the path. Specifically, for each triple in the path, the credibility of the path transition is measured by calculating the dot product between the head entity vector and the relation vector, and then adding the dot product between the relation vector and the tail entity vector. This value reflects the transition probability of reaching the tail entity from the head entity through the relation. The total transition probability of the path is obtained by summing up the transition probabilities of all triples in the path and dividing by the length of the path, thus obtaining the average transition probability of the path. A higher path transition probability indicates that the relations in the path are more credible.

[0086]

[0087] Among them, l is the path length, e j is the head entity, e j+1 is the tail entity, r j is the relationship, and PT(h, p i , t) represents the path transition probability.

[0088] According to the three important factors of the path - path semantic similarity (PS), average path distance (AD), and path transition probability (PT), the importance of the path can be comprehensively evaluated. Specifically, these three factors are concatenated into a three - dimensional vector and input into a two - layer multi - layer perceptron (MLP) network to generate a query vector eq i :

[0089] eq i = MLP([PS(h, p i , t), AD(h, p i , t), PT(h, p i , t)]) (7)

[0090] Then, an additive attention mechanism is adopted to map the path vector and the query vector into a shared latent space respectively. The similarity score between them is calculated through a non - linear activation function, and then the paths are ranked according to the scores to obtain the top - ten paths, and these scores are normalized to obtain the attention weights of each path. Specifically, the additive score S(p i , eq i ) is calculated as follows:

[0091]

[0092] Among them, eq i is the query vector, p i is the i - th path vector, W eq and W p are the mapping matrices for the query vector and the path vector, b is the bias term, and W a is the weight vector in the additive attention mechanism. Based on the additive score, the attention weight of each path is calculated:

[0093]

[0094]

[0095] Among them, a iis the attention weight of the i-th path, representing the relative importance of the path, and N is the number of the top ten paths ranked by scores. Finally, the path vectors are weighted and averaged according to the attention weights to generate a unified path representation vector:

[0096]

[0097] Finally, the unified path vector P can be concatenated with the embedding vector of the relation r to form a comprehensive vector z, which is then input into a two-layer multi-layer perceptron (MLP) network, combined with the logarithmic transformation value of the total number of paths N all to obtain the confidence score of the triple.

[0098] HP(h, r, t) = σ(MLP(z) + β · log(N all + 1)) (11)

[0099] where σ represents the Sigmoid activation function, and β is an adjustable hyperparameter used to control the contribution weight of the number of paths to the confidence score. Finally, HP(h, r, t) represents the confidence score of the triple, which reflects the tightness between the path and the relation and its reliability in the knowledge graph.

[0100] The present invention proposes a hierarchical path modeling method, proposes three path importance factors and combines an adaptive attention mechanism to dynamically allocate path weights, improving the utilization efficiency and evaluation accuracy of path information.

[0101] In some possible embodiments, the relations in the triples sharing the same entity are regarded as interconnected nodes and connected to construct a relation hypergraph; the triples in the urban knowledge graph are reconstructed from the perspective of entity types to obtain a type hypergraph; the semantic consistency is quantified through the relation hypergraph and the type hypergraph to generate a type-structure confidence score; the specific implementation process includes:

[0102] From the perspective of the graph structure, the triples with the same entity are usually semantically closely related. The knowledge graph (KG) can be analogized to a social network, where each triple represents an independent individual, and the "agreement degree" of adjacent triples with respect to the target triple reflects whether it can be naturally integrated into the overall knowledge system. Research shows that there is a high degree of dependence between the entity types in the knowledge graph and the graph structure. Existing research on knowledge graph embedding shows that the learned entity embeddings can form obvious clusters according to entity types in the embedding space, indicating that semantically related entities are often connected by specific relations. Therefore, a basic prerequisite for a triple (head entity, relation, tail entity) to hold is that its entity types must match the relation. If the entities of a certain triple are significantly different in type from neighboring entities, then this triple is more likely to contain inaccurate or noisy information.

[0103] Based on the above observations, the present invention constructs two three - layer hypergraphs, namely, the relation hypergraph and the type hypergraph, to represent the topological relationship between the original knowledge graph (KG) and its type information, and calculates its credibility based on the type - structure dependency relationship of triples. Specifically, each relation triple is regarded as a node, thereby transforming the target knowledge graph into a new structure graph. When constructing the relation hypergraph, a relation - based construction method is adopted, that is, if two triples share the same entity in the original knowledge graph, they are regarded as interconnected nodes and connected in the relation hypergraph. Such a three - layer structure transformation method effectively removes low - level information noise while retaining the original graph structure. In addition, to further characterize the influence of entity types, the present invention constructs a type hypergraph to reconstruct the knowledge graph from the perspective of types. The specific approach is to take each triple in the original knowledge graph as an independent instance, and by learning the features of the head entity type and the tail entity type, replace the original head entity and tail entity with the corresponding entity types, thereby re - defining and constructing the semantic representation of the triple. This method can better capture the consistency between entity types and the graph structure, providing richer semantic information for the credibility evaluation of the knowledge graph. Therefore, in normal triples, rich type information can be extracted using the type hypergraph to reconstruct the semantic representation learned in the relation hypergraph. Based on this characteristic, the present invention measures the consistency of the learned representations from two perspectives and uses it as an effective detection signal to evaluate the credibility of each triple in the original knowledge graph.

[0104] (1) Relation hypergraph. Using the translation hypothesis of the TransE embedding model, extract the embedding representations of entities and relations (e h , e r , e t ) from the knowledge graph; then, integrate these embedding vectors to construct a unified triple - level representation, so as to better capture the semantic information of the triples and the relationship structure between them. Specifically, the embedding representation of a triple is generated by the following formula:

[0105] t i = f concat (e h , r r , r t ) (12)

[0107] where f concat represents concatenating the head entity embedding, relation embedding, and tail entity embedding in the triple. Through this concatenation operation, the generated triple embedding t i can effectively capture the local relationship structure of the input triple. Therefore, these embeddings are used as the initial embeddings of each node in the relation hypergraph.

[0108] Rich contextual information among triple communities is crucial for identifying anonymous triples. To model this global contextual information, a graph neural network architecture based on the multi-head attention mechanism is adopted. By selectively aggregating neighbor triple information, the embedding representation of triples is updated. Specifically, given an anchor triple t in the relational hypergraph i ∈R d , its embedding representation is updated by attention aggregation of features in its neighborhood (e.g., {t1, t2, … t m}, where t m ∈R d ).

[0109] To enhance the expressive power of triple embeddings, a two-layer graph attention network is adopted, and the multi-head attention mechanism is used to aggregate neighbor triple features related to the anchor triple. Let H be the number of attention heads, and each attention head k calculates its corresponding attention score The formula is as follows:

[0110]

[0111] where f att represents the scoring function of the attention mechanism, is the learnable transformation matrix of the k-th attention head, and d ′ = d / H is the single-head dimension. Then, for the attention coefficients of each pair of triples, they are normalized by the softmax function:

[0112]

[0113] where, represents the normalized attention coefficient.

[0114] For each head k, the neighbor triples are weighted and summed using the normalized weights to obtain the aggregated representation of each head:

[0115]

[0116] where, is the output representation of the k-th head. The outputs of all heads are concatenated and mapped to the final output space through a learnable linear transformation matrix :

[0117]

[0118] where, z i is the updated embedding representation of the anchor triple t i , and f concatDenotes concatenation. Finally, to enhance the expressive power of triple embeddings, the final embedding representation can be further processed through the non-linear activation function ReLU:

[0119] t i ′ = σ(z i ) (17)

[0121] where t i ′ is the finally updated triple embedding representation.

[0122] (2) Type hypergraph. Currently, most knowledge graph embedding methods mainly focus on learning the internal relationship structure between entities in triples. However, the semantic information contained in entity types plays a crucial role in characterizing entity features and improving the quality of knowledge graph embeddings. In a knowledge graph, entity types usually show an uneven distribution. Specifically, entities usually associate with different type sets according to their roles and contexts in different triples, and even the roles played by the same entity in multiple triples may be different. Therefore, how to select the type that best conforms to the current entity's semantics from the types of the head entity or the tail entity becomes the core issue in reconstructing triple semantics. To address this challenge, the present invention designs a relationship-driven mechanism that, through the guidance of relationship information, intelligently selects the most relevant entity types, thereby ensuring that the core features of entities can be more precisely captured during the semantic construction process.

[0123] Given a triple (h, r, t), to extract richer semantic information from the key types of entities h and t, first concatenate the type embeddings of each entity into a matrix. Use τ to represent the entity type set, and define the entity type of entity h in the form of {v h,1 ,τ h,2 ,…τ h,i}, where τ h,i ∈τ is the i-th entity type, and |τ h | is the total number of entity types regarding entity h. The type matrices of the head entity and the tail entity are respectively Then project the entity embedding into the entity type space, and then calculate the similarity measure between this projection and the entity type embedding to obtain the entity type embedding E τ [h, i]. These type embeddings E τ [h, i] will be passed into the attention module together with the relationship embedding e r to generate the comprehensive representation of the entity types of h and t Taking entity h as an example, first calculate the interaction between each type and the relationship:

[0124] att h,i = f att(r, E τ [h, i])

[0125]

[0126] where f att is a neural network module for feature transformation, aiming to capture the correlation between entity types and relationships and generate corresponding attention weights is the normalized attention weight of type τ h,i . Next, the final entity type representation is calculated by taking the weighted sum of all type embeddings :

[0127]

[0128] For the tail entity t, its comprehensive representation is calculated in the same way. Subsequently, the type-based triple embedding y i can be calculated by concatenating the head entity type embedding the relation embedding e r and the tail entity type embedding :

[0129]

[0130] In the type hypergraph, given an anchor triple Following the global perspective of the relation hypergraph, the embedding representation is updated by attending to the neighbor features of y i (e.g., where {y1, y2, …, y m}, where ). Specifically, a two-layer graph attention network is used for calculation:

[0131]

[0132] where x i ′ is the type embedding triple after aggregating neighbor information and updating, represents the attention weight, W is the weight matrix, and σ is the activation function. In this way, the type information of neighbors can be aggregated to update the embedding representation of the anchor point.

[0133] Type hypergraphs are used to characterize the distribution of entity types, while relational hypergraphs model the correlations between relationships. Since there is a close connection between entity types and the topological structure of a knowledge graph, entities of specific types tend to be connected through specific relationships. Therefore, in normal triples, rich type information can be extracted using type hypergraphs to reconstruct the semantic representations learned in relational hypergraphs. Based on this property, by measuring the consistency of the learned representations from both perspectives, it is used as an effective detection signal to evaluate the credibility of each triple in the original knowledge graph.

[0134] TC(h,r,t) = sim(t′ i ,x′ i ) (23)

[0135] Among them, TC(h,r,t) represents the credibility of each triple in the original knowledge graph, and sim represents the consistency processing.

[0136] The present invention aggregates neighbor information through a multi-head attention mechanism to construct a relational hypergraph, and at the same time uses a graph attention network to enhance the matching degree between entity types and relationships, further aggregates neighbor information to construct a type hypergraph, thereby quantifying the consistency between entity types and the graph structure, and finally improving the accuracy of triple credibility evaluation.

[0137] In some possible embodiments, based on the entity-relationship credibility score, the path credibility score, and the type-structure credibility score, a comprehensive credibility score is obtained for evaluating the correctness of urban knowledge; the specific implementation process includes:

[0138] Finally, the comprehensive credibility score of each triple is calculated based on the extracted features, and it is defined as a scoring function for evaluating the correctness in the trustworthy characteristics of the triple. Specifically, the credibility of the triple is measured by three core features: the entity-relationship score, the path score, and the type-structure score. These features respectively reveal the relationship strength between entities, the path information between entities, and the consistency between entity types and the structure, providing multi-dimensional verification for the correctness of the triple. During the calculation process, weight factors λ1 and λ2 are applied to the path score HP(h,r,t) and the type-structure score TC(h,r,t) to adjust their contributions in the final score. Subsequently, all weighted feature values are normalized through the Sigmoid activation function, thereby calculating a credibility score T(h,r,t) between 0 and 1. This score reflects the credibility of the triple. The higher the score, the higher the correctness and stronger the credibility of the triple. Through this method, a quantitative evaluation of the triple's correctness is achieved, providing reliable theoretical and practical support for the trustworthiness evaluation of the knowledge graph.

[0139] T(h,r,t) = σ((EA(h,t) + λ1·HP(h,r,t) + λ2TC(h,r,t))) (24)

[0140] The present invention realizes the multi - dimensional and fine - grained evaluation of the credibility of knowledge graph triples through methods such as the deep fusion of triple internal structure information and external type information, hierarchical path modeling, and type - graph structure consistency evaluation, and focuses on evaluating the correctness in trustworthy features. Specifically, it includes: constructing a multi - level abnormal signal system, dynamically allocating path weights using hierarchical path modeling and adaptive attention mechanisms, and quantifying the consistency between entity types and graph structures through relational hypergraphs and type hypergraphs. In hierarchical path modeling, the semantic hierarchical structure in the path is captured by hierarchical encoding and attention mechanisms to dynamically weight path information, improving the utilization efficiency of path information; in type - graph structure consistency evaluation, the relevance between relationships and the distribution of entity types are respectively modeled through relational hypergraphs and type hypergraphs, the most relevant entity types are selected using a relationship - driven attention mechanism, and neighbor information is aggregated by combining graph attention networks to quantify the consistency between entity types and graph structures. These methods together provide a systematic solution for the trustworthy evaluation of urban knowledge.

[0141] As described above in combination with Figure 1 、 Figure 2 a detailed introduction to the method for evaluating the correctness of urban knowledge based on multi - level abnormal signals provided by the embodiments of the present invention has been given. Next, the system for evaluating the correctness of urban knowledge based on multi - level abnormal signals provided by the embodiments of the present invention will be introduced in combination with the accompanying drawings.

[0142] Figure 3 is a schematic structural diagram of the system for evaluating the correctness of urban knowledge based on multi - level abnormal signals shown in the embodiments of the present invention. Referring to Figure 3 the system of the present invention includes:

[0143] A system for evaluating the correctness of urban knowledge based on multi - level abnormal signals includes:

[0144] An entity - relationship credibility evaluation module, which is configured to: based on the obtained urban knowledge graph, generate an entity sub - graph through depth - first search, and calculate the association strength of entity pairs through resource transfer amount to obtain an entity - relationship credibility score;

[0145] A path credibility evaluation module, which is configured to: based on the urban knowledge graph, evaluate the importance of path information to generate a query vector; calculate the weight of the additive score between the query vector and the path vector, combine the path vector, and calculate a path representation vector; splice the path representation vector with the relationship to obtain a comprehensive vector; based on the comprehensive vector and the total number of all paths, obtain a path credibility score;

[0146] A type-structure credibility evaluation module, which is configured to: regard the relationships in triples sharing the same entity as interconnected nodes, connect them, and construct a relationship hypergraph; reconstruct the triples in the urban knowledge graph from the perspective of entity types to obtain a type hypergraph; quantify the semantic consistency through the relationship hypergraph and the type hypergraph, and generate a type-structure credibility score.

[0147] A comprehensive credibility evaluation module, which is configured to: obtain a comprehensive credibility score based on the entity relationship credibility score, the path credibility score, and the type-structure credibility score, and use it to evaluate the correctness of urban knowledge.

[0148] In some embodiments, the entity relationship credibility evaluation module is specifically configured to: construct a feature vector based on the resource amount from the head entity to the tail entity, the in-degree of the head node, the out-degree of the head node, the in-degree of the tail node, the out-degree of the tail node, and the number of relationships connecting the head entity and the tail entity; and obtain the entity relationship credibility score by using a logistic regression model based on the feature vector.

[0149] In some embodiments, the path credibility evaluation module is specifically configured to: calculate the path semantic similarity based on the semantic similarity between the head entity and the tail entity in each triple in the path; calculate the average path distance based on the sum of the head entity vector and the relationship vector minus the tail entity vector in each triple in the path; obtain the path transition probability based on the dot product between the head entity vector and the relationship vector plus the dot product between the relationship vector and the tail entity vector in each triple in the path; and generate a query vector based on the path semantic similarity, the average path distance, and the path transition probability.

[0150] The present invention enhances the semantic information of entities through entity types, and identifies potential anomalies by evaluating the consistency degree between the type and the graph structure, thereby providing a basis for knowledge trustworthiness evaluation.

[0151] In some embodiments, the path credibility evaluation module is further specifically configured to: input the comprehensive vector into a two-layer perceptron network, and combine it with the logarithmic transformation value of the total number of all paths to obtain the path credibility score.

[0152] In some embodiments, the type-structure credibility assessment module is specifically configured as follows: utilizing the translation hypothesis of the TransE embedding model to extract the embedding representations of entities and relationships from the urban knowledge graph to obtain the embedding representation of the first anchor triple; using a multi-head attention mechanism to aggregate the embedding representations of neighbor triplets related to the embedding representation of the first anchor triple to obtain an attention score, and normalize it to obtain the attention coefficient of each pair of triplets; based on the attention coefficient of each pair of triplets, weightedly aggregate the output results of the multi-head attention mechanism, and process them through a nonlinear activation function to obtain the updated embedding representation of the first anchor triple to construct a relationship hypergraph.

[0153] In some embodiments, the type-structure credibility assessment module is further configured to: construct a type matrix of the head entity and a type matrix of the tail entity based on the triples in the urban knowledge graph; project the head entity embedding into the head entity type space, calculate the similarity measure between the projection and the head entity type matrix, and obtain the head entity type embedding; project the tail entity embedding into the tail entity type space, calculate the similarity measure between the projection and the tail entity type matrix, and obtain the tail entity type embedding; concatenate the head entity type embedding, the relationship embedding, and the tail entity type embedding to obtain the embedding representation of the second anchor triple; based on the embedding representation of the second anchor triple, aggregate and update the neighbor information to obtain the type embedding triple to construct a type hypergraph.

[0154] This paper proposes a hierarchical path modeling method, proposes three path importance factors and combines them with an adaptive attention mechanism to dynamically allocate path weights, thereby improving the utilization efficiency and evaluation accuracy of path information.

[0155] The present invention aggregates neighbor information through a multi-head attention mechanism to construct a relationship hypergraph. At the same time, it uses a graph attention network to enhance the matching between entity types and relationships, further aggregates neighbor information, and constructs a type hypergraph, thereby quantifying the consistency between entity types and graph structures, and ultimately improving the accuracy of credibility assessment of triples.

[0156] The present invention enhances the semantic information of entities through entity types and identifies potential anomalies by evaluating the consistency between types and graph structures, thereby providing a basis for knowledge trustworthiness evaluation.

[0157] According to an embodiment of the present invention, the city knowledge correctness evaluation system based on multi-level abnormal signals can correspond to the method described in the embodiment of the present invention, and the above and other operations and / or functions of each module of the city knowledge correctness evaluation system based on multi-level abnormal signals are respectively to achieve Figure 1 、 Figure 2 For the sake of brevity, the corresponding processes of each method in are not repeated here.

[0158] See Figure 4 The structural diagram of the computer device shown. The computer device includes a processor, a communication interface, and a computer-readable storage medium. Among them, the processor, the communication interface, and the computer-readable storage medium can be connected through a bus or other means. Among them, the communication interface is used to receive and send data. The computer-readable storage medium can be stored in the memory of the computer device. The computer-readable storage medium is used to store computer programs. The computer programs include program instructions. The processor is used to execute the program instructions stored in the computer-readable storage medium. The processor (or CPU (Central Processing Unit, central processor)) is the computing core and control core of the computer device. It is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions to implement the corresponding steps in the embodiment of the method for evaluating the correctness of urban knowledge based on multi-level abnormal signals.

[0159] This embodiment provides a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in the computer device and is used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and, of course, the extended storage medium supported by the computer device. The computer-readable storage medium provides a storage space, and this storage space stores the processing system of the computer device.

[0160] Moreover, one or more instructions suitable for being loaded and executed by the processor are also stored in this storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory; optionally, it can also be at least one computer-readable storage medium located far from the aforementioned processor.

[0161] In one embodiment, one or more instructions are stored in the computer-readable storage medium; the processor loads and executes one or more instructions stored in the computer-readable storage medium to implement the corresponding steps in the embodiment of the method for evaluating the correctness of urban knowledge based on multi-level abnormal signals described above.

[0162] This embodiment provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and these computer instructions are stored in the computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the corresponding steps in the embodiment of the method for evaluating the correctness of urban knowledge based on multi-level abnormal signals described above.

[0163] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories and optical memories, etc.) containing computer-usable program code.

[0164] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or the combination of blocks.

[0165] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or the combination of blocks.

[0166] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or the combination of blocks.

[0167] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0168] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for evaluating the correctness of urban knowledge based on multi-level abnormal signals, characterized in that Including: Based on the obtained urban knowledge graph, generate an entity subgraph through depth-first search, calculate the association strength of entity pairs through the resource transfer amount, and obtain the entity relationship credibility score; Based on the urban knowledge graph, evaluate the importance of path information and generate a query vector; Calculate the weight of the additive score between the query vector and the path vector, combine the path vector, and calculate the path representation vector; Concatenate the path representation vector with the relationship to obtain a comprehensive vector; Based on the comprehensive vector and the total number of all paths, obtain the path credibility score; Regard the relationships in the triples sharing the same entity as interconnected nodes and connect them to construct a relationship hypergraph; Reconstruct the triples in the urban knowledge graph from the perspective of entity types to obtain a type hypergraph; Quantify the semantic consistency through the relationship hypergraph and the type hypergraph to generate a type-structure credibility score; Based on the entity relationship credibility score, the path credibility score, and the type-structure credibility score, obtain a comprehensive credibility score for evaluating the correctness of urban knowledge.

2. The method for evaluating the correctness of urban knowledge based on multi-level abnormal signals according to claim 1, wherein The method for calculating the association strength of entity pairs through the resource transfer amount and obtaining the entity relationship credibility score includes: constructing a feature vector based on the resource amount from the head entity to the tail entity, the in-degree of the head node, the out-degree of the head node, the in-degree of the tail node, the out-degree of the tail node, and the number of relationships connecting the head entity and the tail entity; Based on the feature vector, using a logistic regression model to obtain the entity relationship credibility score.

3. The method for evaluating the correctness of urban knowledge based on multi-level abnormal signals according to claim 1, wherein The method for evaluating the importance of path information based on the urban knowledge graph and generating a query vector includes: calculating the path semantic similarity based on the semantic similarity between the head entity and the tail entity in each triple of the path; calculating the average path distance based on the sum of the head entity vector and the relationship vector in each triple of the path minus the tail entity vector; calculating the path transition probability based on the dot product between the head entity vector and the relationship vector in each triple of the path plus the dot product between the relationship vector and the tail entity vector; generating a query vector based on the path semantic similarity, the average path distance, and the path transition probability.

4. The method for evaluating the correctness of urban knowledge based on multi-level abnormal signals according to claim 1, wherein The method for obtaining the path credibility score based on the comprehensive vector and the total number of all paths includes: inputting the comprehensive vector into a two-layer perceptron network and combining the logarithmic transformation value of the total number of all paths to obtain the path credibility score.

5. The method for evaluating the correctness of urban knowledge based on multi-level abnormal signals according to claim 1, wherein The method for regarding the relationships in the triples sharing the same entity as interconnected nodes and connecting them to construct a relationship hypergraph includes: Using the translation hypothesis of the TransE embedding model, extract the embedding representations of entities and relationships from the urban knowledge graph to obtain the embedding representation of the first anchor triple; Adopt a multi-head attention mechanism to aggregate the embedding representations of neighbor triples related to the embedding representation of the first anchor triple, obtain attention scores, and normalize them to obtain the attention coefficients for each pair of triples; Based on the attention coefficients for each pair of triples, weighted-aggregate the output results of the multi-head attention mechanism, and after processing by a non-linear activation function, obtain the updated embedding representation of the first anchor triple to construct a relationship hypergraph.

6. The method for evaluating the correctness of urban knowledge based on multi-level abnormal signals according to claim 1, characterized in that Reconstructing the triples in the urban knowledge graph from the perspective of entity types to obtain a type hypergraph; the method includes: Based on the triples in the urban knowledge graph, constructing a type matrix of the head entity and a type matrix of the tail entity; By projecting the head entity embedding into the head entity type space, calculating the similarity metric between the projection and the head entity type matrix, and obtaining the head entity type embedding; By projecting the tail entity embedding into the tail entity type space, calculating the similarity metric between the projection and the tail entity type matrix, and obtaining the tail entity type embedding; Concatenating the head entity type embedding, the relationship embedding, and the tail entity type embedding to obtain the embedding representation of the second anchor triple; Based on the embedding representation of the second anchor triple, aggregating neighbor information and updating to obtain a type embedding triple for constructing a type hypergraph.

7. An urban knowledge correctness evaluation system based on multi-level abnormal signals, characterized in that Including: An entity relationship credibility evaluation module, which is configured to: based on the obtained urban knowledge graph, generate an entity subgraph through depth-first search, and calculate the association strength of entity pairs through the resource transfer amount to obtain an entity relationship credibility score; A path credibility evaluation module, which is configured to: based on the urban knowledge graph, evaluate the importance of path information and generate a query vector; Calculating the weight of the additive score between the query vector and the path vector, combining the path vector, and calculating the path representation vector; concatenating the path representation vector with the relationship to obtain a comprehensive vector; based on the comprehensive vector and the total number of all paths, obtaining a path credibility score; A type-structure credibility evaluation module, which is configured to: regard the relationships in the triples sharing the same entity as interconnected nodes and connect them to construct a relationship hypergraph; reconstruct the triples in the urban knowledge graph from the perspective of entity types to obtain a type hypergraph; quantifying the semantic consistency through the relationship hypergraph and the type hypergraph to generate a type-structure credibility score; A comprehensive credibility evaluation module, which is configured to: based on the entity relationship credibility score, the path credibility score, and the type-structure credibility score, obtain a comprehensive credibility score for evaluating the correctness of urban knowledge.

8. A computer device, characterized in that A processor, adapted to execute a computer program; A computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by the processor, the steps in the method for evaluating the correctness of urban knowledge based on multi-level abnormal signals according to any one of claims 1-6 are implemented.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and the computer program is adapted to be loaded and executed by the processor to implement the steps in the method for evaluating the correctness of urban knowledge based on multi-level abnormal signals according to any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program, and when the computer program is executed by the processor, the steps in the method for evaluating the correctness of urban knowledge based on multi-level abnormal signals according to any one of claims 1-6 are implemented.

Images