An artificial intelligence-based financial knowledge graph construction method and system

Abstract

The application discloses a kind of financial knowledge graph construction method and system based on artificial intelligence, it is related to artificial intelligence data processing field.The method includes: multi-dimensional semantic analysis is carried out to heterogeneous financial data source, and structured semantic fragment is extracted;Financial entity perception unit is constructed, multiple granularity entity is identified and unique code is generated;Based on event cascade relationship and attachment structure, generate preliminary relationship graph, inject semantic translation label;Through semantic separation and label reconstruction mechanism, the relationship of graph is regularized, and the financial relationship network of semantic consistency is formed;Evolution increment iteration is executed in combination with new corpus, and node and edge set are dynamically updated;The iteration result is evaluated in semantic consistency and structural integrity, and the stable financial knowledge graph structure body is output.Through the introduction of multi-factor semantic analysis model, causal relationship modeling mechanism and graph evolution iteration strategy, the systematic improvement of financial knowledge graph in structure expression precision, semantic reasoning ability and dynamic adaptability is realized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence data processing, specifically to a method and system for constructing a financial knowledge graph based on artificial intelligence. Background Technology

[0002] With the continuous growth of financial data and its significantly enhanced heterogeneity and multi-source characteristics, traditional knowledge construction methods based on rule configuration and static extraction strategies are no longer sufficient to meet the high-precision modeling requirements of the dynamic evolution of financial entities, relationships, and semantics. In typical application scenarios such as financial regulation, risk identification, public opinion analysis, and investment decision-making, there is an urgent need to construct financial knowledge graphs with dynamic update capabilities, semantic consistency guarantees, and structural traceability features to achieve in-depth analysis and predictive support for financial event chains and entity evolution paths.

[0003] In existing technologies, most methods for constructing financial knowledge graphs suffer from the following technical bottlenecks: First, the semantic parsing granularity is coarse, failing to fully consider the heterogeneous characteristics of financial data in dimensions such as structural sparsity, label ambiguity, and semantic perturbation, resulting in limited accuracy in extracting basic entities from the graph. Second, symmetrical aggregation strategies are commonly used in entity recognition, failing to differentiate and model the asymmetric relationships between multiple financial attributes, affecting the distinguishability and accuracy of semantic recognition and entity assignment. Third, the relationship construction stage lacks causal relationship modeling based on event cascading patterns and upstream and downstream semantic dependency structures, making it difficult to effectively characterize the evolution triggering mechanism between financial entities. Fourth, during the graph evolution process, a dynamic evaluation mechanism based on node stability and global fusion degree is not introduced, leading to problems such as local mutations and structural drift in the graph evolution path, restricting the timeliness and controllability of the graph.

[0004] Therefore, there is an urgent need to propose a financial knowledge graph construction method that integrates structural perturbation modeling, semantic variation adjustment, and graph incremental evolution mechanism to improve the intelligence level and dynamic adaptability of the entire graph construction process, and achieve accurate modeling and high-quality knowledge generation of complex financial semantic networks. Summary of the Invention

[0005] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a method and system for constructing a financial knowledge graph based on artificial intelligence, so as to solve the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for constructing a financial knowledge graph based on artificial intelligence, comprising: We acquire raw data from heterogeneous financial data sources, and perform multidimensional semantic analysis on the data sources based on the source domain complexity factor, structural sparsity factor, and semantic perturbation potential of each data unit. We then extract structured semantic fragments and construct a semantic analysis dataset. Based on the semantic parsing dataset, a financial entity perception unit is constructed. By combining the asymmetric fusion weight of features and the attribute coupling suppression factor, an entity assignment mechanism for multiple financial semantic domains is established to identify multi-granularity entities and generate unique entity codes, thereby generating a set of financial entities. Based on a set of financial entities, an entity connection structure oriented towards relationship-induced triggering paths is constructed through an event cascade probability matrix and upstream and downstream dependency coefficients to generate a preliminary financial relationship graph. Semantic translation meta-labels are injected into the edge attributes of the financial relationship graph. By using semantic separation adjustment, the semantically inconsistent mapping relationships in the preliminary financial relationship graph are deeply normalized, and the entity label sparse tensor reconstruction mechanism is used to eliminate pseudo-related connections, forming a semantically normalized financial relationship network. Based on semantically regularized financial relationship networks and newly added financial corpora, the evolution trajectory of nodes and edge set expansion paths in the financial relationship graph structure are dynamically updated through evolutionary incremental iteration, continuously evolving a financial knowledge graph driven by local stability and global fusion, thus forming a financial knowledge graph with evolutionary capabilities. By using semantic robustness assessment factors and structural regression integrity indicators, the semantic consistency of the iterated financial knowledge graph is evaluated and the structural integrity is verified, generating a stable output financial knowledge graph structure.

[0007] The present invention is further configured such that constructing the semantic parsing dataset includes: The raw data from the acquired heterogeneous financial data sources are divided into data units according to data source, field type, and semantic content; Based on the source domain complexity factor of each data unit, the degree of hierarchical nesting and semantic fusion redundancy of the unit in the source structure are quantitatively identified. Based on the structural sparsity factor, a structural measure is performed on the field distribution completeness and local missing features between fields of this data unit. Based on semantic perturbation potential, perturbation analysis is performed on the degree of contextual ambiguity and semantic variation of text data fields; By combining the above factors, a multidimensional semantic parsing matrix is ​​constructed, and a parsing operation that combines semantic and structural dimensions is performed on each data unit. The structured semantic fragments with clear semantic boundaries obtained after parsing are extracted in a standardized manner and uniformly collected to form a semantic parsing dataset.

[0008] The present invention is further configured such that the source domain complexity factor is a structural feature index generated based on the field nesting level, label reuse frequency and naming perturbation degree in the data unit; The structural sparsity factor is a distribution sparsity index constructed based on the degree of field value range jumps, the proportion of field missing values, and the dispersion of field distribution. Semantic perturbation potential is a semantic heterogeneity index calculated by combining the magnitude of semantic variation of the field, the degree of semantic drift in the context, and the density of semantic interference items.

[0009] The present invention is further configured such that the financial entity sensing unit comprises: Based on the semantic labels and attribute fragments in the semantic parsing dataset, entity candidate units that meet the conditions for multidimensional semantic recognition are extracted; Combining feature asymmetric fusion weights, directional matching is performed on the semantic coupling degree and attribute fusion strength among candidate entity units; By combining attribute coupling suppression factors, redundant relationships caused by high-frequency attribute linkages are identified and weakened, thereby suppressing misidentified entity association paths. Establish an entity assignment mechanism for multiple financial semantic domains, and divide entity candidate units into granular categories based on semantic category mapping and attribute aggregation structure; By generating unique entity codes and combining the results of the assignment mechanism, a set of financial entities is constructed by generating unique and traceable coded identifiers for each entity.

[0010] The present invention is further configured such that the feature asymmetric fusion weight is constructed by establishing a weighting matrix based on attribute semantic density, fragment overlap, label activation index, lexical offset length and co-occurrence order difference, and using a fusion strategy with directional feature distribution; The attribute coupling suppression factor constructs a constraint matrix based on the polysemous strength of the attribute context, the cross-semantic attribution ratio, the degree of nonlinear interaction, and the attribution fuzzy boundary parameters. It encodes and limits the attribute linkage relationship by combining the asymmetric structural distribution parameters within the coupling dimension.

[0011] The present invention is further configured such that, based on the set of financial entities, an event cascading probability matrix and upstream and downstream dependency coefficients are constructed to characterize the event influence propagation relationship and semantic subordination structure between entities; Based on the event cascade probability matrix and upstream and downstream dependency coefficients, relationship triggering paths are established between entities to form an entity connection network with causal guidance structure, generating a preliminary financial relationship graph. Semantic translation meta-tags are injected into the edge attributes of the graph to express the semantic evolution direction, interaction mode, and structural role dependency type of entity relationships.

[0012] The present invention is further configured such that the formation of the semantically regular financial relationship network includes: By adjusting semantic separation, the semantically inconsistent mapping relationships in the preliminary financial relationship graph are normalized. The adjustment factor, which is constructed by structural connectivity parameters, semantic mapping residual parameters and label matching quantity, is used to identify the connection paths with semantic conflicts in the graph edge set. Based on the identified connection paths, a mapping adjustment matrix is ​​generated, and an entity label sparse tensor reconstruction model is constructed by combining the evolution state index of the entity label space to reconstruct the entity labels in the preliminary financial relationship graph and eliminate spurious connections. Based on the reconstructed entity label tensor and adjustment matrix, the edge set structure is updated, conflicting edge connections are removed, and a financial relationship network with semantic consistency and label structure stability is formed.

[0013] The present invention is further configured such that the continuously evolving financial knowledge graph driven by local stability and global fusion includes: Based on semantically regular financial relationship networks and newly added financial corpora, we construct evolutionary input units, establish node evolution driving factors and edge set expansion prediction tensors, and form graph evolution conditions with structural dynamism and semantic adaptability. By combining the local stability of nodes with the global fusion of the graph, the evolution triggering state of the graph structure is determined, the regions to be updated are selected, and incremental connection construction is performed. Under the constraint of evolution threshold, the node set and edge set in the graph structure are iteratively updated to generate a financial knowledge graph with semantic consistency, structural coherence and adaptive evolution capabilities.

[0014] The present invention is further configured such that the financial knowledge graph structure for generating stable output includes: The semantic consistency assessment and structural integrity verification of the evolved financial knowledge graph structure are performed. Semantic consistency assessment analyzes the degree of collaborative association between semantic tags in the graph based on semantic robustness assessment factors; Structural integrity verification measures the degree to which the node connectivity relationships of the graph are preserved before and after evolution based on the structural regression integrity index. By combining semantic robustness assessment factors and structural regression completeness indicators, the output stability of the current graph structure is determined, thus forming a stable financial knowledge graph structure.

[0015] This invention also provides an artificial intelligence-based financial knowledge graph construction system, the system comprising: Semantic parsing module: Acquires raw data from heterogeneous financial data sources, performs multidimensional semantic parsing on the data sources based on the source domain complexity factor, structural sparsity factor, and semantic perturbation potential energy of each data unit, extracts structured semantic fragments, and constructs a semantic parsing dataset; Financial Entity Recognition Module: Based on the semantic parsing dataset, a financial entity perception unit is constructed. By combining feature asymmetric fusion weights and attribute coupling suppression factors, an entity assignment mechanism for multiple financial semantic domains is established to identify multi-granularity entities and generate unique entity codes, thereby generating a set of financial entities. Relationship Connection Construction Module: Based on the set of financial entities, an entity connection structure oriented towards the relationship-induced triggering path is constructed through the event cascade probability matrix and upstream and downstream dependency coefficients, generating a preliminary financial relationship graph, and semantic translation meta-tags are injected into the edge attributes of the financial relationship graph; Semantic regularization module: It performs deep regularization on the semantically inconsistent mapping relationships in the preliminary financial relationship graph through semantic separation adjustment, and uses the entity label sparse tensor reconstruction mechanism to eliminate pseudo-related connections, forming a semantically regularized financial relationship network. Evolutionary Iteration Module: Based on the semantically regularized financial relationship network and newly added financial corpus, the module dynamically updates the node evolution trajectory and edge set expansion path in the financial relationship graph structure through incremental evolutionary iteration, continuously evolving the financial knowledge graph driven by local stability and global fusion, forming a financial knowledge graph with evolutionary capabilities. Consistency Assessment and Integrity Verification Module: Through semantic robustness assessment factors and structural regression integrity indicators, the module performs semantic consistency assessment and structural integrity verification on the financial knowledge graph after iteration, generating a stable output financial knowledge graph structure.

[0016] This invention provides a method and system for constructing a financial knowledge graph based on artificial intelligence. The method acquires raw data from heterogeneous financial data sources, performs multi-dimensional semantic analysis on the data sources based on the source domain complexity factor, structural sparsity factor, and semantic perturbation potential energy of each data unit, extracts structured semantic fragments, and constructs a semantic analysis dataset. Based on the semantic analysis dataset, a financial entity perception unit is constructed, and an entity assignment mechanism oriented towards multiple financial semantic domains is established by combining feature asymmetric fusion weights and attribute coupling suppression factors to identify multi-granularity entities and generate unique entity codes, thus generating a set of financial entities. Based on the set of financial entities, an entity connection structure oriented towards relationship-induced triggering paths is constructed through an event cascade probability matrix and upstream / downstream dependency coefficients, generating a preliminary financial relationship graph. The financial relationship graph is then further refined. Semantic translation meta-tags are injected into the edge attributes; semantic separation is used to deeply regularize the semantically inconsistent mapping relationships in the initial financial relationship graph, and a sparse tensor reconstruction mechanism based on entity tags is used to eliminate pseudo-related connections, forming a semantically regularized financial relationship network; based on the semantically regularized financial relationship network and newly added financial corpus, the evolutionary trajectory of nodes and the edge set expansion path in the financial relationship graph structure are dynamically updated through evolutionary incremental iteration, continuously evolving a financial knowledge graph driven by local stability and global fusion, forming a financial knowledge graph with evolutionary capabilities; semantic robustness assessment factors and structural regression completeness indicators are used to evaluate the semantic consistency and verify the structural completeness of the iterated financial knowledge graph, generating a stable output financial knowledge graph structure. The beneficial effects include: 1. Achieve deep semantic analysis of heterogeneous financial data: By constructing a multi-dimensional analytical model based on source domain complexity factor, structural sparsity factor and semantic perturbation potential, the ability to analyze the hidden semantic structure in multi-source data is effectively improved, and the accuracy of the structured semantic expression of the original data is enhanced. 2. Construct a financial relationship graph with causal triggering structure: Integrate the event cascade probability matrix and upstream and downstream dependency coefficients to characterize the semantic propagation path and relationship triggering chain between entities, forming a preliminary financial relationship graph with semantic translation meta-labels, and enhancing the graph's expressive power in semantic reasoning and causal modeling. 3. Supports dynamic evolution and incremental updates of the graph structure: Based on the evolutionary incremental iteration mechanism driven by local stability and global fusion, the node evolution trajectory and edge set expansion path are dynamically adapted and updated to ensure the adaptability and continuity of the graph when facing new corpus, and to build a financial knowledge graph with evolutionary capabilities.

[0017] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 A flowchart illustrating an artificial intelligence-based financial knowledge graph construction method is shown as an exemplary embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the structure of an artificial intelligence-based financial knowledge graph construction system, which is an exemplary embodiment of the present invention. Detailed Implementation

[0019] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0020] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0021] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0022] Example 1: A method for constructing a financial knowledge graph based on artificial intelligence, such as Figure 1 As shown, it includes: We acquire raw data from heterogeneous financial data sources, and perform multidimensional semantic analysis on the data sources based on the source domain complexity factor, structural sparsity factor, and semantic perturbation potential of each data unit. We then extract structured semantic fragments and construct a semantic analysis dataset. Based on the semantic parsing dataset, a financial entity perception unit is constructed. By combining the asymmetric fusion weight of features and the attribute coupling suppression factor, an entity assignment mechanism for multiple financial semantic domains is established to identify multi-granularity entities and generate unique entity codes, thereby generating a set of financial entities. Based on a set of financial entities, an entity connection structure oriented towards relationship-induced triggering paths is constructed through an event cascade probability matrix and upstream and downstream dependency coefficients to generate a preliminary financial relationship graph. Semantic translation meta-labels are injected into the edge attributes of the financial relationship graph. By using semantic separation adjustment, the semantically inconsistent mapping relationships in the preliminary financial relationship graph are deeply normalized, and the entity label sparse tensor reconstruction mechanism is used to eliminate pseudo-related connections, forming a semantically normalized financial relationship network. Based on semantically regularized financial relationship networks and newly added financial corpora, the evolution trajectory of nodes and edge set expansion paths in the financial relationship graph structure are dynamically updated through evolutionary incremental iteration, continuously evolving a financial knowledge graph driven by local stability and global fusion, thus forming a financial knowledge graph with evolutionary capabilities. By using semantic robustness assessment factors and structural regression integrity indicators, the semantic consistency of the iterated financial knowledge graph is evaluated and the structural integrity is verified, generating a stable output financial knowledge graph structure.

[0023] The present invention is further configured such that constructing the semantic parsing dataset includes: The raw data from the heterogeneous financial data sources are divided into data units according to data source, field type, and semantic content. Specifically, data sources include exchanges, corporate announcements, and social comments; field types include numerical, category, and text; and semantic content includes financial events, subject relationships, and behavioral indicators. Based on the source domain complexity factor of each data unit, the degree of hierarchical nesting and semantic fusion redundancy of that unit in the source structure are quantitatively identified; specifically, the source domain complexity factor... Measuring the semantic nesting depth, tag reuse, and naming stability of data units within the original source structure is a quantitative indicator of structural hierarchical complexity. Based on the structural sparsity factor, a structural measure is performed on the completeness of field distribution and the local missing features between fields for this data unit; specifically, the structural sparsity factor... Measuring the discontinuity, jump behavior, and concentration of missing data among data fields reflects the integrity and stability of structural information; Based on semantic perturbation potential, perturbation analysis is performed on the degree of contextual ambiguity and semantic variation of text-based data fields; specifically, semantic perturbation potential... Assessing the degree of semantic inconsistency, contextual ambiguity, and noise in text fields is a negative indicator of semantic stability. A multidimensional semantic parsing matrix is ​​constructed by combining the above factors, and a combined semantic and structural dimension parsing operation is performed on each data unit. Specifically, the multidimensional semantic parsing matrix is ​​a joint expression structure formed by summarizing the three factors, used to perform linked extraction of semantic and structural dimensions. By using the above three types of factors to construct the multidimensional semantic parsing matrix, linked parsing is performed on each data unit to identify structured fragments with stable semantic boundaries. The formula for constructing the multidimensional semantic parsing matrix is: ,in, The overall score of the analytical index matrix for the data unit. , , The weight coefficients of the three types of factors satisfy the following conditions: The matrix, determined through experimental optimization, is used to screen data units with clear semantic boundaries, rich structural information, and low semantic noise, serving as the basis for structured semantic fragment extraction. The structured semantic fragments with clear semantic boundaries obtained after parsing are extracted in a standardized manner and uniformly collected to form a semantic parsing dataset.

[0024] The present invention further specifies that the source domain complexity factor is a structural feature index generated based on the field nesting level, label reuse frequency, and naming perturbation degree in the data unit; specifically, the calculation formula for the source domain complexity factor is: ,in, This represents the total number of fields in the data unit. For fields Nesting level, take integer. For fields Tag reuse frequency indicates the number of times the field appears repeatedly in multiple contexts. For fields The degree of naming perturbation is calculated based on the normalized value of the edit distance of the field, ranging from [0.1, 1.0]. The source domain complexity factor is the cube root of the weights of nesting level, tag reuse and naming inconsistency as three different perspectives, reflecting the nonlinear coupling characteristics between multiple structural factors. The structural sparsity factor is a distribution sparsity index constructed based on the degree of abrupt changes in field value ranges, the proportion of missing fields, and the dispersion of field distribution; specifically, the formula for calculating the structural sparsity factor is: ,in, For fields The degree of range jump is calculated based on the ratio of the maximum and minimum range spans to the number of types, ranging from [1.0, 20.0]. For fields The missing percentage, ranging from [0.0, 1.0]. For fields The discreteness of the distribution; Semantic perturbation potential is a semantic heterogeneity index calculated by combining the magnitude of semantic variation of a field, the degree of semantic shift in the context, and the density of semantic interference items. Specifically, the formula for calculating semantic perturbation potential is as follows: ,in, Total number of text fields For fields The semantic variation range refers to the extent to which the semantic center of the same term shifts in different contexts. The degree of contextual semantic drift represents the change in semantic distribution within the context window in the historical corpus. The semantic interference density represents the proportion of invalid or ambiguous words per unit length. This refers to the field weight.

[0025] The present invention is further configured such that the financial entity sensing unit comprises: Based on the semantic labels and attribute fragments in the semantic parsing dataset, entity candidate units that meet the conditions for multidimensional semantic recognition are extracted; By combining feature asymmetric fusion weights, the semantic coupling degree and attribute fusion strength among candidate entity units are directionally matched; specifically, a feature asymmetric fusion weight matrix is ​​constructed. It characterizes the directional feature differences between candidate entities in the intersection of semantic structure and attributes, reflecting the asymmetric coupling in the fusion direction. By combining attribute coupling suppression factors, redundant relationships caused by high-frequency attribute linkages are identified and weakened, suppressing misidentified entity association paths. Specifically, an attribute coupling suppression mechanism is established to identify the interference of high-frequency co-occurring attributes on entity uniqueness, and entity recognition accuracy is improved by suppressing coupled redundant relationships. Attribute coupling suppression factors The strength of interference from attribute redundancy is measured; the higher the value, the stronger the suppression. Inconsistent mappings in complex semantic attribution are controlled through a combination of power-law and nonlinear methods. An entity assignment mechanism oriented towards multiple financial semantic domains is established, and entity candidate units are granularly divided according to semantic category mapping and attribute aggregation structure. Specifically, by combining semantic category and attribute aggregation patterns, candidate units are granularly divided and semantically labeled to achieve refined assignment across multiple semantic domains. The entity assignment confidence score is used to measure the rationality and uniqueness of a candidate unit assigned to a specific semantic domain, supporting granular division and classification boundary delineation. Its formula is as follows: ,in, Number the candidate units of the entity. For the first The number of attribute fields contained in the unit. For fields Category semantic fit For fields Semantic salience in context, As a property coupling inhibition factor, The system assigns confidence scores to entities to measure the rationality and uniqueness of a candidate unit in a specific semantic domain. It supports granular partitioning and classification boundary delineation, realizes a hierarchical assignment strategy oriented towards semantic domains, and supports multi-granularity entity aggregation and dynamic semantic domain mapping. By generating unique entity codes and combining the results of the assignment mechanism, a unique and traceable coded identifier is generated for each entity, constructing a set of financial entities. Specifically, based on factors such as structural nesting, semantic category, and contextual features, a unique and traceable coded identifier is constructed for each identified entity. The construction formula is as follows: ,in, For entities, Number the semantic domain in which the entity resides. For the depth of entity structure nesting, This is the combined value of entity semantic tags, representing the length of the set of semantic class tags after mapping. It serves as the location fingerprint for entity fragments, representing the concatenation of the starting position and the context window number, thereby enhancing the uniqueness, stability, and traceability of entity identifiers and providing strong support for the consistency management of node relationships in knowledge graphs.

[0026] The present invention is further configured such that the feature asymmetric fusion weights are constructed by establishing a weighting matrix based on attribute semantic density, fragment overlap, label activation index, lexical offset length, and co-occurrence order difference, and employing a fusion strategy with directional feature distribution; specifically, the construction formula for the feature asymmetric fusion weight matrix is ​​as follows: ,in, As candidate entity units, The semantic density ratio represents the ratio of the semantic coverage of words in two unit attribute fields. Tag activation difference represents the difference in the degree of activation of semantic tags in the context. The degree of overlap represents the percentage of intersection between attribute segments of two units. This is the lexical offset length, representing the difference in offset values ​​between the occurrence positions of the first field and its corresponding positions in the context. The co-occurrence order difference is the encoding difference based on the order of events. The above parameters are calibrated experimentally. Through the asymmetric feature fusion mechanism, the entity recognition's ability to perceive structural orientation and fusion dominance is enhanced, thereby improving the entity boundary discrimination ability. The attribute coupling suppression factor constructs a constraint matrix based on the polysemous strength of attribute context, semantic attribution crossover ratio, degree of nonlinear interaction, and attribution ambiguity boundary parameters. It encodes and limits attribute linkage relationships by combining asymmetric structural distribution parameters within the coupling dimension. Specifically, the calculation formula for the attribute coupling suppression factor is as follows: ,in, For attribute fields, For attributes The intensity of contextual polysemy, The semantic attribution crossover ratio represents the proportion of overlap between two attributes appearing in multiple semantic tags. The degree of non-linear interaction between attribute pairs represents the proportion of the two attributes appearing in an asymmetric semantic structure. The attribute fuzzy boundary parameter represents the uncertainty score of the semantic domain to which the attribute belongs. It is calibrated by experiments and introduces an attribute coupling suppression mechanism to effectively reduce the redundant path interference caused by high-frequency attribute linkage and reduce the entity mismatch rate.

[0027] The present invention is further configured such that, based on the set of financial entities, an event cascading probability matrix and upstream and downstream dependency coefficients are constructed to characterize the event influence propagation relationship and semantic subordination structure between entities; Based on the event cascade probability matrix and upstream / downstream dependency coefficients, relationship triggering paths are established between entities, forming an entity connection network with a causal guidance structure. A preliminary financial relationship graph is generated, and semantic translation meta-tags are injected into the edge attributes of the graph to express the semantic evolution direction, interaction mode, and structural role dependency type of entity relationships. Specifically, the formula for constructing the event cascade probability matrix is ​​as follows: ,in, , Assign a number to the financial entity. The event sequence coordination factor represents the entity. and The percentage of events participated in simultaneously Time-induced density, representing the event originating from an entity. spread to The inverse of the time window density, , Semantic redundancy of an entity represents the degree of label repetition in the context. The local path perturbation coefficient represents the magnitude of the ambiguity distribution at intermediate nodes in the event propagation path. , The entity semantic domain tension coefficient represents the degree of conflict between the semantic domain in which each entity resides and the event. The event triggering complexity represents the damping coefficient for cross-semantic domain diffusion between events. This represents a non-exponential linear symbolic mapping, that is, it does not use a real exponential function, but activates through the tension difference. Larger indicates a larger entity Triggering Entity The stronger the event cascading relationship, the more likely the upstream and downstream dependency coefficients are to be constructed using the following formula: ,in, , The depth of the entity structure represents the number of nesting levels. The cross-domain tensor strength for attribute mapping represents the norm of the tensor of overlapping entity attributes across semantic domains. This refers to the label structure deviation, indicating the degree of inconsistency in the order of entity label structures. , Semantic dependency centrality represents the centrality weight of an entity within its semantic domain. This coefficient characterizes the hierarchical and structural relationships between entities, reflecting their importance in the upstream and downstream positioning of the semantic path. A larger value indicates a stronger entity... More dependent on This constitutes a semantic subordinate relationship; the calculation logic for the relationship-induced trigger path weight is as follows: ,in, For semantic activation functions, ,in, The semantic path coherence score represents the distance from the path to the path. arrive Semantic tag matching degree on the path The structural role offset coefficient represents the entity. and The degree of semantic misalignment of the roles ultimately triggers path weights. The path weights are determined by the joint event induction capability and the degree of upstream and downstream subordination, while semantic and structural consistency adjustment factors are introduced to form directional path weights for subsequent relation graph edge weight injection; the edge construction attribute semantic translation meta-label constructor is also included. ,in, The strength of semantic evolution trend indicates the semantic label's evolution from... Towards Migration trend score The structural role dependency factor represents the stability score of role relationships in an event chain. This is a mapping function used to map numerical ranges to a set of semantic translation labels. These labels are used to assign semantic direction and structural role annotations to edge attributes in the graph, clarifying the evolution of relationships between entities and their structural attribution dependencies.

[0028] The present invention is further configured such that the formation of the semantically regular financial relationship network includes: By adjusting semantic separation, the semantically inconsistent mapping relationships in the preliminary financial relationship graph are normalized. The adjustment factor, which is constructed by structural connectivity parameters, semantic mapping residual parameters and label matching quantity, is used to identify the connection paths with semantic conflicts in the graph edge set. Based on the identified connection paths, a mapping adjustment matrix is ​​generated, and an entity label sparse tensor reconstruction model is constructed by combining the evolution state index of the entity label space to reconstruct the entity labels in the preliminary financial relationship graph and eliminate spurious connections. Based on the reconstructed entity label tensor and adjustment matrix, the edge set structure is updated, conflicting edge connections are eliminated, and a financial relationship network with semantic consistency and label structure stability is formed. Specifically, after the initial financial relationship graph is generated, some edge connections between entities exhibit semantic inconsistencies: that is, semantic label conflicts exist between entities on certain edge connections, such as the coexistence of "derived" and "independent"; pseudo-related connections: connections without actual semantic basis due to label ambiguity or structural overlap; label space evolution mismatch: inconsistencies in the evolution state of entity labels during graph updates or inter-period extrapolation. To eliminate the above problems and form a semantically regular financial relationship network, this invention proposes semantic separation adjustment: identifying edge paths with semantic conflicts; mapping adjustment matrix and label sparse tensor reconstruction model: reconstructing the structure of conflicting edges and their associated entity labels; edge set structure update: eliminating pseudo-related edges to form a stable and consistent structural network; constructing a semantic conflict adjustment factor and defining ternary relationship paths. The formula is: ,in, For entity pairs , Structural connectivity score, For the edge The semantic label mapping residual magnitude represents the offset magnitude from the expected semantic meaning. This is the penalty factor for insufficient label matching, representing the inverse ratio of the number of overlapping labels in the path. The entity semantic transition index represents the transition from... arrive The semantic drift intensity, The degree of relational role mismatch represents the deviation between the edge role and the actual dependency in the semantic space. A higher value indicates a stronger semantic conflict and structural mismatch on the edge, and it should be prioritized for normalization or removal; the defined dimension is... From the sparse matrix, generate the mapping adjustment matrix: ,in, For mapping adjustment matrix, This indicates that the connection path pairs need to be regularized. This matrix serves as a semantic conflict threshold, used to filter edge pairs that need to be fed into the label reconstruction model; a third-order tensor is constructed. Let represent the tensor state of entity-label-time, and define the entity-label reconstruction function: ,in, For entities With tags The strength of historical associations, For tags In time semantic activity For entities In time The semantic domain drift, For tags Global drift sensitivity indicates whether a space is susceptible to changes in its surroundings. The synonym interference intensity between entity labels is represented by and the fuzzy label density is represented by . (The reconstructed value is missing from the original text.) This represents the credibility strength of the associated labels of the corrected entity at a certain time dimension, used to judge the semantic rationality of edge connections; the edge set structure is updated, conflicting connections are removed, and the reconstructed label tensor is generated. With adjustment matrix Combined for re-evaluating edge sets, define the edge update function: ,in, The preset semantic consistency threshold means that edge connections are retained only when they are not identified as conflicting edges, or when the two entities have high consistency in the label space.

[0029] The present invention is further configured such that the continuously evolving financial knowledge graph driven by local stability and global fusion includes: Based on semantically regularized financial relationship networks and newly added financial corpora, evolutionary input units are constructed, and node evolution driving factors and edge set expansion prediction tensors are established to form graph evolution conditions with structural dynamism and semantic adaptability. Specifically, node evolution driving factors are constructed. Define nodes The evolutionary driving factors are ,in, For nodes The semantic convolutional response strength between the new corpus and the new data. The semantic mutation index between a node and its adjacent entities represents the semantic tension between the old and new connections. The number of historical label migrations for a node represents the discrete state jump count. The node structure perturbation response degree reflects changes in node degree and the direction of newly added connections; larger values ​​indicate a higher response degree. This indicates that the node is located in a region of high structural activity and semantic reconstruction, making it suitable for participation in evolution; an edge set is constructed to expand the prediction tensor, and the prediction tensor is defined. Predicting entity pairs New relationships may be added between them Feasibility: ,in, For nodes With relation type Frequency of occurrence in the new corpus For nodes Correspondence tags The degree of decoupling of historical mutual information, For nodes , The number of hops in the path structure layer. For relation type The evolutionary impedance coefficient represents its sensitivity to inducing global structural changes. High Representative Relationship It has evolutionary and expansion value between node pairs; By combining node local stability with global graph fusion, the evolution trigger state of the graph structure is determined, regions to be updated are selected, and incremental connection construction is performed; specifically, this involves combining node local stability... Global fusion with the map Construct the trigger function: ,in, External attachment strength of a node, and power representing the edge weight pointing to that node; local stability of a node. Defined as: ,in, , This represents the node connectivity before and after the graph update, with 1 indicating connectivity and 0 indicating disconnection; global fusion degree. Defined as: , representing the degree of semantic overlap aggregation across all nodes, where, For node pairs , The size of the intersection of semantic tags The structural distance between node pairs represents the shortest path. For nodes The set of neighbors, if Then the node Considered as entering the evolutionary region, The preset experience threshold; Under the constraint of the evolution threshold, the node set and edge set in the graph structure are iteratively updated to generate a financial knowledge graph with semantic consistency, structural coherence and adaptive evolution capabilities. Specifically, once the evolution region is determined, the following update is performed: Add a new node: Introduce a node whose label status is greater than a threshold. The new entity; Add edge: Keep all And the relationship edges satisfy the condition that the path is not closed; Delete edge: remove Outgoing edges from nodes are used to prevent structural degradation. , , The preset threshold is determined experimentally.

[0030] The present invention is further configured such that the financial knowledge graph structure for generating stable output includes: The evolved financial knowledge graph structure undergoes semantic consistency assessment and structural integrity verification. Specifically, the semantic consistency assessment analyzes the degree of synergistic association between semantic tags in the graph based on semantic robustness assessment factors. Semantic robustness assessment factors are used... The degree of collaborative association between semantic labels in the graph is modeled to measure the robustness of label fusion after the injection of new information. The formula is as follows: ,in, For entities The aggregation depth of the bound semantic tags, For tags Frequency of contextual references between them For tags The greater the semantic overlap, the more redundant it is. The cross-label linkage probability of label pairs is used to measure the density of collaborative semantics and mutual support capabilities within the label space. The higher the value, the more robust the semantics. Structural integrity verification measures the degree to which the node connectivity relationships in the graph are preserved before and after evolution based on the structural regression integrity index; specifically, a structural regression integrity index is constructed. This formula is used to evaluate the continuity of node connections and structural inheritance capabilities before and after graph evolution. , For the map in the first Nodes in version The inward edge retention rate, For the first Nodes in version The recurrence rate of outgoing edges, The frequency of local structural disturbances at the nodes represents the number of structural jumps. For version The proportion of structural relay role played by middle nodes in edge set changes; this indicator confirms the consistency of knowledge graph inheritance through the quantitative connectivity of historical structure; The output stability of the current knowledge graph structure is determined by combining semantic robustness assessment factors and structural regression completeness indicators, thus forming a stable financial knowledge graph structure. Specifically, a structure stability function is defined. By integrating semantic and structural evaluation results, it is determined whether the graph meets the output conditions:

[0031] As a semantic robustness assessment factor, This is an indicator of structural regression integrity. This represents the number of structural conflicts caused by adding new edges to the current graph. Let be the structural conflict gain coefficient, if The graph structure was identified as a stable output. The preset threshold is determined experimentally.

[0032] Example 2: Please see Figure 2 This exemplary AI-based financial knowledge graph construction system includes: Semantic parsing module: Acquires raw data from heterogeneous financial data sources, performs multidimensional semantic parsing on the data sources based on the source domain complexity factor, structural sparsity factor, and semantic perturbation potential energy of each data unit, extracts structured semantic fragments, and constructs a semantic parsing dataset; Financial Entity Recognition Module: Based on the semantic parsing dataset, a financial entity perception unit is constructed. By combining feature asymmetric fusion weights and attribute coupling suppression factors, an entity assignment mechanism for multiple financial semantic domains is established to identify multi-granularity entities and generate unique entity codes, thereby generating a set of financial entities. Relationship Connection Construction Module: Based on the set of financial entities, an entity connection structure oriented towards the relationship-induced triggering path is constructed through the event cascade probability matrix and upstream and downstream dependency coefficients, generating a preliminary financial relationship graph, and semantic translation meta-tags are injected into the edge attributes of the financial relationship graph; Semantic regularization module: It performs deep regularization on the semantically inconsistent mapping relationships in the preliminary financial relationship graph through semantic separation adjustment, and uses the entity label sparse tensor reconstruction mechanism to eliminate pseudo-related connections, forming a semantically regularized financial relationship network. Evolutionary Iteration Module: Based on the semantically regularized financial relationship network and newly added financial corpus, the module dynamically updates the node evolution trajectory and edge set expansion path in the financial relationship graph structure through incremental evolutionary iteration, continuously evolving the financial knowledge graph driven by local stability and global fusion, forming a financial knowledge graph with evolutionary capabilities. Consistency Assessment and Integrity Verification Module: Through semantic robustness assessment factors and structural regression integrity indicators, the module performs semantic consistency assessment and structural integrity verification on the financial knowledge graph after iteration, generating a stable output financial knowledge graph structure.

[0033] It should be noted that the AI-based financial knowledge graph construction system and the AI-based financial knowledge graph construction method provided in the above embodiments belong to the same concept. The specific methods by which each module and unit performs its operations have been described in detail in the method embodiments and will not be repeated here. In practical applications, the AI-based financial knowledge graph construction system provided in the above embodiments can be configured to have different functional modules assigned to it as needed. That is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above, and this is not a limitation here.

[0034] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0035] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0036] In this application, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0037] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0038] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0039] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0040] In the several embodiments provided in this application, it should be understood that the disclosed system can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0041] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0042] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0043] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0044] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for constructing a financial knowledge graph based on artificial intelligence, characterized in that, include: The raw data from heterogeneous financial data sources is acquired. Based on the source domain complexity factor, structural sparsity factor, and semantic perturbation potential of each data unit, multidimensional semantic analysis is performed on the data sources to extract structured semantic fragments and construct a semantic analysis dataset. The source domain complexity factor is a structural feature index generated based on the field nesting level, label reuse frequency, and naming perturbation degree in the data unit. The structural sparsity factor is a distribution sparsity index constructed based on the field value range jump degree, field missing ratio, and field distribution discreteness. The semantic perturbation potential is a semantic heterogeneity index calculated by combining the field semantic variation amplitude, context semantic drift degree, and semantic interference item density. Based on the semantic parsing dataset, a financial entity perception unit is constructed. Combining feature asymmetric fusion weights and attribute coupling suppression factors, an entity assignment mechanism for multiple financial semantic domains is established to identify multi-granularity entities and generate unique entity codes, thus generating a set of financial entities. Specifically, the feature asymmetric fusion weights are used to perform directional matching of the semantic coupling degree and attribute fusion strength among candidate entity units. The attribute coupling suppression factor is used to identify and weaken redundant relationships caused by high-frequency attribute linkages, suppressing misidentified entity association paths. The entity assignment mechanism refers to combining semantic categories and attribute aggregation patterns to perform granular division and semantic identification of candidate units, achieving refined assignment for multiple semantic domains. Based on a set of financial entities, an entity connection structure oriented towards relationship-induced triggering paths is constructed using an event cascade probability matrix and upstream / downstream dependency coefficients. A preliminary financial relationship graph is generated, and semantic translation meta-tags are injected into the edge attributes of the financial relationship graph to express the semantic evolution direction, interaction mode, and structural role dependency type of entity relationships. The event cascade probability matrix and upstream / downstream dependency coefficients are used to characterize the event influence propagation relationship between entities and the semantic subordination structure between entities. The semantically inconsistent mapping relationships in the preliminary financial relationship graph are deeply normalized by semantic separation adjustment, and the sparse tensor reconstruction mechanism of entity labels is used to eliminate pseudo-related connections, forming a semantically normalized financial relationship network. The sparse tensor reconstruction mechanism of entity labels is to construct an entity label sparse tensor reconstruction model by combining the evolution state index of the entity label space, and reconstruct the entity labels in the preliminary financial relationship graph to eliminate pseudo-related connections. Based on semantically regularized financial relationship networks and newly added financial corpora, the evolution trajectory of nodes and edge set expansion paths in the financial relationship graph structure are dynamically updated through evolutionary incremental iteration, continuously evolving a financial knowledge graph driven by local stability and global fusion, thus forming a financial knowledge graph with evolutionary capabilities. By using semantic robustness assessment factors and structural regression integrity indicators, the semantic consistency of the iterated financial knowledge graph is evaluated and the structural integrity is verified, generating a stable output financial knowledge graph structure.

2. The method for constructing a financial knowledge graph based on artificial intelligence according to claim 1, characterized in that, The construction of the semantic parsing dataset includes: The raw data from the acquired heterogeneous financial data sources are divided into data units according to data source, field type, and semantic content; Based on the source domain complexity factor of each data unit, the degree of hierarchical nesting and semantic fusion redundancy of the unit in the source structure are quantitatively identified. Based on the structural sparsity factor, a structural measure is performed on the field distribution completeness and local missing features between fields of this data unit. Based on semantic perturbation potential, perturbation analysis is performed on the degree of contextual ambiguity and semantic variation of text data fields; By combining the above factors, a multidimensional semantic parsing matrix is ​​constructed, and a parsing operation that combines semantic and structural dimensions is performed on each data unit. The structured semantic fragments with clear semantic boundaries obtained after parsing are extracted in a standardized manner and uniformly collected to form a semantic parsing dataset.

3. The method for constructing a financial knowledge graph based on artificial intelligence according to claim 2, characterized in that, The construction of financial entity sensing units includes: Based on the semantic labels and attribute fragments in the semantic parsing dataset, entity candidate units that meet the conditions for multidimensional semantic recognition are extracted; By generating unique entity codes and combining them with an assignment mechanism to output unique and traceable code identifiers for each entity, a set of financial entities is constructed.

4. The method for constructing a financial knowledge graph based on artificial intelligence according to claim 3, characterized in that, The asymmetric feature fusion weights are constructed by establishing a weighting matrix based on attribute semantic density, fragment overlap, label activation index, lexical offset length, and co-occurrence order difference, and employing a fusion strategy with directional feature distribution. The attribute coupling suppression factor constructs a constraint matrix based on the polysemous strength of the attribute context, the cross-semantic attribution ratio, the degree of nonlinear interaction, and the attribution fuzzy boundary parameters. It encodes and limits the attribute linkage relationship by combining the asymmetric structural distribution parameters within the coupling dimension.

5. The method for constructing a financial knowledge graph based on artificial intelligence according to claim 1, characterized in that, The formation of semantically regular financial relationship networks includes: By adjusting semantic separation, the semantically inconsistent mapping relationships in the preliminary financial relationship graph are normalized. The adjustment factor, which is constructed by structural connectivity parameters, semantic mapping residual parameters and label matching quantity, is used to identify the connection paths with semantic conflicts in the graph edge set. Based on the identified connection paths, a mapping adjustment matrix is ​​generated, and an entity label sparse tensor reconstruction model is constructed by combining the evolution state index of the entity label space to reconstruct the entity labels in the preliminary financial relationship graph and eliminate spurious connections. Based on the reconstructed entity label tensor and adjustment matrix, the edge set structure is updated, conflicting edge connections are removed, and a financial relationship network with semantic consistency and label structure stability is formed.

6. The method for constructing a financial knowledge graph based on artificial intelligence according to claim 5, characterized in that, Continuously evolving financial knowledge graphs, driven by local stability and global fusion, include: Based on semantically regular financial relationship networks and newly added financial corpora, we construct evolutionary input units, establish node evolution driving factors and edge set expansion prediction tensors, and form graph evolution conditions with structural dynamism and semantic adaptability. By combining the local stability of nodes with the global fusion of the graph, the evolution triggering state of the graph structure is determined, the regions to be updated are selected, and incremental connection construction is performed. Under the constraint of evolution threshold, the node set and edge set in the graph structure are iteratively updated to generate a financial knowledge graph with semantic consistency, structural coherence and adaptive evolution capabilities.

7. The method for constructing a financial knowledge graph based on artificial intelligence according to claim 6, characterized in that, The structure for generating stable financial knowledge graphs includes: The semantic consistency assessment and structural integrity verification of the evolved financial knowledge graph structure are performed. Semantic consistency assessment analyzes the degree of collaborative association between semantic tags in the graph based on semantic robustness assessment factors; Structural integrity verification measures the degree to which the node connectivity relationships of the graph are preserved before and after evolution based on the structural regression integrity index. By combining semantic robustness assessment factors and structural regression completeness indicators, the output stability of the current graph structure is determined, thus forming a stable financial knowledge graph structure.

8. An artificial intelligence-based financial knowledge graph construction system, used to implement the artificial intelligence-based financial knowledge graph construction method according to any one of claims 1-7, characterized in that, include: Semantic parsing module: Acquires raw data from heterogeneous financial data sources, performs multidimensional semantic parsing on the data sources based on the source domain complexity factor, structural sparsity factor, and semantic perturbation potential energy of each data unit, extracts structured semantic fragments, and constructs a semantic parsing dataset; Financial Entity Recognition Module: Based on the semantic parsing dataset, a financial entity perception unit is constructed. Combining feature asymmetric fusion weights and attribute coupling suppression factors, an entity assignment mechanism for multiple financial semantic domains is established to identify multi-granularity entities and generate unique entity codes, thereby generating a set of financial entities. Relationship Connection Construction Module: Based on the set of financial entities, an entity connection structure oriented towards the relationship-induced triggering path is constructed through the event cascade probability matrix and upstream and downstream dependency coefficients, generating a preliminary financial relationship graph, and semantic translation meta-tags are injected into the edge attributes of the financial relationship graph; Semantic regularization module: It performs deep regularization on the semantically inconsistent mapping relationships in the preliminary financial relationship graph through semantic separation adjustment, and uses the entity label sparse tensor reconstruction mechanism to eliminate pseudo-related connections, forming a semantically regularized financial relationship network. Evolutionary Iteration Module: Based on the semantically regularized financial relationship network and newly added financial corpus, the module dynamically updates the node evolution trajectory and edge set expansion path in the financial relationship graph structure through incremental evolutionary iteration, continuously evolving the financial knowledge graph driven by local stability and global fusion, forming a financial knowledge graph with evolutionary capabilities. Consistency Assessment and Integrity Verification Module: Through semantic robustness assessment factors and structural regression integrity indicators, the module performs semantic consistency assessment and structural integrity verification on the financial knowledge graph after iteration, generating a stable output financial knowledge graph structure.

Images