Entity relationship extraction and search system and method constructed based on knowledge graph

Abstract

The invention discloses an entity relationship extraction and search system and method based on knowledge graph construction, and relates to the technical field of artificial intelligence and information retrieval, and the method comprises the following steps: obtaining a cognitive load mapping matrix and knowledge graph data of a target object; according to the cognitive load mapping matrix and the node weight, performing multi-scale hierarchical projection processing on the knowledge graph data to generate each hierarchical view containing homonymous nodes at different hierarchical positions, and establishing a mutual mapping channel for the homonymous nodes among the hierarchical views; wherein the node weight is obtained by performing aggregation processing on the cognitive load mapping matrix. According to the scheme, the multi-scale projection is generated through clustering analysis, the normalized weight and the dynamic threshold interval are combined, the multi-level view adaptive to the target cognitive ability can be dynamically generated according to the cognitive load mapping matrix, meanwhile, cross-level semantic association is reserved through the mutual mapping channel, and therefore the cognitive load is reduced, and meanwhile semantic information loss is avoided.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and information retrieval technology, specifically to an entity relationship extraction and search system and method based on knowledge graph construction. Background Technology

[0002] Against the backdrop of the rapid development of modern information retrieval and artificial intelligence technologies, knowledge graphs, as a tool that can structurally represent entities, concepts, and relationships between entities, have been widely used in semantic search, intelligent question answering, and structured information retrieval in recent years. Knowledge graphs represent entities through nodes, represent relationships between entities through edges, and use graph structures to reveal semantic connections between information, thereby enabling computers to understand and process complex knowledge systems.

[0003] However, in the process of implementing the technical solution of the invention in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems:

[0004] Existing knowledge graph-assisted search methods are typically based on entity recognition and relation extraction techniques. They generate knowledge graphs by parsing text or structured data and then perform search queries on these graphs. The advantage of these methods is that they can capture complex relationships between entities using graph structures, providing richer semantic information than traditional retrieval methods, and enabling deep querying and reasoning based on relationships. However, these methods usually assume that the target audience can fully understand the semantics of nodes and relationships. When targeting people with cognitive impairments, these methods lack mechanisms to quantify the target audience's cognitive abilities, making it difficult to generate presentation strategies that adapt to the individual cognitive abilities of the target audience for the search results. Furthermore, when generating search suggestions, there is a lack of dynamic alternative path design for when the target audience fails to understand the information, which can easily lead to erroneous search results. Summary of the Invention

[0005] The purpose of this invention is to provide an entity relationship extraction and search system and method based on knowledge graph construction, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] In a first aspect, this invention discloses an entity relationship extraction and search method based on knowledge graph construction, applied to knowledge graph-assisted retrieval for people with cognitive impairment, including the following steps:

[0008] Obtain the cognitive load mapping matrix and knowledge graph data of the target object;

[0009] Based on the cognitive load mapping matrix and node weights, the knowledge graph data is subjected to multi-scale hierarchical projection processing to generate hierarchical views containing nodes with the same name at different hierarchical positions, and a mutual mapping channel is established between nodes with the same name in each hierarchical view.

[0010] The node weights are obtained by aggregating the cognitive load mapping matrix.

[0011] Based on the node weight and hierarchical position, nodes with the same name are selected in priority order to generate the main prompt sequence of each level view;

[0012] Scan the nodes with the same name in each level of view that are not occupied by the main prompt sequence, and calculate their consistency score in different level views in combination with the mutual mapping channel;

[0013] Determine whether the consistency score is greater than a preset discrimination threshold; if so, generate a candidate subsequence based on the corresponding nodes with the same name.

[0014] By combining the views at each level and their corresponding main prompt sequence and alternative sub-sequences, search results are generated.

[0015] Secondly, this invention discloses an entity relationship extraction and search system based on knowledge graph construction, comprising:

[0016] The data acquisition module is used to acquire the cognitive load mapping matrix and knowledge graph data of the target object;

[0017] The hierarchical view generation module is used to perform multi-scale hierarchical projection processing on the knowledge graph data according to the cognitive load mapping matrix and node weights, generate hierarchical views containing nodes with the same name at different hierarchical positions, and establish mutual mapping channels between nodes with the same name in each hierarchical view.

[0018] The prompt sequence generation module is used to select nodes with the same name in priority order according to the node weight and hierarchical position, and then generate the main prompt sequence of each level view;

[0019] Scan nodes with the same name that are not occupied by the main prompt sequence in the same level view, and calculate their consistency score in different level views in combination with the mutual mapping channel;

[0020] Determine whether the consistency score is greater than a preset discrimination threshold; if so, generate a candidate subsequence based on the corresponding nodes with the same name.

[0021] The search results generation module is used to combine the views at each level and their corresponding main prompt sequence and alternative sub-sequences to generate search results.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] 1. This solution generates multi-scale projections through cluster analysis. Combined with normalized weights and dynamic threshold ranges, it can dynamically generate multi-level views that adapt to the target cognitive ability based on the cognitive load mapping matrix. At the same time, it retains the semantic associations across levels through the mutual mapping channel, thereby reducing cognitive load while avoiding the loss of semantic information and improving the adaptability and accuracy of retrieval results.

[0024] 2. This solution establishes cross-view associations through mutual mapping channels and conducts comprehensive evaluation by combining hierarchical threshold differences and weight distribution. It realizes a dynamic filtering mechanism for candidate nodes across hierarchical views. When cognitively impaired target objects have difficulty understanding the main prompt sequence, it can generate candidate sub-paths that conform to their cognitive load characteristics, significantly improving the fault tolerance and continuity of retrieval results. Attached Figure Description

[0025] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0026] Figure 1 This is a flowchart illustrating the steps of the entity relationship extraction and search method based on knowledge graph construction according to the present invention.

[0027] Figure 2 This is a schematic diagram of the process for performing multi-scale hierarchical projection processing on the knowledge graph data provided by the present invention.

[0028] Figure 3 A schematic diagram illustrating the process of generating search results provided by this invention;

[0029] Figure 4 A schematic diagram of the process for generating an iterative weight propagation graph provided by the present invention;

[0030] Figure 5 This is a schematic diagram illustrating the module functions of the entity relationship extraction and search system based on knowledge graph construction provided by the present invention. Detailed Implementation

[0031] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0032] Application Overview:

[0033] In existing technologies, knowledge graph-assisted search methods are typically based on entity recognition and relation extraction techniques. They generate knowledge graphs by parsing text or structured data and then perform search queries on these graphs. These approaches can capture complex relationships between entities using graph structures, providing richer semantic information than traditional retrieval methods and enabling deep querying and reasoning based on relationships. However, existing methods usually assume that the target object can fully understand the semantics of nodes and relationships, lacking a quantitative assessment mechanism for the cognitive abilities of people with cognitive impairments. This makes it difficult for search results to be adapted to individual cognitive abilities, and the lack of dynamic alternative path design when generating search suggestions can easily lead to erroneous results due to misunderstandings.

[0034] To address these issues, research has revealed that the core challenge lies in quantifying the cognitive capacity of the target audience and dynamically adjusting retrieval strategies. Firstly, quantifying cognitive load requires establishing a mapping relationship between the target audience's interactive data and knowledge graph nodes. This necessitates the introduction of a cognitive load mapping matrix to characterize the fit between the target audience and the nodes. Secondly, the need for hierarchical projection processing stems from the sensitivity of cognitively impaired individuals to information presentation methods. Multi-scale projection can generate views of varying complexity. Finally, the implementation of dynamic alternative paths requires establishing a mechanism for linking views. The proposal of mutual projection channels aims to solve the problem of consistent judgment across hierarchical nodes. These considerations ultimately lead to a technical approach that combines cognitive load mapping, multi-scale projection, and dynamic path adjustment.

[0035] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0036] Example 1:

[0037] Please see Figure 1 A knowledge graph-based entity relationship extraction and search method, applied to knowledge graph-assisted retrieval for people with cognitive impairment, includes the following steps:

[0038] Obtain the cognitive load mapping matrix and knowledge graph data of the target object;

[0039] Based on the cognitive load mapping matrix and node weights, the knowledge graph data is subjected to multi-scale hierarchical projection processing to generate hierarchical views containing nodes with the same name at different hierarchical positions, and a mutual mapping channel is established between nodes with the same name in each hierarchical view.

[0040] The node weights are obtained by aggregating the cognitive load mapping matrix.

[0041] Based on node weight and hierarchical position, nodes with the same name are selected in priority order to generate the main prompt sequence of each level view;

[0042] Scan nodes with the same name that are not occupied by the main prompt sequence in each level of view, and calculate their consistency score in different levels of view by combining the mutual mapping channel;

[0043] Determine whether the consistency score is greater than the preset discrimination threshold; if so, generate candidate subsequences based on the corresponding nodes with the same name.

[0044] By combining the views at each level and their corresponding main prompt sequence and alternative sub-sequences, search results are generated.

[0045] Among them, the cognitive load mapping matrix refers to a multi-dimensional weight matrix constructed based on the real-time performance of the target object in information processing and task interaction;

[0046] Knowledge graph data refers to a structured collection of data built based on knowledge in a specific domain;

[0047] Node weight refers to a quantitative indicator used to characterize the relevance, importance, and cognitive load fit between a target object and each node with the same name in a knowledge graph.

[0048] Multi-scale hierarchical projection processing refers to the process of splitting and reorganizing entities and relationships in a knowledge graph level by level based on the overall structure of the knowledge graph and using the cognitive load mapping matrix and node weights.

[0049] Hierarchical position refers to the structural position of the same node with the same name in different hierarchical views;

[0050] Nodes with the same name refer to nodes in knowledge graph data that have the same semantic identifier or the same entity name;

[0051] Hierarchical view refers to a hierarchical projection representation generated based on knowledge graph data and cognitive load mapping matrix;

[0052] Mutual mapping channels refer to cross-level connection paths formed by establishing a one-to-one correspondence between nodes with the same name at different levels in a multi-level view based on a knowledge graph.

[0053] The main prompt sequence refers to the set of nodes with the same name that have the highest priority in each level of view and their serialized arrangement, determined by the cognitive load mapping matrix of the target object during the knowledge graph-assisted retrieval process.

[0054] Consistency score is an indicator used to measure whether the structural characteristics, weight distribution and mutual mapping relationship of nodes with the same name are consistent in different hierarchical views;

[0055] The preset discrimination threshold is a numerical limit used to measure the consistency of nodes with the same name in cross-level views. Specifically, it is obtained by distributing the consistency scores of all nodes with the same name and dividing the results into percentages.

[0056] Alternate subsequences refer to node sequences that can be presented in place of the main prompt sequence when the main prompt sequence is not correctly understood by the target object or does not produce the expected effect.

[0057] Search results refer to the structured prompts generated by the system based on knowledge graph data and cognitive load mapping matrix, which are used to assist people with cognitive impairment in their searches.

[0058] This solution quantifies the cognitive characteristics of the target object through a cognitive load mapping matrix, generates view levels adapted to different cognitive levels through multi-scale projection, and utilizes mutual mapping channels to achieve dynamic path adjustment across levels. This hierarchical and progressive structure ensures the reliability of the core retrieval path and provides a fault-tolerance mechanism through alternative sub-sequences, effectively solving the comprehension bias problem that is prone to occur in information retrieval by people with cognitive impairments. Through the above technical solution, this application can dynamically adjust the presentation level and retrieval path of the knowledge graph according to the cognitive ability of the target object, ensuring the accurate transmission of core information while providing semantically consistent alternative retrieval schemes. This not only improves the information acquisition efficiency of people with cognitive impairments, but also reduces the probability of erroneous retrieval results through the synergistic effect of multi-level views, achieving accurate matching between the knowledge retrieval process and the cognitive characteristics of the target object.

[0059] The above describes a complete solution for entity relationship extraction and search based on knowledge graphs. The following section details how to obtain the cognitive load mapping matrix and knowledge graph data for the target object, specifically including:

[0060] Acquire interaction data and knowledge graph data of the target object;

[0061] Isomorphic mapping is performed on the interaction data and knowledge graph data respectively;

[0062] Perform bidirectional iterative projection, including forward and reverse mapping, on the isomorphic mapping result until the bidirectional iterative projection result converges, and output the cognitive load mapping matrix;

[0063] In this matrix, the rows of the cognitive load mapping matrix represent the nodes with the same name in the knowledge graph data, the columns of the cognitive load mapping matrix represent the interaction data, and the elements of the cognitive load mapping matrix represent the compatibility strength between the target object and the nodes with the same name in the current environment.

[0064] Among them, isomorphic mapping processing refers to the process of converting two types of data from different sources—interaction data of the target object and knowledge graph data—into structurally corresponding and comparable mapping representations.

[0065] Bidirectional iterative projection refers to the process of continuously adjusting the matching relationship between the interaction data of the target object and the knowledge graph data through iterative calculations in two directions: forward mapping and backward mapping, until the mapping relationship converges.

[0066] The above content will be described in detail below:

[0067] The interactive interface is used to obtain the interaction data of the target object in real time during the operation process. The interaction data includes, but is not limited to, click records, mouse or touch operations, page dwell time, input content and response selection.

[0068] Knowledge graph data is obtained by accessing a pre-established knowledge graph database. The knowledge graph data includes, but is not limited to, the semantics and attribute information of nodes with the same name, the reference relationship between nodes with the same name, the adjacency relationship, and the semantic association between nodes with the same name.

[0069] The interaction data and knowledge graph data are vectorized and encoded respectively to obtain the feature vectors of the interaction data and the feature vectors of the knowledge graph data. These are then used as the results of isomorphic mapping, thereby mapping the interaction data and knowledge graph data to the feature space of the same dimension.

[0070] Perform a bidirectional iterative projection, including both forward and reverse mappings, on the isomorphic mapping result until the bidirectional iterative projection result converges, and output the cognitive load mapping matrix:

[0071] The expression for the bidirectional iterative projection process is as follows:

[0072]

[0073] In the formula, This represents the intermediate adaptation matrix output by the forward mapping. This represents the cognitive load mapping matrix after back-mapping correction. Indicates the first The cognitive load mapping matrix generated in the next iteration Represents the feature vector of the interactive data. Represents the feature vector of knowledge graph data. This represents a forward mapping function that projects the feature vectors of the interactive data onto the node space and generates the adaptation strength. This represents the inverse mapping function, which projects the feature vectors of the knowledge graph data onto the target object space and corrects the fit strength.

[0074] Until the bidirectional iterative projection results converge:

[0075]

[0076] In the formula, Describing the Frobenius norm, This represents the convergence threshold.

[0077] This solution establishes a dynamic association between interactive data and knowledge graph data through bidirectional iterative projection, enabling the cognitive load mapping matrix to continuously capture changes in the cognitive state of the target object. Through the above technical solution, this application can generate dynamically quantified cognitive adaptation indicators based on the real-time interactive behavior of the target object, solving the problem of insufficient adaptability of search results caused by the lack of a cognitive ability quantification mechanism in the prior art.

[0078] The above describes how to obtain the cognitive load mapping matrix and knowledge graph data of the target object. The following describes how to perform multi-scale hierarchical projection processing on the knowledge graph data based on the cognitive load mapping matrix and node weights, generating hierarchical views containing nodes with the same name at different hierarchical positions, and establishing mutual mapping channels between nodes with the same name in each hierarchical view. Please refer to [link / reference]. Figure 2 , Figure 2 This is a flowchart illustrating the multi-scale hierarchical projection processing of knowledge graph data provided in an embodiment of this application. The multi-scale hierarchical projection processing of knowledge graph data specifically includes:

[0079] Cluster analysis is performed on the elements of the cognitive load mapping matrix to determine the set of projection scales, and a corresponding hierarchical threshold interval is generated for each projection scale.

[0080] The normalized weight of the corresponding node with the same name is calculated based on the adaptation strength and node weight, and then compared with the threshold region of each level. Based on the comparison results, different classification strategies are applied to the nodes with the same name, thereby generating a view of each level containing nodes with the same name at different level positions.

[0081] Compare the nodes with the same name in different view levels one by one, and establish mutual mapping channels for nodes with the same name and the same node type.

[0082] Cluster analysis refers to the process of grouping elements in the cognitive load mapping matrix according to their similarity.

[0083] The projection scale set refers to a set of scale parameters used to generate multi-level views;

[0084] The hierarchical threshold range refers to the set of threshold ranges defined for each projection scale;

[0085] Adaptation strength refers to the degree to which nodes with the same name are preferentially selected or presented by the system in the current hierarchical view and surrounding environment;

[0086] Normalized weights indicate the relative fit strength of nodes with the same name in a hierarchical view;

[0087] Classification strategy refers to the operational rules used to perform hierarchical processing and view generation for nodes with the same name.

[0088] The above content will be described in detail below:

[0089] For each element of the cognitive load mapping matrix, calculate its similarity to elements of other nodes with the same name. The specific calculation formula is as follows:

[0090]

[0091] In the formula, Indicates nodes with the same name Corresponding elements and nodes with the same name The similarity between corresponding elements Represents the nodes with the same name in the cognitive load mapping matrix. The corresponding element, Represents the nodes with the same name in the cognitive load mapping matrix. The corresponding element, This represents the maximum value of an element in the cognitive load mapping matrix. This represents the minimum value of an element in the cognitive load mapping matrix;

[0092] Clustering algorithms (such as hierarchical clustering, k-means clustering, or density clustering) are used to group similarity into clusters.

[0093] For each cluster, statistics are performed, and the center value of each cluster is extracted. The separation degree between clusters is obtained by calculating the distance between the center values ​​of adjacent clusters. It is then determined whether the separation degree is greater than a preset screening threshold. If so, the corresponding cluster is treated as an independent projection scale, thereby generating a set of projection scales. A corresponding hierarchical threshold interval is generated for each projection scale. The specific calculation formula is as follows:

[0094]

[0095] In the formula, Represents clusters The threshold at the level, Represents clusters The threshold at the level, Represents clusters The mean of the contained elements. Represents clusters The standard deviation of the included elements , Indicates system parameters;

[0096] The normalized weight of the corresponding node with the same name is calculated based on the adaptation strength and node weight. The specific calculation formula is as follows:

[0097]

[0098] In the formula, Indicates nodes with the same name Normalized weights, Indicates nodes with the same name The node weights, Indicates nodes with the same name The strength of the fit, Indicates nodes with the same name , Indicates the total number of nodes with the same name;

[0099] The node weights are obtained by averaging all elements in the column containing the corresponding node with the same name in the cognitive load mapping matrix.

[0100] The normalized weights of nodes with the same name are compared with the threshold regions of each level. Based on the comparison results, different classification strategies are applied to nodes with the same name, thereby generating hierarchical views containing nodes with the same name at different levels.

[0101] Determine whether the normalized weight is greater than the threshold at the level. If so, mark the corresponding node with the same name as the reserved type at the corresponding projection scale and present it in full in the subsequent corresponding hierarchical view.

[0102] If the normalized weight is less than the threshold of the level, the corresponding node with the same name is marked as a collapsed type under the corresponding projection scale. It will not be displayed separately in the subsequent corresponding level view, but will be aggregated to the adjacent node with the same name of the retained type, and only the reference relationship will be retained.

[0103] In other cases, the corresponding nodes with the same name are marked as aggregation types under the corresponding projection scale. In the subsequent corresponding hierarchical view, the aggregation type with the same name with the highest normalized weight is taken as the aggregation target, and the semantics of other aggregation type nodes with the same name are merged with the aggregation target.

[0104] The normalized weights of nodes of the same name of different types within the same projection scale are sorted to obtain the relative proportional positions within the same projection scale, and these are used as the hierarchical positions of the corresponding nodes of the same name, thereby generating a hierarchical view containing nodes of the same name with different hierarchical positions.

[0105] Compare the nodes with the same name in different view levels one by one, and establish mutual mapping channels for nodes with the same name and the same node type.

[0106] This solution generates multi-scale projections through cluster analysis. By combining normalized weights and hierarchical threshold ranges, it can flexibly allocate node levels according to the cognitive state of the target object. At the same time, it preserves the semantic associations across levels through mutual mapping channels, avoiding the semantic fragmentation problem caused by hierarchical division. Through the above technical solution, this application can dynamically generate multi-level views adapted to the target's cognitive ability based on the cognitive load mapping matrix, and maintain semantic consistency across levels through mutual mapping channels. This reduces cognitive load while avoiding the loss of semantic information, thereby improving the adaptability and accuracy of retrieval results.

[0107] The above describes how to perform multi-scale hierarchical projection processing on knowledge graph data based on the cognitive load mapping matrix and node weights to generate hierarchical views containing nodes with the same name at different hierarchical positions, and establish mutual mapping channels between nodes with the same name in different hierarchical views. The following describes how to select nodes with the same name in priority order based on node weights and hierarchical positions to generate the main prompt sequence for each hierarchical view, specifically including:

[0108] In each hierarchical view, nodes with the same name are sorted by priority according to their hierarchical position in the hierarchical view. The priority sorting results are then adjusted by weighting the nodes with the same name. The top Z nodes with the same name in the adjusted priority sorting results are selected to generate the main prompt sequence of the corresponding hierarchical view.

[0109] The value of Z is related to the elements of the cognitive load mapping matrix of all nodes with the same name displayed in the corresponding hierarchical view, and its specific calculation formula is as follows:

[0110]

[0111] In the formula, This indicates the number of nodes with the same name selected from the first Z positions. Indicates nodes with the same name The strength of the fit, This indicates the total number of nodes with the same name displayed in the corresponding view hierarchy. This represents the maximum fit strength of all nodes with the same name displayed in the corresponding hierarchical view within the cognitive load mapping matrix. This indicates a rounding operation, ensuring that Z is an integer.

[0112] This solution prioritizes nodes with the same name in each hierarchical view according to their hierarchical position and then adjusts them by weighting them in the cognitive load mapping matrix. This ensures that when generating the main prompt sequence, the logical order of nodes with the same name in the hierarchical structure is preserved, while prioritizing nodes with high compatibility with the target object. This enhances the relevance and effectiveness of the prompt sequence. Furthermore, this solution dynamically correlates the number of nodes with the same name with their compatibility strength and the total number of nodes with the same name in the hierarchical view. This not only ensures that the main prompt sequence covers core information but also avoids excessive redundant nodes, improving the accuracy and understandability of the prompts. It achieves dual optimization of the hierarchical structure of nodes with the same name and the cognitive load of the target object, making the generated main prompt sequence both logically consistent and personally adaptable, thereby improving the system's response efficiency and prompt success rate to target object interactions.

[0113] The above describes how to select nodes with the same name in priority order based on node weight and hierarchical position to generate the main prompt sequence for each level of view. The following describes scanning for nodes with the same name in each level of view that are not occupied by the main prompt sequence, and calculating their consistency score across different level views using the mutual mapping channel. Specifically, this includes:

[0114] For nodes with the same name in the main prompt sequence of each level view, scan for nodes with the same name in other level views that are not occupied by the corresponding main prompt sequence, and filter out the corresponding nodes with the same name based on the mutual mapping channel.

[0115] Perform a difference analysis on the hierarchical threshold range of the same-name node and the corresponding hierarchical view containing the same-name node, and then calculate the consistency score by weighting it with the normalized weight.

[0116] Among them, difference analysis refers to the process of comparing the hierarchical position and hierarchical threshold range of nodes with the same name in different hierarchical views;

[0117] Weighted calculation refers to the process of combining the normalized weight of a node with the same name in the current view with the hierarchical difference of its corresponding node with the same name in other hierarchical views.

[0118] The above content will be described in detail below:

[0119] For nodes with the same name in the main prompt sequence of each level view, scan for nodes with the same name in other level views that are not occupied by the corresponding main prompt sequence, and filter out the corresponding nodes with the same name based on the mutual mapping channel.

[0120] The difference analysis is performed on the hierarchy threshold range and hierarchy position of the hierarchy view containing the same-named node and its corresponding same-named node. The specific calculation formula is as follows:

[0121]

[0122] In the formula, Indicates nodes with the same name In hierarchy view and hierarchical view The results of the difference analysis, Represents a hierarchical view The threshold at the level, Represents a hierarchical view The threshold at the level, Represents a hierarchical view The threshold at the level, Represents a hierarchical view The threshold at the level, Indicates nodes with the same name In hierarchy view hierarchical position Indicates nodes with the same name In hierarchy view hierarchical position The weighting coefficients representing the differences in threshold levels. The weighting coefficients represent the differences in hierarchical positions, and the results of the difference analysis are weighted with the corresponding normalized weights to generate a consistency score.

[0123] Determine whether the consistency score is greater than the preset discrimination threshold. If so, obtain the number of corresponding nodes with the same name in the main prompt sequence that meet the preset discrimination threshold, and generate a candidate subsequence with the minimum number of nodes with the same name.

[0124] This solution establishes cross-view associations through mutual mapping channels and conducts a comprehensive evaluation by combining hierarchical threshold differences and weight distribution to effectively identify candidate nodes with cross-level semantic continuity. Through the above technical solution, this application realizes a dynamic screening mechanism for candidate nodes across hierarchical views. When cognitively impaired target objects have difficulty understanding the main prompt sequence, it can automatically generate candidate paths that conform to their cognitive load characteristics. By quantitatively evaluating the consistency of nodes in different views, it avoids semantic breaks caused by hierarchical differences and significantly improves the fault tolerance and continuity of search results.

[0125] The above describes scanning for nodes with the same name that are not occupied by the main prompt sequence in each level of the view, and calculating their consistency score across different levels of the view by combining the mutual mapping channel. The following describes how to generate search results by integrating each level of the view and its corresponding main prompt sequence and alternative sub-sequences. Please refer to [link / reference]. Figure 3 , Figure 3 This is a schematic diagram of the process for generating search results provided in an embodiment of this application. Generating search results specifically includes:

[0126] After generating the main prompt sequence and alternative sub-sequences, obtain feedback data on the interaction data;

[0127] The feedback data is compared with the main cue sequence and alternative subsequences, and the cognitive load mapping matrix, the views at each level, the main cue sequence, and the alternative subsequences are updated based on the comparison results.

[0128] The updated cognitive load mapping matrix, the updated views at each level, the updated main prompt sequence, and the updated alternative subsequences are combined to generate search results.

[0129] Feedback data refers to the collection of behavior and selection information of the target object after interacting with the system.

[0130] The above content will be described in detail below:

[0131] After generating the main prompt sequence and alternative sub-sequences, obtain feedback data on the interaction data. The feedback data includes, but is not limited to, whether the target object correctly understands the prompt, whether the operation is successful or not, the order of selection or click, and the matching status with the alternative sub-sequences.

[0132] The feedback data is compared node-by-node with the main cue sequence and candidate subsequences, and the deviation between the comparison results and the cognitive load mapping matrix is ​​calculated. The specific calculation formula is as follows:

[0133]

[0134] In the formula, Indicates nodes with the same name deviation, Indicates nodes with the same name The comparison results Indicates nodes with the same name The strength of the fit, Indicates the adjustment factor;

[0135] The bias is superimposed on the elements of the corresponding nodes with the same name in the cognitive load mapping matrix to update the cognitive load mapping matrix, and the node weights and normalized weights are recalculated.

[0136] Based on the updated cognitive load mapping matrix, the views of each level are regenerated, and the comparison results are weighted with the recalculated normalized weights. Then, the weighted weights are compared with the level threshold range of the regenerated views of each level, the classification results of the same-name nodes are adjusted, and the updated views of each level are generated.

[0137] The main prompt sequence and alternative subsequences are updated based on the updated level threshold range of each level view, the corresponding updated level position, the recalculated node weights, and the weighted result of the recalculated normalized weights.

[0138] The updated cognitive load mapping matrix, updated views at each level, updated main cue sequence, updated alternative subsequences, and comparison results are used to generate an iterative weight propagation graph through a gating fusion mechanism.

[0139] The comparison results are used as a gating signal input to the updated cognitive load mapping matrix, and a weighted correction operation is performed on the elements of the corresponding nodes with the same name, thereby updating the recalculated node weights.

[0140] The updated node weights are then propagated along the hierarchical view path of each level of view:

[0141] The propagation starts from the node with the same name that has been adjusted, and passes the weight changes level by level along the path according to the mutual mapping channel between the updated views of each level with a preset attenuation factor. The propagation rule is to perform cumulative weighting on adjacent nodes with the same name, and at the cross-view mutual mapping point, simultaneously map the changes to the nodes with the same name in the corresponding level view, and perform a weighted average solution on the node weights of each node with the same name after propagation within the same level view to generate the global weight value of the view at that level.

[0142] Map the propagated node weights back to the updated main cue sequence and candidate subsequences:

[0143] Using the node ID as the key, locate all sequences containing the same-name node in the updated main prompt sequence and candidate subsequences, and then correct the updated main prompt sequence and candidate subsequences according to the node weights after propagation of the corresponding same-name node;

[0144] After the correction is completed, the propagation process of the recalculated node weights, the updated global weight values ​​of each level view, the corrected main prompt sequence and alternative subsequences are comprehensively recorded, and an iterative weight propagation graph is generated according to the normalized data structure.

[0145] Using the iterative weight propagation graph as a constraint, the updated hierarchical views, the updated main suggestion sequence, and the updated alternative subsequences are integrated to generate search results:

[0146] Recalculate the view weights for each updated view hierarchy:

[0147] The comprehensive priority score for each updated hierarchical view is calculated based on the global weight value of the corresponding updated hierarchical view in the iterative weight propagation graph and the node weight of the same-named node within the corresponding updated hierarchical view after propagation, combined with the adaptation strength of the same-named node in the updated cognitive load mapping matrix. The specific calculation formula is as follows:

[0148]

[0149] In the formula, Represents the updated hierarchical view The overall priority score, This represents the updated hierarchical view recorded in the iterative weight propagation graph. The global weight value, This represents the updated hierarchical view recorded in the iterative weight propagation graph. Same-name node The node weights after propagation This represents the nodes with the same name in the updated cognitive load mapping matrix. The strength of the fit, This represents the updated hierarchical view recorded in the iterative weight propagation graph. The total number of nodes with the same name in the middle. , and These represent the corresponding impact factors;

[0150] Determine whether the overall priority score is greater than the preset threshold. If so, use the updated hierarchical view as the final hierarchical view and sort it in descending order according to its overall priority score.

[0151] The weighted cumulative calculation of the node weights of the same-named nodes involved in the updated candidate subsequences in the comparison results and iterative weight propagation graph is performed to obtain the sequence comprehensive score of each updated candidate subsequence, and the updated candidate subsequences are sorted in descending order according to the sequence comprehensive score.

[0152] The sorted final hierarchical view, the sorted updated alternative subsequence, and the updated main prompt sequence are structurally combined to generate search results.

[0153] This solution introduces a feedback-driven dynamic update mechanism, enabling search results to adaptively optimize as the cognitive state of the target object changes. Through the above technical solution, this application can effectively solve the problem of mismatch between search results and the cognitive ability of the target object in the prior art. The dynamic adjustment mechanism reduces the probability of erroneous searches. At the same time, the real-time update function of alternative subsequences provides multi-path search support for people with cognitive impairment, significantly improving the fault tolerance and smoothness of the search process.

[0154] The above describes how to generate search results by integrating views at various levels and their corresponding main prompt sequences and alternative subsequences. The following describes how to update the cognitive load mapping matrix, views at various levels, main prompt sequences, and alternative subsequences based on the comparison results, specifically including:

[0155] Calculate the deviation between the comparison results and the cognitive load mapping matrix, and update the cognitive load mapping matrix based on the deviation;

[0156] Based on the updated cognitive load mapping matrix, the views at each level are regenerated, and the comparison results are mapped to the regenerated views at each level. The classification results of nodes with the same name are adjusted, and the updated views at each level are generated.

[0157] Update the main prompt sequence and alternative subsequences based on the updated views at each level.

[0158] Here, bias refers to the difference between the adaptation strength recorded by the same node in the cognitive load mapping matrix and the comparison result.

[0159] This part has already been described in detail above, so I will not repeat it here.

[0160] This solution introduces a feedback-driven matrix update mechanism to capture changes in the cognitive state of the target object in real time. Furthermore, the dynamic updating of the main prompt sequence and alternative sub-sequences provides more flexible retrieval path selection for target objects with cognitive impairments. Through the above technical solutions, this application effectively solves the problem in existing technologies where the retrieval results do not match the cognitive abilities of the target object due to the lack of a dynamic adjustment mechanism. Through real-time feedback-driven matrix updates and hierarchical view reconstruction, the system can adaptively optimize the knowledge graph presentation strategy, reducing erroneous retrieval results caused by misunderstandings in target objects with cognitive impairments. Simultaneously, the dynamic update mechanism of the main prompt sequence and alternative sub-sequences provides multiple switchable paths for the search process, enhancing the system's fault tolerance and flexibility during target object interaction.

[0161] The above describes how the cognitive load mapping matrix, hierarchical views, main cue sequence, and alternative sub-sequences are updated based on the comparison results. The following describes how the updated cognitive load mapping matrix, updated hierarchical views, updated main cue sequence, and updated alternative sub-sequences are combined to generate search results, specifically including:

[0162] The updated cognitive load mapping matrix, the updated views at each level, the updated main cue sequence, the updated alternative subsequences, and the comparison results are used to generate an iterative weight propagation graph through a gating fusion mechanism.

[0163] Using the iterative weight propagation graph as a constraint, the updated hierarchical views, the updated main prompt sequence, and the updated alternative subsequences are integrated to generate search results.

[0164] Among them, the gating fusion mechanism refers to a processing mechanism used to weighted integrate and adaptively adjust multi-source data in a system;

[0165] An iterative weight propagation graph is a structured data representation used to implement multi-level weight adjustments and feedback cascading in a system.

[0166] This part has already been described in detail above, so I will not repeat it here.

[0167] This solution achieves dynamic weight updates through a gating fusion mechanism and constructs cross-level associations using an iterative weight propagation graph. This allows search results to adaptively adjust based on the target object's interactive behavior, effectively avoiding path selection bias caused by cognitive differences. Through the above technical solution, this application can dynamically optimize the generation path of search results based on the target object's real-time feedback. While maintaining the semantic relevance of the knowledge graph, it effectively adapts to the cognitive load variation characteristics of people with cognitive impairments, significantly reducing the probability of erroneous retrieval due to cognitive bias and improving the accuracy of interactive search in complex knowledge systems.

[0168] The above describes how to generate search results by integrating the updated cognitive load mapping matrix, updated views at each level, updated main cue sequence, and updated candidate subsequences. The following describes how to generate an iterative weight propagation graph by using a gating fusion mechanism with the updated cognitive load mapping matrix, updated views at each level, updated main cue sequence, updated candidate subsequences, and comparison results. Please refer to [link / reference]. Figure 4 , Figure 4 This is a schematic diagram of the process for generating an iterative weight propagation graph provided in an embodiment of this application. Generating the iterative weight propagation graph specifically includes:

[0169] The comparison results are used as gating signals and input into the updated cognitive load mapping matrix to obtain the updated node weights.

[0170] Based on the updated inter-view mapping channels, the updated node weights are propagated along the hierarchical view channel paths;

[0171] The propagated node weights are mapped back to the updated main cue sequence and the updated alternative subsequences to generate an iterative weight propagation graph.

[0172] Among them, the gating signal refers to the data input used to adjust and control the cognitive load mapping matrix;

[0173] Hierarchical view channel paths refer to the set of weight propagation paths established between nodes with the same name and their associated nodes within each hierarchical view.

[0174] This part has already been described in detail above, so I will not repeat it here.

[0175] This solution, through a gating fusion mechanism and an iterative weight propagation graph, can dynamically adjust the weight distribution based on real-time comparison results. Simultaneously, it utilizes hierarchical view channel paths to achieve cross-level information transmission, thereby improving the logical coherence and semantic consistency of search results. Through the above technical solutions, this application effectively solves the problem of insufficient search result adaptability caused by the lack of dynamic weight adjustment in existing technologies. By using structured combinations constrained by the iterative weight propagation graph, it ensures the collaborative optimization of the main prompt sequence and alternative sub-sequences, significantly improving the efficiency and accuracy of retrieval for people with cognitive impairments.

[0176] The above describes how the updated cognitive load mapping matrix, updated hierarchical views, updated main cue sequence, updated candidate subsequences, and comparison results are used to generate an iterative weight propagation graph through a gating fusion mechanism. The following describes how, using the iterative weight propagation graph as a constraint, the updated hierarchical views, updated main cue sequence, and updated candidate subsequences are integrated to generate search results, specifically including:

[0177] Using the updated cognitive load mapping matrix, comparison results, and iterative weight propagation graph as constraints, the updated hierarchical views and updated candidate subsequences are reordered and combined with the updated main prompt sequence in a structured manner to generate search results.

[0178] Reordering refers to the process of readjusting the presentation order of the updated hierarchical views and the updated alternative subsequences.

[0179] Structured assembly refers to the process of systematically integrating updated data of various types according to specific rules and hierarchical logic.

[0180] This part has already been described in detail above, so I will not repeat it here.

[0181] This solution, through a gating fusion mechanism and an iterative weight propagation graph, can adjust the weight distribution in real time based on interactive feedback, and generate multi-path retrieval results through reordering and structured combination, thereby improving its adaptability to people with cognitive impairment. Through the above technical solution, this application can generate structured retrieval paths with hierarchical logical relationships based on the dynamic update results of the cognitive load mapping matrix. While ensuring the priority of the main prompt sequence, it provides alternative sub-sequences for comprehension failure scenarios, effectively reducing the retrieval error rate for people with cognitive impairment, and enhancing the interpretability and traceability of retrieval results.

[0182] Example 2:

[0183] Please see Figure 5 An entity relationship extraction and search system based on knowledge graphs includes:

[0184] The data acquisition module is used to acquire the cognitive load mapping matrix and knowledge graph data of the target object;

[0185] The hierarchical view generation module is used to perform multi-scale hierarchical projection processing on knowledge graph data based on the cognitive load mapping matrix and node weights, generate hierarchical views containing nodes with the same name at different hierarchical positions, and establish mutual mapping channels between nodes with the same name in each hierarchical view.

[0186] The prompt sequence generation module is used to select nodes with the same name in priority order according to node weight and hierarchical position, and then generate the main prompt sequence of each level view;

[0187] Scan nodes with the same name that are not occupied by the main prompt sequence in the same level view, and calculate their consistency score in different level views by combining the mutual mapping channel;

[0188] Determine whether the consistency score is greater than the preset discrimination threshold; if so, generate candidate subsequences based on the corresponding nodes with the same name.

[0189] The search results generation module is used to combine the views at each level and their corresponding main prompt sequence and alternative sub-sequences to generate search results.

[0190] This embodiment has the same technical effects as Embodiment 1.

[0191] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. All data mentioned in this application have undergone normalization and other preprocessing to achieve dimensional uniformity during calculation.

[0192] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A knowledge graph-based entity relation extraction and search method, applied to knowledge graph-assisted retrieval for people with cognitive impairment, characterized in that... Includes the following steps: Obtain the cognitive load mapping matrix and knowledge graph data of the target object; Based on the cognitive load mapping matrix and node weights, the knowledge graph data is subjected to multi-scale hierarchical projection processing to generate hierarchical views containing nodes with the same name at different hierarchical positions, and a mutual mapping channel is established between nodes with the same name in each hierarchical view. The node weights are obtained by aggregating the cognitive load mapping matrix. Based on the node weight and hierarchical position, nodes with the same name are selected in priority order to generate the main prompt sequence of each level view; Scan the nodes with the same name in each level of view that are not occupied by the main prompt sequence, and calculate their consistency score in different level views in combination with the mutual mapping channel; Determine whether the consistency score is greater than a preset discrimination threshold; if so, generate a candidate subsequence based on the corresponding nodes with the same name. By combining the views at each level and their corresponding main prompt sequence and alternative sub-sequences, search results are generated.

2. The entity relation extraction and search method based on knowledge graph construction according to claim 1, characterized in that: Obtaining the cognitive load mapping matrix and knowledge graph data of the target object specifically includes: Acquire interaction data and knowledge graph data of the target object; The interactive data and the knowledge graph data are respectively subjected to isomorphic mapping processing; Perform bidirectional iterative projection, including forward and reverse mapping, on the isomorphic mapping result until the bidirectional iterative projection result converges, and output the cognitive load mapping matrix; Wherein, the rows of the cognitive load mapping matrix represent the nodes with the same name in the knowledge graph data, the columns of the cognitive load mapping matrix represent the interaction data, and the elements of the cognitive load mapping matrix characterize the compatibility strength between the target object and the nodes with the same name in the current environment.

3. The entity relation extraction and search method based on knowledge graph construction according to claim 2, characterized in that: Based on the cognitive load mapping matrix and node weights, the knowledge graph data undergoes multi-scale hierarchical projection processing to generate hierarchical views containing nodes with the same name at different hierarchical positions. Specifically, establishing mutual mapping channels between nodes with the same name in each hierarchical view includes: Cluster analysis is performed on the elements of the cognitive load mapping matrix to determine the set of projection scales, and a corresponding hierarchical threshold interval is generated for each projection scale. The normalized weight of the corresponding node with the same name is calculated based on the adaptation strength and the node weight, and compared with the threshold region of each level. Different classification strategies are applied to the nodes with the same name based on the comparison results, thereby generating a view of each level containing nodes with the same name at different level positions. Compare the nodes with the same name in different view levels one by one, and establish mutual mapping channels for nodes with the same name and the same node type.

4. The entity relation extraction and search method based on knowledge graph construction according to claim 3, characterized in that: Scanning nodes with the same name in each view level that are not occupied by the main prompt sequence, and calculating their consistency scores in different view levels in conjunction with the mutual mapping channel, specifically includes: For nodes with the same name in the main prompt sequence of each level view, scan for nodes with the same name in other level views that are not occupied by the corresponding main prompt sequence, and filter out the corresponding nodes with the same name from them according to the mutual mapping channel; Perform a difference analysis on the hierarchical threshold range of the same-name node and the corresponding hierarchical view containing the same-name node, and then calculate the consistency score by weighting it with the normalized weight.

5. The entity relation extraction and search method based on knowledge graph construction according to claim 2, characterized in that: By combining the views at each level and their corresponding main prompt sequence and alternative sub-sequences, the search results generated specifically include: After generating the main prompt sequence and the alternative sub-sequences, obtain the feedback data of the interaction data; The feedback data is compared with the main prompt sequence and the alternative sub-sequences, and the cognitive load mapping matrix, the views at each level, the main prompt sequence, and the alternative sub-sequences are updated based on the comparison results. The updated cognitive load mapping matrix, the updated views at each level, the updated main prompt sequence, and the updated alternative subsequences are combined to generate search results.

6. The entity relation extraction and search method based on knowledge graph construction according to claim 5, characterized in that: The update of the cognitive load mapping matrix, the views at each level, the main cue sequence, and the alternative sub-sequences based on the comparison results specifically includes: Calculate the deviation between the comparison result and the cognitive load mapping matrix, and update the cognitive load mapping matrix based on the deviation; Based on the updated cognitive load mapping matrix, the views at each level are regenerated, and the comparison results are mapped to the regenerated views at each level. The classification results of nodes with the same name are adjusted, and the updated views at each level are generated. The main prompt sequence and the alternative sub-sequences are updated based on the updated hierarchical views.

7. The entity relation extraction and search method based on knowledge graph construction according to claim 5, characterized in that: Based on the updated cognitive load mapping matrix, updated views at each level, updated main cue sequence, and updated alternative subsequences, the generated search results specifically include: The updated cognitive load mapping matrix, the updated views at each level, the updated main cue sequence, the updated alternative subsequences, and the comparison results are used to generate an iterative weight propagation graph through a gating fusion mechanism. Using the iterative weight propagation graph as a constraint, the updated hierarchical views, the updated main prompt sequence, and the updated alternative subsequences are integrated to generate search results.

8. The entity relation extraction and search method based on knowledge graph construction according to claim 7, characterized in that: The updated cognitive load mapping matrix, updated views at each level, updated main cue sequence, updated candidate subsequences, and comparison results are used through a gating fusion mechanism to generate an iterative weight propagation graph, specifically including: The comparison results are used as gating signals and input into the updated cognitive load mapping matrix to obtain the updated node weights. Based on the updated inter-view mapping channels, the updated node weights are propagated along the hierarchical view channel paths; The propagated node weights are mapped back to the updated main cue sequence and the updated alternative subsequences to generate an iterative weight propagation graph.

9. The entity relation extraction and search method based on knowledge graph construction according to claim 7, characterized in that: Using the iterative weight propagation graph as a constraint, the updated hierarchical views, the updated main prompt sequence, and the updated alternative subsequences are integrated to generate search results, specifically including: Using the updated cognitive load mapping matrix, comparison results, and the iterative weight propagation graph as constraints, the updated hierarchical views and updated candidate subsequences are reordered and combined with the updated main prompt sequence in a structured manner to generate search results.

10. An entity relation extraction and search system based on knowledge graphs, characterized in that, include: The data acquisition module is used to acquire the cognitive load mapping matrix and knowledge graph data of the target object; The hierarchical view generation module is used to perform multi-scale hierarchical projection processing on the knowledge graph data according to the cognitive load mapping matrix and node weights, generate hierarchical views containing nodes with the same name at different hierarchical positions, and establish mutual mapping channels between nodes with the same name in each hierarchical view. The prompt sequence generation module is used to select nodes with the same name in priority order according to the node weight and hierarchical position, and then generate the main prompt sequence of each level view; Scan nodes with the same name that are not occupied by the main prompt sequence in the same level view, and calculate their consistency score in different level views in combination with the mutual mapping channel; Determine whether the consistency score is greater than a preset discrimination threshold; if so, generate a candidate subsequence based on the corresponding nodes with the same name. The search results generation module is used to combine the views at each level and their corresponding main prompt sequence and alternative sub-sequences to generate search results.

Images