Decision information generation method, electronic device, and computer program product

Abstract

This disclosure provides a method for generating decision information, an electronic device, and a computer program product. The method for generating decision information includes: determining the feature values ​​of multi-dimensional node features of entity nodes in a knowledge graph; performing a weighted summation of the feature values ​​of the multi-dimensional node features to obtain an importance score for each entity node; and using entity nodes with importance scores greater than a threshold value as decision-dependent nodes in decision rules, and generating decision information about the knowledge graph according to the decision rules.

Description

Technical Field

[0001] This disclosure relates to the technical fields of knowledge graphs, and in particular to a method for generating decision information, an electronic device, and a computer program product. Background Technology

[0002] In the process of generating decision information based on input information using knowledge graphs, constructing knowledge graphs and analyzing the importance of each entity node are key technologies for realizing information extraction, knowledge reasoning, and intelligent decision-making.

[0003] Related technologies primarily rank the importance of entity nodes based on the number of direct connections or evaluate their pivotal position in the knowledge graph through iterative calculations. However, these solutions all rely on the topological characteristics of the knowledge graph, neglecting the semantic connotations and attribute values ​​of nodes and lacking intuitive business semantic support. Furthermore, they fail to fully consider the hierarchical inheritance relationships within the knowledge graph, leading to distorted importance assessments, affecting the credibility and acceptability of decision-making results, and hindering the in-depth application of knowledge graphs in decision support. Summary of the Invention

[0004] This disclosure provides a method for generating decision information, an electronic device, and a computer program product.

[0005] According to one aspect of this disclosure, a method for generating decision information is provided, comprising: determining feature values ​​of multi-dimensional node features of entity nodes in a knowledge graph; performing a weighted summation of the feature values ​​of the multi-dimensional node features to obtain an importance score of the entity node; and using entity nodes with importance scores greater than a score threshold as decision dependency nodes in a decision rule, and generating decision information about the knowledge graph according to the decision rule.

[0006] Based on one technical solution, the entity node is quantitatively evaluated from multiple dimensions by integrating structural features, content features, and hierarchical features. This significantly improves the semantic accuracy and business interpretability of the importance score, thereby generating a response solution that is more in line with the actual scenario and has greater decision-making value.

[0007] In some implementations, determining the feature values ​​of multi-dimensional node features of entity nodes in a knowledge graph includes: determining a first feature value of structural features, a second feature value of content features, and a third feature value of hierarchical features of the entity nodes based on the hierarchical relationship tree of the entity nodes. The hierarchical relationship tree can characterize the node type, descriptive text, and hierarchical structure information of the entity nodes. The structural features characterize the degree of association between the entity nodes and other nodes in the knowledge graph, the content features characterize the information richness of the entity nodes, and the hierarchical features characterize the importance of the entity nodes at the hierarchy in the knowledge graph.

[0008] According to one technical solution, based on the hierarchical relationship tree, the structure, content and hierarchical features are extracted in a unified manner, which realizes the systematic characterization of the multi-dimensional semantics and topological attributes of entity nodes, and significantly improves the completeness of feature representation and the accuracy of importance assessment.

[0009] In some implementations, determining a first feature value of the structural characteristics of the entity node includes: determining, based on the hierarchical structure information, a superior node and a subordinate node directly connected to the entity node; configuring an in-degree weight for the association between the superior node and the entity node based on the node type of the superior node, and configuring an out-degree weight for the association between the subordinate node and the entity node; and using the sum of the in-degree weight and the out-degree weight as the first feature value.

[0010] In some implementations, determining the first feature value of the structural features of the entity node further includes: determining whether the entity node has implicitly associated nodes based on the descriptive text of the entity node, wherein the implicitly associated nodes are not recorded as entity nodes in the knowledge graph, but have an association relationship with the entity node; when the implicitly associated nodes exist, determining the relationship weight between the implicitly associated nodes and the entity node; multiplying the relationship weight by a decay factor to obtain the target relationship weight; and calculating the sum of the in-degree weight, the out-degree weight, and the target relationship weight, and normalizing the sum to obtain the first feature value.

[0011] In some implementations, determining a second feature value of the content features of the entity node includes: preprocessing the descriptive text of the entity node to obtain target text that meets the format requirements; calculating the probability of occurrence of each character in the target text, wherein the probability of occurrence of a character is the ratio of the number of occurrences of the character to the total number of characters in the target text; calculating the probability of occurrence of the character using the information entropy formula to obtain the information entropy of the entity node, wherein the information entropy is proportional to the richness of the content of the descriptive text; and normalizing the information entropy to obtain the second feature value of the entity node.

[0012] In some implementations, the description text of the entity node is preprocessed, including: performing word segmentation on the description text to obtain multiple word segmentation units; and removing stop words, meaningless characters and symbols from the multiple word segmentation units to obtain target text, wherein the target text is a set of characters or words that can express complete semantics.

[0013] In some implementations, determining a third feature value of the hierarchical features of the entity node includes: based on the hierarchical structure information, counting a first number of associations traversed from the root node to the entity node, using the first number as the hierarchical depth of the entity node; determining the total number of subordinate nodes directly connected to the entity node; determining a second number of associations traversed from the entity node to the terminal associated node, and determining the complexity of the associated node based on the total number of subordinate nodes and the second number, wherein the terminal associated node is an entity node that has an association with the entity node and is located at the end of the knowledge graph; and calculating the product of the hierarchical depth, the total number of subordinate nodes, and the complexity, and normalizing the product to obtain the third feature value.

[0014] In some implementations, before determining the multi-dimensional node features of entity nodes in the knowledge graph, the process includes: extracting the node type, descriptive text, and hierarchical structure information of the entity nodes from the knowledge graph, wherein the hierarchical structure information represents the association relationship between the entity nodes and other nodes; and performing structured processing on the node type, descriptive text, and hierarchical structure information of the entity nodes to generate a hierarchical relationship tree, wherein all other nodes in the hierarchical relationship tree are associated with the entity nodes.

[0015] According to another aspect of this disclosure, an electronic device is provided, comprising: a memory storing execution instructions; and a processor executing the execution instructions stored in the memory, causing the processor to perform a decision information generation method according to any embodiment of this disclosure.

[0016] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the decision information generation method described in any embodiment of this disclosure. Attached Figure Description

[0017] The accompanying drawings illustrate exemplary embodiments of the present disclosure and, together with the description thereof, serve to explain the principles of the present disclosure. These drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification.

[0018] Figure 1 This is a schematic diagram illustrating an application scenario of the decision information generation method according to the embodiments of this disclosure.

[0019] Figure 2 This is a flowchart of a decision information generation method according to an embodiment of the present disclosure.

[0020] Figure 3This is a flowchart of the importance score acquisition process according to the embodiments of this disclosure.

[0021] Figure 4 This is a flowchart of the first feature value calculation process according to an embodiment of the present disclosure.

[0022] Figure 5 This is a flowchart of the second feature value calculation process according to an embodiment of the present disclosure.

[0023] Figure 6 This is a flowchart of the third feature value calculation process according to the embodiments of this disclosure.

[0024] Figure 7 This is a schematic diagram of a new knowledge graph according to an embodiment of the present disclosure.

[0025] Figure 8 This is a schematic block diagram of the structure of a decision information generation device according to an embodiment of the present disclosure.

[0026] Figure 9 This is a block diagram illustrating the structure of an electronic device according to one embodiment of the present disclosure. Detailed Implementation

[0027] The present disclosure will now be described in further detail with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are for illustrative purposes only and are not intended to limit the scope of the disclosure. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present disclosure are shown in the accompanying drawings.

[0028] It should be noted that, where there is no conflict, the embodiments and features described in this disclosure can be combined with each other. The technical solutions of this disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0029] In the process of generating decision-making information based on input data using knowledge graphs, the importance score of an entity node directly determines its weight and influence in the final decision outcome. Typically, the system sorts, filters, or aggregates candidate entities based on this score, thereby generating highly relevant and reliable decision recommendations. For example, when a user queries "educational resources in a certain area," the system will prioritize returning education-related entities with high importance scores (such as "a local university" or "local basic education policies"), using these as the basis for decision-making. Therefore, the importance score is not only a quantitative representation of node value but also a crucial bridge connecting the knowledge graph with higher-level decision-making logic.

[0030] However, current mainstream technologies for calculating importance scores have significant flaws: they primarily rely on the number of direct connections to nodes or iterative algorithms based on random walks, assessing "hubliness" solely from a topological perspective while completely ignoring the semantic connotations of nodes (such as descriptive text and node category) and their position within the knowledge graph. This approach leads to distorted importance assessments—for example, "education," as a high-level semantic node, might be underestimated due to its relatively few direct connections; "a certain street office," connected to numerous nodes due to its association with trivial matters, might be overestimated. The resulting decision-making information, while structurally sound, lacks semantic depth and business interpretability, making it difficult to adopt in practice and severely limiting the effectiveness of knowledge graphs in intelligent decision-making scenarios.

[0031] Therefore, this disclosure proposes a method for generating decision information.

[0032] Figure 1 This is a schematic diagram illustrating an application scenario of the decision information generation method according to the embodiments of this disclosure. For example... Figure 1 As shown, this application scenario may include a server 100 and a terminal device 200. The server 100 and the terminal device 200 can interact with each other via a network connection. The server 100 can be a cloud server or a physical server, and the terminal device 200 can be a smart device such as a computer, mobile phone, or tablet. The server 100 can receive requests from the terminal device 200 and run the decision information generation method disclosed herein.

[0033] Figure 2 This is a flowchart of a decision information generation method according to an embodiment of this disclosure. Figure 2 As shown, this disclosure proposes a decision information generation method M200, which calculates importance scores by integrating multi-dimensional node features. This solves the problem of distorted importance assessment caused by traditional solutions relying solely on topological structure, improves the semantic relevance and business interpretability of decision information, and enhances the practicality of knowledge graphs in decision-making.

[0034] Step S210: Determine the feature values ​​of the multi-dimensional node features of entity nodes in the knowledge graph.

[0035] A knowledge graph is a data model that describes knowledge information in a real-world scenario. It typically consists of three parts: entity nodes, the relationships between entity nodes, and node feature information (such as node identifiers, descriptive text of entity nodes, and other structured or unstructured data). By integrating discrete knowledge information into an interconnected knowledge network, knowledge graphs can not only accurately depict the complex relationships in real-world scenarios but also support various applications such as semantic search and decision information generation.

[0036] Entity nodes are uniquely identifiable objects with clear semantics in a knowledge graph. These objects can be concrete real-world transactions (such as a university, a city, or a specific mobile phone model) or abstract concepts and events, such as "higher education" or "date of birth." Each entity node has a unique node identifier, node type (e.g., city), and descriptive text.

[0037] Multi-dimensional node features refer to a set of features systematically characterizing entity nodes in a knowledge graph from multiple dimensions. These features aim to more comprehensively and accurately reflect the semantic connotation, structural position, and business value of nodes. These dimensions typically include structural features, semantic features, and hierarchical features. By integrating these heterogeneous features, multi-dimensional node features can effectively overcome the limitations of single topological indicators, providing a more reliable and interpretable foundation for subsequent node importance assessment and intelligent decision-making.

[0038] Among them, structural features characterize the degree of association between entity nodes and other nodes in the knowledge graph, content features characterize the information richness of entity nodes, and hierarchical features characterize the importance of entity nodes in the hierarchy of the knowledge graph. Multi-dimensional node features can be further expanded to include other dimensions based on the actual scenario requirements reflected by the knowledge graph; no restrictions are imposed here.

[0039] It's important to note that the feature values ​​for each dimension in the multi-dimensional node features are the quantified results of the entity node in its corresponding dimension. These values ​​have undergone normalization to be mapped to the same numerical range, such as 0 to 1, to eliminate dimensional differences and ensure the comparability and consistency of features across different dimensions. The calculation process for each dimension's feature values ​​will be detailed later and will not be elaborated upon here.

[0040] Step 220: Perform a weighted summation of the feature values ​​of the multi-dimensional node features to obtain the importance score of the entity node.

[0041] After obtaining the feature values ​​of nodes in each dimension, weights matching the actual decision-making objectives are assigned to the different node features, and the feature values ​​of each dimension are weighted and summed to calculate the importance score of the entity node. The importance score is a quantitative result of the overall importance of the entity node in the knowledge graph. It integrates multi-dimensional information such as the structure and hierarchy of the entity node in the knowledge graph, as well as the semantics of its own descriptive text. It reflects the relative importance of different features in specific application scenarios and can more accurately and interpretably reflect the overall value of the entity node in the knowledge graph, providing a reliable basis for subsequent node importance ranking and decision information generation.

[0042] Step 230: Entity nodes with importance scores greater than the score threshold are used as decision dependency nodes in the decision rules, and decision information about the knowledge graph is generated according to the decision rules.

[0043] The score threshold is a standard for measuring the importance of entity nodes, used to distinguish which entity nodes are worth including in the decision-making process. The score threshold is set according to the actual situation and is not restricted here. Only entity nodes whose importance score is higher than the score threshold will be retained as important reference information for generating decision-making information.

[0044] Decision-dependent nodes refer to key entity nodes in a knowledge graph whose importance score exceeds a preset threshold. They represent the most relevant and critical knowledge content in the knowledge graph and are the core input elements of decision rules. Decision rules are logical judgment mechanisms defined based on the semantic relationships and topological structure of the knowledge graph. By analyzing decision-dependent nodes and their associated paths, decision information with business value (such as risk warnings, related recommendations, or knowledge completion) can be derived, realizing the transformation from high-importance nodes to decision information.

[0045] Decision information refers to solutions to real-world scenarios reflected in the knowledge graph, such as answers to questions, risk warnings, recommended solutions, or situation summaries. Decision information can effectively support users' judgments and actions.

[0046] Figure 3 This is a flowchart illustrating the process of obtaining importance scores according to the embodiments of this disclosure. The following is in conjunction with... Figure 3 The process of obtaining the importance score is explained.

[0047] In step 301, the hierarchical relationship tree of entity nodes is determined based on the knowledge graph.

[0048] Specifically, the node type, description text, and hierarchical structure information of entity nodes are extracted from the knowledge graph; and the node type, description text, and hierarchical structure information of entity nodes are structured to generate a hierarchical relationship tree, in which all nodes are associated with entity nodes.

[0049] A hierarchical relationship tree is a structured data record centered on a target entity node, documenting the hierarchical structure between that entity and all its related (direct and indirect) superior and subordinate nodes. This tree clearly displays the hierarchical relationships, such as superiority, affiliation, and inclusion, of entities within a knowledge system through a parent-child hierarchy. This not only facilitates efficient extraction of hierarchical structure information but also supports rapid reading of additional information such as the identifiers, descriptive texts, and relationships between nodes.

[0050] Each knowledge graph can generate multiple hierarchical relationship trees because it is itself a network structure containing multiple types of relationships and complex connections. Depending on different analysis needs, multiple tree structures centered on specific entities and reflecting different levels of semantics can be extracted from the same graph.

[0051] Below is a hierarchical relationship tree for a scenario: { "nodes": [ { "id": "Beijing", "relation": "city information", "description": "An important city in China, located in North China, and is China's economic center." "children": [ { "id": "education", "relation": "domain category", "description": "Beijing boasts numerous universities and research institutions, making it one of the most important academic and research centers in China and even globally." }, { "id": "Technological Development", "relation": "domain category", "description": "Beijing is home to a large number of research institutions and high-tech enterprises, and is the national center for science and technology innovation." } ] } ] } When constructing a hierarchical relationship tree with "Beijing" as the central node, since there are no parent nodes related to "Beijing" in the knowledge graph, Beijing is treated as a single entity node, clearly organizing the core domain information related to Beijing. "Beijing" expands downwards through the relationship "City Information," resulting in two first-level child nodes: "Education" and "Technological Development," representing two key dimensions of Beijing's social functions. Each child node is accompanied by a clear node identifier (id), node type (relation), and description, explaining its connotation and value. Structurally, the tree adopts a top-down hierarchical approach, simplifying complex real-world knowledge into a parent-child hierarchical relationship: the top level represents specific city entities, and the middle level represents abstract domain categories. This organizational form not only preserves the semantic connections between entities but also explicitly expresses the logical relationships of "belonging" or "covering."

[0052] In step 302, based on the hierarchical relationship tree, the first feature value of the structural feature, the second feature value of the content feature, and the third feature value of the hierarchical feature of the entity node are calculated.

[0053] Figure 4 This is a flowchart illustrating the structural feature calculation process according to an embodiment of this disclosure. The following is in conjunction with… Figure 4 The calculation process of the first eigenvalue is explained.

[0054] In step 401, based on the hierarchical structure information, the superior and subordinate nodes directly connected to the entity node are determined.

[0055] Hierarchical structure information represents the relationships between entity nodes and other nodes. It originates from the hierarchical relationship tree constructed in the knowledge graph based on directed associations (such as hierarchical or belonging semantic relationships like "belongs to" or "located in"). Because these associations are directional, they can clearly express the semantics of "who contains whom" or "who belongs to whom." Therefore, by traversing the adjacent edges of the entity in the hierarchical relationship tree, we can identify the parent node (i.e., the superior node) and the child nodes (i.e., the subordinate nodes) that the entity points to. This process is essentially a structured representation of the local topology in the knowledge graph using the hierarchical relationship tree, thereby accurately extracting the direct hierarchical dependencies of the entity in the knowledge system, providing a foundation for subsequent structural feature calculations.

[0056] It should be noted that the "directly connected parent node" referred to here is the parent node that directly points to the current entity node, and the "directly connected child node" is the child node that the current entity node directly points to. Both are limited to direct associations and do not include indirect or cross-level connections. The number of parent and child nodes is unlimited and may be zero or more. For example, if no node in the knowledge graph directly points to "Beijing," then when "Beijing" is the current entity node being analyzed, its number of parent nodes is 0; however, if "Beijing" directly points to the nodes "Education" and "Science and Technology Development," then its number of child nodes is 2. Even if "Education" further points to "University A" and "University B," these nodes have no direct association with "Beijing," and therefore are not included when calculating the direct child nodes of "Beijing." This definition ensures the locality and accuracy of hierarchical structure information, reflecting only the hierarchical dependencies of entity nodes within a single layer of neighborhood.

[0057] Furthermore, in step 402, in-degree weights are configured for the parent node and out-degree weights are configured for the child node.

[0058] If a parent node exists, an in-degree weight is assigned to each parent node; if there is no parent node, no in-degree weight needs to be configured. If a child node exists, an out-degree weight is assigned to each child node. The in-degree weights of parent nodes can be different, and the out-degree weights of child nodes can also be different, and there are no restrictions such as them adding up to 1. These weights are only used to quantify the relative influence of the current entity node and its neighboring nodes in the hierarchical structure.

[0059] The in-degree weight represents the semantic support or attribution strength of the parent node to the current entity node, reflecting the importance of being "included" or "belonging to"; the out-degree weight represents the breadth of coverage or dominance of the current entity node over its subordinate nodes, reflecting its ability to encompass its children as a parent or container. Together, they characterize the bidirectional structural influence of the entity in the hierarchical relationship, providing a foundation for subsequent comprehensive calculation of node importance.

[0060] The specific values ​​of in-degree and out-degree weights are set according to the needs of the actual application scenario and are not subject to uniform restrictions. The basis for weight configuration may include factors such as business importance, semantic relevance, expert experience, statistical frequency, or domain rules. For example, in an education decision-making scenario, "education," as a lower-level node, may be assigned a higher out-degree weight (e.g., 0.5) to highlight its core position; while in a technology assessment scenario, "technology development" may receive a higher weight. The existence of a higher-level node determines whether the in-degree weight is zero, and similarly, the existence of a lower-level node determines the allocation of the out-degree weight. In short, the weights are configurable parameters designed to flexibly reflect the actual meaning and influence intensity of hierarchical relationships in different contexts.

[0061] In some implementations, the sum of the in-degree and out-degree weights can be used as the first feature value that characterizes the structural features of entity nodes. However, if the entity's description text contains implicitly related nodes that are not explicitly modeled in the knowledge graph, a semantic parsing mechanism needs to be introduced to identify and quantify such implicit relationships from the description text and include them in the calculation of the first feature value. Conversely, if the description text does not contain such implicit information, no extension is needed, and the weight sum corresponding to the explicit hierarchical relationship can be used directly.

[0062] In step 403, it is determined whether the description text of the entity node contains implicitly related nodes.

[0063] Implicitly related nodes are nodes that are not recorded as entity nodes in the knowledge graph, but have a relationship with entity nodes. For example, in the description "Beijing is a historically significant economic center," although "economy" is not represented as an entity node in the graph, it has a clear semantic relationship with "Beijing," thus constituting an implicitly related node. The judgment process typically involves analyzing the descriptive text using natural language processing techniques (such as named entity recognition, keyword extraction, or semantic role labeling) to identify the potential concepts mentioned, and then verifying whether they belong to missing but relevant semantic units in the knowledge graph using domain ontology or dictionaries.

[0064] If the description text is found to contain implicitly related nodes, then step 404 is executed to determine the relation weights of the implicitly related nodes.

[0065] The value of the relation weight depends on the analysis requirements of the implicitly associated node and the current actual scenario, while the type of relation weight depends on the hierarchical relationship between the implicitly associated node and the entity node being analyzed. If the implicit node belongs to a higher-level concept or category of the current entity, it is considered a higher-level node, and its relation weight uses in-degree weight; if it is a lower-level instance, component, or specific manifestation of the current entity (such as "economy" in relation to "Beijing"), it is considered a lower-level node, and its relation weight uses out-degree weight. This weight value can be quantified based on semantic strength, contextual importance, or preset rules, thereby incorporating implicit knowledge into structural feature calculations and enhancing the completeness of node representation and decision relevance.

[0066] In step 405, the relation weight is multiplied by the decay factor to obtain the target relation weight.

[0067] The introduction of the attenuation factor aims to reflect the uncertainty or weakened information inherent in implicitly related nodes due to their lack of explicit modeling. Compared to explicitly established entity relationships in a knowledge graph, associations inferred solely from textual descriptions are typically less reliable and have weaker semantic strength. Therefore, the original relation weights (in-degree or out-degree weights) of implicitly related nodes are multiplied by an attenuation coefficient less than 1 (e.g., 0.6 or 0.8), which can be configured based on the application scenario, text confidence, or domain experience. This attenuation mechanism preserves the semantic supplementary value of implicit associations while preventing them from excessively influencing the calculation of overall structural features, thus achieving a reasonable balance between explicit knowledge and implicit clues.

[0068] In step 406, the in-degree weight, out-degree weight, and target relation weight are added together, and the addition result is normalized to obtain the first feature value.

[0069] When calculating structural features, the in-degree and out-degree of entity nodes can be further considered: by analyzing the hierarchical relationship tree, the number of parent nodes directly connected to the node (i.e., in-degree) is counted, reflecting the degree to which it is depended upon or belongs to by other nodes; at the same time, the number of child nodes directly connected to the node (i.e., out-degree) is counted, reflecting the breadth and influence of its external connections. In-degree and out-degree together characterize the distribution of node connections in the graph, which can more comprehensively reflect its structural position in the network and provide basic data support for the subsequent construction of structural features.

[0070] The calculation of the first eigenvalue of structural features is mainly based on the in-degree and out-degree weights in the hierarchical relationship. However, the in-degree and out-degree of a node can still be retained as auxiliary structural information. Although this type of information does not directly participate in the calculation of the first eigenvalue, it can be used to evaluate the connectivity, centrality, or potential influence of a node when analyzing strategy information or performing contextual understanding, thereby providing richer reference dimensions for decision-making.

[0071] Figure 5 This is a flowchart illustrating the calculation process of the content features according to the embodiments of this disclosure. The following is in conjunction with... Figure 5 The calculation process for the second eigenvalue is explained.

[0072] In step 501, the description text is preprocessed to obtain the target text.

[0073] The preprocessing process includes: segmenting the descriptive text to obtain multiple segmentation units; and removing stop words, meaningless characters and symbols from the multiple segmentation units to obtain the target text, which is a collection of characters or words that can express complete semantics.

[0074] Descriptive text refers to the natural language content attached to entity nodes in a knowledge graph, which is used to explain the basic attributes, functions, backgrounds or features of the entity. For example, "Beijing is the economic center of China". Such text usually comes from encyclopedias, databases or manual annotations, and is an important bridge connecting structured knowledge and human-readable semantics.

[0075] The purpose of preprocessing is to transform the original descriptive text into a cleaner, more compact and semantically focused target text for subsequent feature extraction or semantic analysis. Specifically, it includes: First, perform word segmentation on the text, that is, divide continuous sentences into individual word segmentation units, such as the words "Beijing", "is", "China", "of", "economic", "center" in the above example. Then, remove the stop words, such as high-frequency but semantically meaningless function words like "of", "is", "and", etc., including meaningless characters such as line breaks, tab characters, garbled characters, etc., and irrelevant symbols such as punctuation marks ",", ".", etc.

[0076] For example, after preprocessing the original description "Beijing is the economic and cultural center of China.", the target text may be {"Beijing", "China", "economic", "cultural", "center"}. This process effectively removes noise, retains the core semantic units, and significantly improves the accuracy and efficiency of subsequent semantic calculations.

[0077] In step 502, calculate the character occurrence probability of each character in the target text. Among them, the character occurrence probability is the ratio of the number of occurrences of a character to the total number of characters in the target text.

[0078] In step 503, call the information entropy formula to calculate the character occurrence probability and obtain the information entropy.

[0079] The information entropy formula is , where Xi represents character i, P(Xi) represents the character occurrence probability of character i, represents the information entropy of the target text.

[0080] Information entropy is an indicator to measure the uncertainty of character distribution in a text: when the character distribution is more uniform, the entropy value is higher, indicating a more complex semantic structure and a higher information density; on the contrary, if some characters appear frequently, the entropy value is lower, indicating that the content is more concentrated or redundant. For example, in "Subject learning is very important", the character "学" appears repeatedly, and its information entropy is lower than that of a text like "The educational strength of A University is strong" with a more balanced character distribution. By calculating the information entropy, the semantic richness and expression diversity of the text can be quantitatively described, providing a basis for subsequent judgments of text importance, keyword extraction or core concept recognition, and can also be combined with other features for node semantic complexity evaluation.

[0081] In step 504, normalize the information entropy to obtain the second feature value.

[0082] The calculated information entropy is normalized (e.g., by dividing by the theoretical maximum entropy or mapping to the [0,1] interval), transforming it into standardized content features. The purpose of normalization is to eliminate dimensional differences caused by varying text lengths or character set sizes, making information entropy values ​​comparable. This approach is consistent with the normalization of the first and third feature values. All node feature values ​​are uniformly scaled to the same numerical range, preventing a single feature value from dominating subsequent weighted calculations due to its excessive magnitude, and ensuring fair fusion of multi-source heterogeneous features. This improves the stability, rationality, and cross-scenario applicability of entity node importance score calculation.

[0083] Figure 6 This is a flowchart of the hierarchical feature calculation process according to the embodiments of this disclosure. The following is in conjunction with... Figure 6 The calculation process of the third eigenvalue is explained.

[0084] In step 601, the first number of associations traversed from the root node to the entity node is counted, and the first number is used as the hierarchical depth of the entity node.

[0085] The first quantity refers to the number of relationships (directed edges) traversed from the root node of the knowledge graph along the hierarchical path to reach the current entity node. This quantity directly reflects the nesting level of the current entity in the hierarchical structure and is defined as the hierarchical depth of the entity node. For example, if the root node "Beijing" is directly connected to "Education," and "Education" is then connected to "University A," then when "University A" is the currently viewed entity node, its first quantity is 2, and its hierarchical depth is 2, indicating that it is located at the second level below the root node. The greater the hierarchical depth, the more specific and lower-level the entity is; the smaller the depth, the closer it is to the top-level abstract concept. Since "Beijing" is the root node, its hierarchical depth is 0.

[0086] In step 602, the total number of subordinate nodes directly connected to the entity node is determined.

[0087] This metric represents the total number of direct subordinate nodes pointed to by the current entity node, i.e., the number of child nodes that the node, as a parent node, possesses in the hierarchical relationship tree. This value reflects the breadth or coverage of the entity's branches in the knowledge structure: the more directly connected subordinate nodes, the richer the specific instances, subcategories, or sub-fields it covers. For example, "Beijing" directly connects to two subordinate nodes, "Education" and "Technological Development," so its total number of subordinate nodes is 2. Even if "Education" further connects to nodes such as "University A," these are indirect connections and are not included in the total number of direct subordinate nodes of the current entity. This metric, as an important component of hierarchical characteristics, is used to measure the structural expansion capability of an entity.

[0088] In step 603, a second number of associations traversed from the entity node to the end associated node is determined.

[0089] The second quantity refers to the maximum number of relationships (i.e., directed edges) traversed from the current entity node down the hierarchical relationship to the farthest terminal node. A terminal node is an entity node that is related to an entity node and is located at the end of the knowledge graph; that is, a node that no longer has any lower-level nodes in the hierarchical relationship tree. It represents the most concrete and indivisible instance or concept in the knowledge system. For example, if "Beijing" → "Education" → "University A" constitutes a path, and "University A" has no lower-level nodes, then "University A" is the terminal node. Reaching this terminal node from "Beijing" requires 2 relationships, hence the second quantity is 2. If multiple downward paths exist, the number of edges on the longest path is taken as the second quantity. This metric reflects the maximum hierarchical span that the current entity node can extend to, characterizing its potential depth and refinement capability within the knowledge structure, and is one of the key features for measuring hierarchical complexity.

[0090] In step 604, the complexity of the associated nodes is determined based on the total number of subordinate nodes and the second number.

[0091] The complexity of an entity node's associated nodes is jointly determined by the total number of its subordinate nodes (i.e., the number of direct child nodes) and a second factor (i.e., the maximum path length reaching the final node). This comprehensive measure assesses the breadth and depth of the node's branches within the hierarchical structure. A larger total number of subordinate nodes indicates a wider range of subclasses or instances covered by the entity; a larger second factor indicates a longer and deeper knowledge chain. The combination of these factors reflects the richness and structure of the knowledge organized by the node—for example, "education" with only two child nodes but extending to three levels (e.g., "education → higher education → university A") has a higher complexity than a node with only one child node and one level. This complexity, as part of the hierarchical characteristics, provides a quantitative basis for evaluating an entity's expressive power and information capacity within the knowledge graph.

[0092] Complexity is generally calculated by multiplying the total number of subordinate nodes and the second number. The result of multiplication can more reasonably reflect the comprehensive complexity of a node in the knowledge system, which is both "widely covered" and "deeply layered". Of course, depending on specific needs, addition or other weighted combination methods can also be used, but the product form can better avoid the dominance of a single dimension and effectively represent the richness of structural information.

[0093] In step 605, the product of the hierarchy depth, the total number of lower-level nodes, and the complexity is calculated and normalized to obtain the third feature value.

[0094] Multiplying the hierarchical depth of an entity node, the total number of its subordinate nodes, and its complexity yields a raw value that comprehensively reflects its position, breadth, and structural complexity within the hierarchical structure. This product is then normalized (e.g., mapped to the [0,1] interval) to obtain a standardized third feature value. Normalization ensures that this feature value is on the same scale as the first and second feature values, facilitating subsequent multi-feature value fusion and fair calculation of importance scores.

[0095] In step 303, a weighted summation is performed on each feature value.

[0096] In the feature fusion stage, a weighted summation mechanism is used to comprehensively evaluate multi-dimensional node features. For example, the structural feature weight is set to 0.4 to reflect the node's connectivity and pivotal position in the graph; the content feature weight is set to 0.3 to reflect the information density and semantic value carried by the node's descriptive text; and the hierarchical feature weight is set to 0.3 to represent the node's structural position and complexity within the knowledge system. Through this reasonable weighting, feature values ​​from different sources are fused according to their relative importance in the actual decision-making scenario, achieving a comprehensive, balanced, and interpretable quantitative assessment of the importance of entity nodes, thereby improving the accuracy and credibility of subsequent decision information generation. Of course, the weights of each feature can be set according to the analytical needs of the actual scenario; the above is merely an example.

[0097] In step 304, the importance score is obtained. The importance score is the result of the weighted summation in step 303.

[0098] Finally, entity nodes with importance scores greater than the score threshold are used as decision-dependent nodes for the decision rules. Decision information about the knowledge graph is generated according to the decision rules. The decision information is about the solutions to the actual scenarios reflected by the knowledge graph.

[0099] These high-importance entity nodes represent the most relevant, semantically rich, and structurally critical elements in the knowledge graph that are relevant to the current task. The decision information generated based on decision rules that use them as decision-dependent nodes provides actionable and evidence-based strategic recommendations for the actual scenarios reflected in the knowledge graph (such as urban governance, education planning, or technology layout).

[0100] In some implementations, a display component showing the importance scores of entity nodes is overlaid on the existing knowledge graph to form a new knowledge graph.

[0101] Figure 7 This is a schematic diagram of a new knowledge graph according to an embodiment of this disclosure. For example... Figure 7As shown, a new knowledge graph structure with "Beijing" as the root node is presented. "Beijing" serves as the central entity, connecting multiple child nodes through different types of relationships, forming a hierarchical and semantic knowledge network. Circles represent entity nodes, such as "Economy and Technology," "History and Culture," "Zhongguancun," "Artificial Intelligence," "Peking Opera," and "Forbidden City," which are specific objects or concepts in the knowledge graph. Labels on the connecting lines (such as "Domain," "Development Area," "Key Technology," "Traditional Art Forms," ​​and "Representative Buildings") indicate the node type, i.e., the category of the relationship between entities. For example, "Economy and Technology" connects to "Beijing" through the "Domain" relationship and points to "Artificial Intelligence" through "Key Technology," reflecting a dual structure of hierarchy and attributes. Furthermore, the pop-up box on the right side of the diagram shows that when the "Economy and Technology" node is hit, it dynamically displays its node identifier as "Economy and Technology," its descriptive text as "Including service industries, high-tech industries, etc., reflecting Beijing's status as a national science and technology innovation center," and its calculated importance score of 127.2%. This knowledge graph, which visualizes importance scores, not only clearly presents the knowledge structure but also allows users to intuitively understand the value weight of each node, facilitating the rapid identification of key information and improving the interactive experience and decision-making efficiency.

[0102] Figure 8 This is a schematic block diagram of the decision information generation apparatus according to an embodiment of the present disclosure. Figure 8 As shown, the decision information generation device 800 proposed in this disclosure includes: a feature calculation module 810, used to determine the feature values ​​of multi-dimensional node features of entity nodes in a knowledge graph; an importance score calculation module 820, used to perform weighted summation of the feature values ​​of multi-dimensional node features to obtain the importance score of entity nodes; and a strategy information analysis module 830, used to take entity nodes with importance scores greater than a score threshold as decision dependency nodes in decision rules, and generate decision information about the knowledge graph according to the decision rules.

[0103] The decision information generation device 800 disclosed herein can be in the form of computer software, and each module of the decision information generation device 800 can be in the form of computer software modules.

[0104] The various modules of the decision information generation device 800 disclosed herein are designed to implement the various steps of the decision information generation method. Their execution principles and steps can be referred to the preceding text and will not be repeated here.

[0105] Figure 9 This is a schematic block diagram of an electronic device according to one embodiment of the present disclosure. Figure 9As shown, this disclosure also provides an electronic device 1000, including: a processor 1200 and a memory 1300, the memory 1300 storing execution instructions; the processor 1200 executes the execution instructions stored in the memory 1300, causing the processor 1200 to execute a decision information generation method.

[0106] The hardware architecture of the electronic device 1000 can be implemented using a bus architecture. The bus architecture can include any number of interconnect buses and bridges, depending on the specific application of the hardware and overall design constraints. Bus 1100 connects various circuits, including one or more processors 1200, memory 1300, and / or hardware modules. Bus 1100 can also connect various other circuits 1400, such as peripheral devices, voltage regulators, power management circuits, external antennas, etc.

[0107] Bus 1100 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, this diagram uses only one connection line, but this does not imply that there is only one bus or one type of bus.

[0108] This disclosure also provides a readable storage medium storing a computer program that, when executed by a processor, is used to implement the methods described above. A "readable storage medium" can be any means capable of containing, storing, communicating, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples of a readable storage medium include: an electrical connection with one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable read-only memory (CDROM), etc.

[0109] This disclosure also provides a computer program product, the methods of which can be implemented wholly or partially through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented wholly or partially as a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed, all or part of the processes or functions of this disclosure are performed.

[0110] Computer programs or instructions can be stored in a readable storage medium or transferred from one readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The readable storage medium can be any available medium capable of access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a digital video optical disc; or a semiconductor medium, such as a solid-state drive. The computer-readable storage medium can be a volatile or non-volatile storage medium, or it can include both volatile and non-volatile types of storage media.

[0111] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, electronic devices, readable storage media, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0112] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0113] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0114] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0115] In the description of this specification, the references to terms such as "one embodiment / mode," "some embodiments / modes," "example," "specific example," or "some examples," etc., refer to specific features, structures, or characteristics described in connection with that embodiment / mode or example, which are included in at least one embodiment / mode or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment / mode or example. Moreover, the specific features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments / modes or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments / modes or examples described in this specification, as well as the features of different embodiments / modes or examples.

[0116] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0117] Those skilled in the art should understand that the above embodiments are merely for illustrating the present disclosure and are not intended to limit the scope of the disclosure. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of the present disclosure.

Claims

1. A method for generating decision information, characterized in that, include: Determine the feature values ​​of multi-dimensional node features of entity nodes in a knowledge graph; The importance score of the entity node is obtained by weighted summation of the feature values ​​of the multi-dimensional node features. as well as Entity nodes with importance scores greater than a score threshold are used as decision dependency nodes in the decision rules, and decision information about the knowledge graph is generated according to the decision rules.

2. The decision information generation method according to claim 1, characterized in that, Determine the feature values ​​of multi-dimensional node features of entity nodes in a knowledge graph, including: Based on the hierarchical relationship tree of the entity nodes, a first feature value of the structural features, a second feature value of the content features, and a third feature value of the hierarchical features of the entity nodes are determined. The hierarchical relationship tree can characterize the node type, descriptive text, and hierarchical structure information of the entity nodes. The structural features characterize the degree of association between the entity nodes and other nodes in the knowledge graph, the content features characterize the information richness of the entity nodes, and the hierarchical features characterize the importance of the entity nodes at the level of the knowledge graph.

3. The decision information generation method according to claim 2, characterized in that, Determining the first feature value of the structural features of the entity node includes: Based on the hierarchical structure information, determine the superior and inferior nodes that are directly connected to the entity node; Configure in-degree weights for the association between the parent node and the entity node, and configure out-degree weights for the association between the lower-level node and the entity node based on the node type of the lower-level node; and The sum of the in-degree weight and the out-degree weight is used as the first feature value.

4. The decision information generation method according to claim 3, characterized in that, Determining the first feature value of the structural features of the entity node further includes: Based on the description text of the entity node, determine whether the entity node has implicitly associated nodes, wherein the implicitly associated nodes are not recorded as entity nodes in the knowledge graph, but have an association relationship with the entity node; When the implicitly associated node exists, determine the relationship weight between the implicitly associated node and the entity node; Multiply the relationship weight by the decay factor to obtain the target relationship weight; and Calculate the sum of the in-degree weight, the out-degree weight, and the target relation weight, and normalize the sum to obtain the first feature value.

5. The decision information generation method according to claim 2, characterized in that, Determining the second feature value of the content feature of the entity node includes: The description text of the entity node is preprocessed to obtain target text that meets the format requirements; The probability of occurrence of each character in the target text is calculated, where the probability of occurrence is the ratio of the number of times the character appears to the total number of characters in the target text. The probability of the character's occurrence is calculated using the information entropy formula to obtain the information entropy of the entity node. This information entropy is directly proportional to the richness of the descriptive text. The information entropy is normalized to obtain the second feature value of the entity node.

6. The decision information generation method according to claim 5, characterized in that, Preprocessing the description text of the entity nodes includes: The descriptive text is segmented to obtain multiple segmentation units; and Stop words, meaningless characters, and symbols are removed from the multiple word segmentation units to obtain the target text, which is a set of characters or words that can express complete semantics.

7. The decision information generation method according to claim 2, characterized in that, Determining the third feature value of the hierarchical features of the entity node includes: Based on the hierarchical structure information, a first number of associations traversed from the root node to the entity node is counted, and the first number is used as the hierarchical depth of the entity node. Determine the total number of subordinate nodes directly connected to the entity node; Determine a second number of associations traversed from the entity node to the terminal associated node, and determine the complexity of the associated node based on the total number of subordinate nodes and the second number, wherein the terminal associated node is an entity node that has an association with the entity node and is located at the end of the knowledge graph; and Calculate the product of the level depth, the total number of lower-level nodes, and the complexity, and normalize the product to obtain the third feature value.

8. The decision information generation method according to claim 1, characterized in that, Before determining the multi-dimensional node features of entity nodes in a knowledge graph, the following steps are included: Extract the node type, descriptive text, and hierarchical structure information of the entity nodes from the knowledge graph, wherein the hierarchical structure information represents the association relationship between the entity nodes and other nodes; and The node type, description text, and hierarchical structure information of the entity node are processed in a structured manner to generate a hierarchical relationship tree, in which all nodes are associated with the entity node.

9. An electronic device, characterized in that, include: The memory stores execution instructions; as well as A processor that executes the execution instructions stored in the memory, causing the processor to perform the decision information generation method according to any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the decision information generation method according to any one of claims 1 to 8.

Images