A cross-platform intelligent publishing method based on a knowledge graph

Abstract

The application discloses a cross-platform intelligent publishing method based on a knowledge graph, which comprises the following steps: acquiring publishing rule information, column structure information and field specification information of multiple publishing platforms, and constructing a unified publishing knowledge graph; generating a platform perspective knowledge subgraph based on the unified publishing knowledge graph; constructing a decision-oriented GraphRAG structure for publishing decision based on the platform perspective knowledge subgraph; performing cross-platform publishing feasibility reasoning in the platform perspective knowledge subgraph based on the decision-oriented GraphRAG structure; performing reverse reasoning based on the platform perspective knowledge subgraph and the decision-oriented GraphRAG structure; generating a cross-platform publishing decision result and performing a publishing operation based on the publishing feasibility result and the counterfactual adjustment result, and obtaining a publishing execution result. The application adopts the knowledge graph and the decision-oriented GraphRAG structure to realize cross-platform intelligent publishing decision and execution.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph and intelligent information processing technology, and in particular to a cross-platform intelligent publishing method based on knowledge graph. Background Technology

[0002] With the increasing number of internet platforms, the publication of content, goods, and services is increasingly characterized by cross-platform and multi-channel nature. Existing cross-platform publishing technologies typically rely on manual experience or rule-based configuration methods to adapt and validate the column structures, field specifications, and publishing rules of different platforms. In practice, significant differences exist between platforms in terms of column systems, field requirements, rule constraints, and compliance standards, leading to the need for multiple rounds of manual verification and adjustments before publication. This results in low publishing efficiency and is prone to rule omissions or adaptation errors.

[0003] In recent years, some technical solutions have introduced knowledge graphs or intelligent retrieval technologies to manage publication rules in a structured manner, and combined rule verification or intelligent recommendation methods to assist in publication decisions. These technologies have improved the systematic nature of rule management to some extent, but most solutions still rely on static rule matching or simple reasoning, making it difficult to accurately identify the key constraints that lead to non-publication in situations with complex and intertwined rules. When publication results do not meet platform requirements, existing technologies can usually only provide violation warnings or list the triggering rule information, failing to further pinpoint the set of rules with the least impact, and also struggling to generate executable content adjustment solutions.

[0004] Furthermore, existing cross-platform publishing technologies often lack a unified decision-making framework when facing scenarios involving simultaneous decision-making across multiple platforms, making it difficult to ensure rule consistency while also taking into account platform differences. Particularly when correcting unpublishable results, existing technologies generally lack a reverse reasoning mechanism based on rule-based causal relationships, failing to automatically generate publishing correction paths that minimize adjustment costs. This limits further improvements in the automation and intelligence of cross-platform publishing.

[0005] Therefore, how to provide a cross-platform intelligent publishing method based on knowledge graphs is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a cross-platform intelligent publishing method based on knowledge graphs. This invention constructs a unified publishing knowledge graph and generates platform-perspective knowledge subgraphs. Combined with a decision-oriented GraphRAG structure for publishing decisions, it intelligently infers the feasibility of content to be published across different publishing platforms. In cases where publishing is not possible, it generates counterfactual adjustment paths through reverse reasoning, thereby achieving automated cross-platform publishing decision-making and execution. This invention effectively solves the problems of large differences in rules across multiple platforms, high costs of manual adaptation, and difficulty in correcting unpublishable results. It possesses advantages such as strong rule adaptation capabilities, high degree of decision automation, low content adjustment costs, and good cross-platform consistency.

[0007] According to an embodiment of the present invention, a cross-platform intelligent publishing method based on knowledge graphs includes the following steps: Obtain publication rule information, column structure information, and field specification information from multiple publication platforms, parse them, and construct a unified publication knowledge graph; Based on a unified publication knowledge graph, corresponding platform-perspective knowledge subgraphs are generated for different publication platforms. Based on the knowledge subgraph from the platform perspective, a decision-oriented GraphRAG structure is constructed for publication decisions; The system acquires content to be published, performs semantic parsing, and, based on a decision-oriented GraphRAG structure, performs cross-platform publication feasibility reasoning in the platform's knowledge subgraph to generate publication feasibility results. When the feasibility results indicate that the content to be published is not publishable on at least one publishing platform, reverse reasoning is performed based on the platform-perspective knowledge subgraph and the decision-oriented GraphRAG structure to obtain the counterfactual adjustment result. Based on the feasibility results and counterfactual adjustment results, a cross-platform publication decision is generated, and the publication operation is executed according to the publication decision results to obtain the publication execution results.

[0008] Optionally, the step of obtaining, parsing, and constructing a unified publishing knowledge graph from multiple publishing platforms' publishing rule information, column structure information, and field specification information specifically includes: Obtain the set of publishing platforms participating in cross-platform publishing, wherein the set of publishing platforms includes multiple different publishing platforms; For each publishing platform in the set of publishing platforms, obtain the corresponding publishing rule information, column structure information and field specification information to form the platform's original information set. The platform's original information set is parsed and processed, and an entity set is constructed based on the publishing platform entity, column entity, field entity, and rule constraint entity. Based on the entity set, construct the rule constraint relationships between the entities in the entity set to form a relationship set; The entity sets and rule constraint relationship sets constructed by each publishing platform are uniformly merged to form a unified entity set and a unified rule constraint relationship set. A unified publication knowledge graph is constructed based on a unified set of entities and a unified set of rule constraint relationships.

[0009] Optionally, the step of generating corresponding platform-perspective knowledge subgraphs based on a unified publication knowledge graph for different publication platforms specifically includes: In the unified publication knowledge graph, retrieve the column entities, field entities, and rule constraint entities associated with the current publication platform to form a candidate entity set from the platform's perspective; Within a unified set of rule constraints, rule constraints corresponding to each entity in the candidate entity set from the platform's perspective are selected to form a set of relationships from the platform's perspective. Based on the candidate entity set and the relationship set from the platform perspective, a knowledge subgraph from the platform perspective is constructed. For the current publishing platform, establish a column mapping relationship between the unified column entity and the column structure information corresponding to the publishing platform in the platform's perspective knowledge subgraph, and write the column mapping relationship into the platform's perspective knowledge subgraph; For the current publishing platform, in the platform's perspective knowledge subgraph, establish a field mapping relationship between the unified field entity and the field specification information corresponding to the publishing platform, and write the field mapping relationship into the platform's perspective knowledge subgraph; For the current publishing platform, in the platform's perspective knowledge subgraph, the publishing rule information corresponding to the publishing platform is transformed into rule constraint relationships, and the rule constraint relationships are bound to the platform's perspective knowledge subgraph; After completing the platform perspective entity filtering, rule constraint relationship association, and column mapping relationship and field mapping relationship binding for each publishing platform in the publishing platform set, a platform perspective knowledge subgraph set is formed.

[0010] Optionally, the construction of a decision-oriented GraphRAG structure based on a platform-perspective knowledge subgraph for publication decisions specifically includes: For each platform-view knowledge subgraph in the platform-view knowledge subgraph set, based on the entity nodes and rule constraints contained in the platform-view knowledge subgraph, the platform-view knowledge subgraph is divided into multiple community subgraphs. For each community subgraph, the column entities, field entities, and rule constraint entities contained in the community subgraph are collected to generate corresponding structured community summary information, forming a community summary set; For the knowledge subgraph from the platform perspective, perform local subgraph retrieval within the knowledge subgraph from the platform perspective to select the local subgraphs that are relevant to the publication decision. The rule constraint relationships contained in the local subgraph are associated with the corresponding rule evidence units to form a rule evidence association set; A decision-oriented GraphRAG structure is constructed based on community summary sets, local subgraphs, and rule-based evidence association sets. For each platform-view knowledge subgraph in the platform-view knowledge subgraph set, a decision-oriented GraphRAG structure related to the corresponding publishing platform is formed, and the decision-oriented GraphRAG structures are aggregated to form a decision-oriented GraphRAG structure set.

[0011] Optionally, the step of obtaining the content to be published, performing semantic parsing, and based on a decision-oriented GraphRAG structure, performing cross-platform publication feasibility reasoning in the platform-view knowledge subgraph to generate publication feasibility results specifically includes: Obtain the content to be published, parse and process the content to be published, extract text information, attribute information and material information, and construct a representation of content elements; Based on the content element representation, the column entities and field entities that match the content element representation are retrieved in the unified publication knowledge graph to form a candidate publication element set. For each publishing platform, select a decision-oriented GraphRAG structure that corresponds to that platform; Based on the decision-oriented GraphRAG structure, the set of candidate publication elements is mapped to the entity nodes in the corresponding local subgraph of the decision-oriented GraphRAG structure, and the rule constraint relationship is identified in the local subgraph. Based on rule constraints, the content element representation is matched and judged with the column constraints, field constraints and rule triggering conditions of the current publishing platform to generate a publishing feasibility judgment result. Generate a corresponding publication feasibility assessment result for each publication platform in the collection of publication platforms.

[0012] Optionally, when the feasibility results indicate that the content to be published is not publishable on at least one publishing platform, the reverse reasoning based on the platform-perspective knowledge subgraph and the decision-oriented GraphRAG structure to obtain the counterfactual adjustment result specifically includes: When the feasibility results indicate that the content to be published is not publishable on at least one publishing platform, the corresponding unpublishable platforms are determined from the set of publishing platforms, forming a set of unpublishable platforms. For each publishing platform in the set of non-publishable platforms, select a platform-perspective knowledge subgraph and a decision-oriented GraphRAG structure corresponding to the publishing platform; Perform reverse reasoning on the rule constraints that are triggered in the current publishing platform to form a set of rule constraints that result in the inability to publish. Within the set of rule constraints, the minimum rule blocking subgraph is identified by progressively eliminating rule constraint relationships that do not affect the non-publication determination result. For the minimum rule blocking subgraph, based on the rule constraint relationships contained in the minimum rule blocking subgraph, content adjustment actions are generated and aggregated to form a set of counterfactual adjustment actions; Under the condition of minimizing constraints, a combination of adjustment actions is selected from the set of counterfactual adjustment actions to form a counterfactual adjustment path; The minimum rule blocking subgraph and the counterfactual adjustment path are used together as the counterfactual adjustment results of the corresponding publishing platform to form a set of counterfactual adjustment results.

[0013] Optionally, identifying the minimum rule-blocking subgraph by progressively eliminating rule constraints that do not affect the non-publication determination result from the rule constraint set specifically includes: For the set of rule constraints that cause publication to be unpublishable, an initial rule subgraph is constructed in the platform-perspective knowledge subgraph corresponding to the current publishing platform; Using the initial rule subgraph as the processing object, a verification rule subgraph is formed based on the rule constraint relationships that constitute the initial rule subgraph; For the verification rule subgraph, a publication feasibility determination process is adopted to perform a publication feasibility determination on the current publication platform and obtain the publication feasibility determination result; The publication feasibility assessment results are processed to obtain the minimum rule blocking subgraph.

[0014] Optionally, the step of selecting a combination of adjustment actions from the set of counterfactual adjustment actions to form a counterfactual adjustment path, while satisfying the minimization constraint, specifically includes: For the generated set of counterfactual adjustment actions, the set of counterfactual adjustment actions is determined; Based on the set of counterfactual adjustment actions, a set of candidate adjustment paths is constructed, where each candidate adjustment path consists of several content adjustment actions arranged in sequence. For each candidate adjustment path, a publication feasibility determination process is performed on the content to be published after applying the candidate adjustment path in the current publishing platform to obtain the corresponding publication feasibility determination result. In the candidate adjustment path set, the publication feasibility judgment results obtained for each candidate adjustment path are filtered to obtain the feasible adjustment path set. Under the condition of minimizing constraints, the candidate adjustment path with the fewest content adjustment actions is selected from the set of feasible adjustment paths as the counterfactual adjustment path.

[0015] Optionally, the step of generating a cross-platform publication decision result based on the publication feasibility result and counterfactual adjustment result, and executing the publication operation according to the publication decision result, to obtain the publication execution result specifically includes: Based on the cross-platform publication feasibility results and counterfactual adjustment results set, a comprehensive judgment is made on each publication platform in the publication platform set to determine the target publication platform set; Based on the column mapping relationship in the platform perspective knowledge subgraph corresponding to each target publishing platform, the target column is determined. For the target publishing platform, when the corresponding publishing feasibility assessment result is that it can be published, the content adjustment action set is determined to be an empty action set; When the corresponding publication feasibility assessment result is that it cannot be published and there is a counterfactual adjustment result corresponding to the target publication platform, the counterfactual adjustment path contained in the counterfactual adjustment result is determined as the set of content adjustment actions of the target publication platform. Based on the target publishing platform, target section, and set of content adjustment actions, generate corresponding publishing decision results; Based on the publication decision, the content to be published is adjusted accordingly according to the set of content adjustment actions, and the corresponding publication execution results are generated.

[0016] The beneficial effects of this invention are: This invention constructs a unified publication knowledge graph and generates platform-perspective knowledge subgraphs based on it. This unifies the modeling and semantic alignment of the column structures, field specifications, and rule constraints of different publication platforms, ensuring a consistent data foundation and rule expression for cross-platform publications. Based on this unified expression method, platform differences can be preserved and invoked in a structured manner, thereby reducing redundant configurations and information conflicts caused by scattered rules and inconsistent definitions during multi-platform adaptation, and improving the standardization and maintainability of cross-platform publication processing.

[0017] This invention constructs a decision-oriented GraphRAG structure based on a platform-perspective knowledge subgraph and generates structured community summary information using community partitioning. Combined with local subgraph retrieval and rule-based evidence unit association, it provides a structured evidence chain consistent with the target section, field requirements, and rule constraints for publication feasibility reasoning. Through the decision-oriented GraphRAG structure, the semantic parsing results of the content to be published can be associated and reasoned with platform rule constraints within the same graph structure, thereby improving the consistency of matching rule triggering conditions, field constraints, and section constraints during cross-platform publication feasibility determination.

[0018] When the feasibility results indicate that the content to be published is unpublishable on at least one publishing platform, this invention uses a platform-perspective knowledge subgraph and a decision-oriented GraphRAG structure to reverse-engineer the rule constraints that cause the unpublishability, identifies the minimum rule-blocking subgraph that makes the publishing conditions unsatisfactory, and generates a counterfactual adjustment path under the condition of satisfying the minimum constraint, thus obtaining a counterfactual adjustment result. Therefore, the system can provide an executable sequence of content adjustment actions based on the location of key constraints, and generate a cross-platform publishing decision based on the publication feasibility results and counterfactual adjustment results, completing the publishing operation. This forms a closed-loop processing flow from judgment and correction to execution, thereby improving the automation level of cross-platform publishing and reducing the frequency of manual correction. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a cross-platform intelligent publishing method based on knowledge graphs proposed in this invention; Figure 2 This is a schematic diagram of the decision-oriented GraphRAG structure in a cross-platform intelligent publishing method based on knowledge graphs proposed in this invention; Figure 3 This is a schematic diagram of the minimum rule blocking subgraph recognition process in a cross-platform intelligent publishing method based on knowledge graphs proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figures 1-3 A cross-platform intelligent publishing method based on knowledge graphs includes the following steps: Obtain publication rule information, column structure information, and field specification information from multiple publication platforms, parse them, and construct a unified publication knowledge graph; Based on a unified publication knowledge graph, corresponding platform-perspective knowledge subgraphs are generated for different publication platforms. Based on the knowledge subgraph from the platform perspective, a decision-oriented GraphRAG structure is constructed for publication decisions; The system acquires content to be published, performs semantic parsing, and, based on a decision-oriented GraphRAG structure, performs cross-platform publication feasibility reasoning in the platform's knowledge subgraph to generate publication feasibility results. When the feasibility results indicate that the content to be published is not publishable on at least one publishing platform, reverse reasoning is performed based on the platform-perspective knowledge subgraph and the decision-oriented GraphRAG structure to obtain the counterfactual adjustment result. Based on the feasibility results and counterfactual adjustment results, a cross-platform publication decision is generated, and the publication operation is executed according to the publication decision results to obtain the publication execution results.

[0022] In this embodiment, the step of obtaining, parsing, and constructing a unified publishing knowledge graph from multiple publishing platforms, including publishing rule information, column structure information, and field specification information, specifically includes: Obtain a set of publishing platforms participating in cross-platform publishing. The set of publishing platforms includes multiple different publishing platforms, each of which is used to host the publication, display, or distribution of the content to be published. The number of publishing platforms is determined based on the actual number of platforms participating in the publishing. For each publishing platform in the set of publishing platforms, obtain the corresponding publishing rule information, column structure information, and field specification information, and then combine the publishing rule information, column structure information, and field specification information to form the platform's original information set; The platform's original information set is parsed and processed. The platform entity, column entity, field entity, and rule constraint entity are identified and extracted from the platform's original information set. The corresponding entity set is constructed based on the platform entity, column entity, field entity, and rule constraint entity. Based on the entity set, rule constraint relationships between the entities in the entity set are constructed. Different entities representing the publishing platform, column, field and rule constraints are associated according to their mutual constraint relationships to form a set of relationships used to represent the rule constraint relationships between entities. The rule constraint relationships are used to characterize the dependency relationship, applicability relationship or restriction relationship between different entities. The entity sets and rule constraint relationship sets constructed for each publishing platform are uniformly integrated. The entities, column entities, field entities and rule constraint entities with the same or corresponding semantics in different publishing platforms are integrated, and the rule constraint relationships constructed in each publishing platform are merged to form a unified entity set and a unified rule constraint relationship set. Based on a unified set of entities and a unified set of rule constraints, a unified publishing knowledge graph is constructed to represent the publishing platform, columns, fields, and rule constraints, enabling the unified publishing knowledge graph to describe the relationships between publishing platforms, columns, fields, and rule constraints in a structured manner.

[0023] In this embodiment, the step of generating corresponding platform-perspective knowledge subgraphs based on a unified publication knowledge graph for different publication platforms specifically includes: Based on the collection of publishing platforms and the unified publishing knowledge graph, for each publishing platform in the collection of publishing platforms, a platform-view knowledge subgraph corresponding to the publishing platform is constructed, so that each platform-view knowledge subgraph can be used to carry the relevant entity and rule constraint information of the publishing platform in the unified publishing knowledge graph. In the unified publication knowledge graph, the column entities, field entities, and rule constraint entities associated with the current publication platform are retrieved, and the retrieved column entities, field entities, and rule constraint entities are aggregated to form a set of candidate entities from the platform perspective corresponding to the publication platform. In a unified set of rule constraint relationships, rule constraint relationships corresponding to each entity in the candidate entity set from the platform's perspective are selected, and the selected rule constraint relationships are aggregated to form a set of platform perspective relationships corresponding to the current publishing platform. Based on the candidate entity set and the relationship set from the platform perspective, a platform perspective knowledge subgraph corresponding to the current publishing platform is constructed, so that the platform perspective knowledge subgraph can be used to represent the structured association of each entity and its rule constraint relationship under the publishing platform. For the current publishing platform, in the platform's perspective knowledge subgraph, establish a column mapping relationship between the unified column entity and the column structure information corresponding to the publishing platform, and write the column mapping relationship into the platform's perspective knowledge subgraph so that the platform's perspective knowledge subgraph can reflect the correspondence between the unified column structure and the column structure of the publishing platform. For the current publishing platform, in the platform's perspective knowledge subgraph, establish a field mapping relationship between the unified field entity and the corresponding field specification information of the publishing platform, and write the field mapping relationship into the platform's perspective knowledge subgraph so that the platform's perspective knowledge subgraph can reflect the correspondence between the unified field specification and the field specification of the publishing platform; For the current publishing platform, in the platform's perspective knowledge subgraph, the publishing rule information corresponding to the publishing platform is transformed into rule constraint relationships, and the rule constraint relationships are bound to the platform's perspective knowledge subgraph, so that the platform's perspective knowledge subgraph can reflect the publishing rule constraint requirements of the publishing platform. After completing the platform perspective entity filtering, rule constraint relationship association, and column mapping relationship and field mapping relationship binding for each publishing platform in the publishing platform set, a platform perspective knowledge subgraph set corresponding one-to-one with each publishing platform in the publishing platform set is formed. The platform perspective knowledge subgraph set is used to represent the perspective-based knowledge structure of different publishing platforms under the unified publishing knowledge graph.

[0024] In this embodiment, the construction of a decision-oriented GraphRAG structure based on a platform-perspective knowledge subgraph for publication decisions specifically includes: For each platform-view knowledge subgraph in the platform-view knowledge subgraph set, based on the entity nodes and rule constraint relationships contained in the platform-view knowledge subgraph, and according to the connection relationship between column entities, field entities and rule constraint entities, the platform-view knowledge subgraph is divided into multiple distinct community subgraphs. For each community subgraph obtained from the knowledge subgraph from the platform perspective, the column entities, field entities, and rule constraint entities contained in the community subgraph are collected, and corresponding structured community summary information is generated based on the column entities, field entities, and rule constraint entities. The structured community summary information includes at least column constraint information, field constraint information, and rule triggering condition information related to the publication decision, thereby forming a community summary set of the knowledge subgraph from the platform perspective. For the platform-perspective knowledge subgraph, the column entities or field entities involved in the structured community summary information are used as the retrieval starting point. Local subgraph retrieval is performed in the platform-perspective knowledge subgraph. The scope of the local subgraph retrieval is limited to entity nodes that are connected to the retrieval starting point within a preset relationship depth and through rule-constrained relationships. Local subgraphs related to the publication decision are selected from the platform-perspective knowledge subgraph. For a local subgraph, the rule constraint relationships contained in the local subgraph are associated with the corresponding rule evidence units. The rule evidence units are derived from the published rule information and correspond one-to-one with the rule constraint relationship, so that each rule constraint relationship is associated with at least one rule evidence unit and the rule evidence unit can be traced back to the corresponding published rule information source, thereby forming a rule evidence association set. Based on the community summary set, local subgraph, and rule evidence association set, the community summary set, local subgraph, and rule evidence association set are integrated to construct a decision-oriented GraphRAG structure for publication decisions. The decision-oriented GraphRAG structure is used to uniformly carry community summary information, rule constraint relationships, and their corresponding rule evidence units related to the current publication platform. For each platform-view knowledge subgraph in the platform-view knowledge subgraph set, a decision-oriented GraphRAG structure related to the corresponding publishing platform is formed. The decision-oriented GraphRAG structures are then aggregated to form a set of decision-oriented GraphRAG structures that correspond one-to-one with each publishing platform in the publishing platform set, serving as an improved GraphRAG structure for publishing decision-making.

[0025] In this embodiment, the step of obtaining the content to be published, performing semantic parsing, and based on a decision-oriented GraphRAG structure, performing cross-platform publication feasibility reasoning in the platform-perspective knowledge subgraph to generate publication feasibility results specifically includes: The process involves acquiring content to be published, parsing and processing the content, extracting text information, attribute information, and material information to describe the content, and constructing a content element representation based on the text information, attribute information, and material information to characterize the content to be published. Based on the content element representation, the column entities and field entities that match the content element representation are retrieved in the unified publication knowledge graph, and the retrieved column entities and field entities are aggregated to form a candidate publication element set, wherein the candidate publication element set is a subset of the entity set in the unified publication knowledge graph. For each publication platform in the publication platform set, a decision-oriented GraphRAG structure corresponding to the publication platform is selected, wherein the decision-oriented GraphRAG structure includes a community summary set, a local subgraph, and a rule evidence association set related to the publication platform; Based on the decision-oriented GraphRAG structure, the candidate publication element set is mapped to the entity nodes in the corresponding local subgraph of the decision-oriented GraphRAG structure, and the rule constraint relationship corresponding to the candidate publication element set and having an association relationship is identified in the local subgraph. Based on the identified rule constraint relationships, the content element representation is matched and judged with the column constraint conditions, field constraint conditions and rule trigger conditions corresponding to the current publishing platform, and the publishing feasibility judgment result corresponding to the publishing platform is generated according to the matching judgment result. For each publishing platform in the publishing platform set, a corresponding publishing feasibility determination result is generated. The publishing feasibility determination result is used to characterize the publishing feasibility status of the content to be published on the corresponding publishing platform. The publishing feasibility determination result includes at least a determination status of whether it can be published. When the determination status is that it cannot be published, it also includes at least one of the rule constraint information associated with the non-publishability determination, the column constraint condition or the field constraint condition that is not met.

[0026] In this embodiment, when the feasibility results indicate that the content to be published is not publishable on at least one publishing platform, the reverse reasoning based on the platform-perspective knowledge subgraph and the decision-oriented GraphRAG structure to obtain the counterfactual adjustment result specifically includes: When at least one of the publication feasibility results across platforms is deemed unpublishable, the corresponding unpublishable platform is determined from the set of publication platforms to form a set of unpublishable platforms. For each publishing platform in the set of non-publishable platforms, a platform-view knowledge subgraph and a decision-oriented GraphRAG structure corresponding to the publishing platform are selected. The decision-oriented GraphRAG structure includes a set of community summaries, a local subgraph, and a set of rule-based evidence associations related to the publishing platform. Based on the local subgraphs in the knowledge subgraph from the platform perspective and the rule evidence association set contained in the decision-oriented GraphRAG structure, reverse reasoning is performed on the rule constraint relationships triggered in the current publishing platform to identify the rule constraint relationships that are causally related to the non-publishing determination results, and the rule constraint relationships are aggregated to form a set of rule constraints that lead to non-publishing. The rule constraints include at least one of the following: column access constraints, field mandatory constraints, field format and value constraints, condition triggering constraints, mutual exclusion and dependency constraints, content compliance and posting ban constraints, material specification constraints, and platform interface and parameter constraints. Within the set of rule constraints, the minimum rule blocking subgraph is identified by progressively eliminating rule constraint relationships that do not affect the non-publication determination result. For the identified minimum rule blocking subgraph, based on the rule constraint relationship contained in the minimum rule blocking subgraph, a content adjustment action to eliminate the corresponding rule constraint is generated, and the generated content adjustment actions are collected to form a counterfactual adjustment action set, wherein the content adjustment action includes at least one of the following: column selection, field filling, field modification, material adjustment or rule triggering condition adjustment of the content to be published; Under the condition of minimizing constraints, a combination of adjustment actions is selected from the set of counterfactual adjustment actions to form a counterfactual adjustment path; The minimum rule blocking subgraph and the counterfactual adjustment path are used together as the counterfactual adjustment result for the corresponding publishing platform. For each publishing platform in the set of platforms that cannot be published, the corresponding counterfactual adjustment result is generated separately. The counterfactual adjustment results corresponding to each publishing platform are collected to form a counterfactual adjustment result set.

[0027] In this embodiment, the step of identifying the minimum rule-blocking subgraph by gradually eliminating rule constraint relationships that do not affect the non-publication determination result in the rule constraint set specifically includes: For the set of rule constraints that cause publication to be unpublishable, an initial rule subgraph is constructed in the platform perspective knowledge subgraph corresponding to the current publishing platform. This subgraph contains the rule constraint relationships in the set of rule constraints and their directly associated entity nodes. The directly associated entity nodes include column entities, field entities, or content element entities connected to the rule constraint relationships. Using the initial rule subgraph as the processing object, select individual rule constraints from the rule constraints that constitute the initial rule subgraph one by one in a preset order as the rule constraints to be verified. While keeping the other rule constraints and associated entity nodes in the initial rule subgraph unchanged, remove the rule constraints to be verified from the initial rule subgraph to form the corresponding verification rule subgraph. The preset order includes sorting according to the priority of the type of rule constraint relationship. The priority of the type of rule constraint relationship includes at least the following: content compliance and prohibition constraint relationship, column access constraint relationship, field mandatory constraint relationship, field format and value constraint relationship, condition trigger constraint relationship, mutual exclusion and dependency constraint relationship, material specification constraint relationship, and platform interface and parameter constraint relationship. When the type of rule constraint relationship is the same, it is sorted according to the severity level of the rule constraint relationship. When both the type and severity level of the rule constraint relationship are the same, it is sorted according to the distance of the rule constraint relationship from the column entity or field entity in the platform's knowledge subgraph, from closest to farthest. For the verification rule subgraph, a publication feasibility determination process is adopted to perform a publication feasibility determination on the current publication platform and obtain the publication feasibility determination result corresponding to the conditions of the verification rule subgraph. The publication feasibility determination result is used to characterize whether the current publication platform meets the publication conditions. The publication feasibility judgment result obtained from the verification rule subgraph is processed. When the publication feasibility judgment result is that it cannot be published, the corresponding rule constraint relationship to be verified is determined to be a non-critical rule constraint relationship, and the rule constraint relationship to be verified is removed from the initial rule subgraph. When the publication feasibility assessment result is that it can be published, the rule constraint relationship to be verified is identified as the key rule constraint relationship, and the rule constraint relationship to be verified is retained in the initial rule subgraph; After verifying the relationship between each rule constraint in the rule constraint set, a minimal rule blocking subgraph is obtained, consisting of the key rule constraint relationship and its associated entity nodes, which makes the publication conditions unsatisfactory.

[0028] In this embodiment, the step of selecting a combination of adjustment actions from the set of counterfactual adjustment actions to form a counterfactual adjustment path while satisfying the minimization constraint specifically includes: For the generated set of counterfactual adjustment actions, the set of counterfactual adjustment actions is determined. Each counterfactual adjustment action in the set of counterfactual adjustment actions is used to adjust the rule constraint relationship contained in the minimum rule blocking subgraph for the content to be published. The counterfactual adjustment action includes at least one of the following: column selection adjustment, field filling adjustment, field content modification, material content adjustment, or rule triggering condition adjustment for the content to be published. Based on the set of counterfactual adjustment actions, a set of candidate adjustment paths is constructed, which consists of multiple content adjustment actions combined in a predetermined order. Each candidate adjustment path is composed of several content adjustment actions arranged in sequence, which is used to represent a complete adjustment plan for the content to be published. For each candidate adjustment path in the candidate adjustment path set, a publication feasibility determination process is performed on the content to be published after applying the candidate adjustment path in the current publication platform to obtain the corresponding publication feasibility determination result, which is used to characterize whether the content to be published meets the publication conditions after applying the candidate adjustment path. In the candidate adjustment path set, the publication feasibility judgment results obtained for each candidate adjustment path are screened, and the candidate adjustment path that can change the publication feasibility judgment result from unpublishable to publishable is selected. The screened candidate adjustment paths are then gathered to form a feasible adjustment path set. Under the condition of minimizing constraints, the candidate adjustment path with the fewest content adjustment actions is selected from the set of feasible adjustment paths as the counterfactual adjustment path. The counterfactual adjustment path is the sequence of adjustment actions with the fewest required content adjustment actions under the premise of meeting the publication conditions.

[0029] In this embodiment, the step of generating a cross-platform publication decision result based on the publication feasibility result and the counterfactual adjustment result, and executing the publication operation according to the publication decision result to obtain the publication execution result specifically includes: Based on the cross-platform publication feasibility results and counterfactual adjustment results set, a comprehensive judgment is made on each publication platform in the publication platform set to determine the target publication platform set. The target publication platform set includes publication platforms whose publication feasibility judgment result is that they can be published, and publication platforms that can meet the publication conditions after applying the corresponding counterfactual adjustment results. For each target publishing platform in the target publishing platform set, based on the column mapping relationship in the platform perspective knowledge subgraph corresponding to the target publishing platform, the target column corresponding to the target publishing platform is determined, and the target column is the column entity in the platform perspective knowledge subgraph. For the target publishing platform, when the corresponding publishing feasibility assessment result is that it can be published, the set of content adjustment actions for the target publishing platform is determined to be an empty set of actions. When the corresponding publication feasibility assessment result is that it cannot be published and there is a counterfactual adjustment result corresponding to the target publication platform, the counterfactual adjustment path contained in the counterfactual adjustment result is determined as the set of content adjustment actions of the target publication platform. Based on the target publishing platform, target column, and set of content adjustment actions, a corresponding publishing decision result is generated, wherein the publishing decision result is used to characterize the specific publishing plan to be implemented for the content to be published under the target publishing platform and target column. The generation of publication decision results refers to combining the determined target publication platform, the target column corresponding to the target publication platform, and the set of content adjustment actions determined for the target publication platform to form a publication decision result that includes platform identification information, column identification information, and content adjustment instruction information. Based on the publication decision, the content to be published is adjusted accordingly according to the set of content adjustment actions, and the adjusted content to be published is submitted to the corresponding section of the target publication platform to complete the publication operation on the target publication platform and generate the corresponding publication execution result.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to a business scenario of centralized content publishing across multiple platforms. In this scenario, the content to be published needs to be simultaneously published to multiple different types of publishing platforms, which have significant differences in their column systems, field specifications, content compliance requirements, and rule constraints. Traditionally, content publishers need to perform rule comparisons, field entry, and manual verification for each platform separately. When a publishing restriction occurs, they can only modify each item based on the error messages returned by the platform, resulting in a long overall publishing cycle and a lot of repetitive work.

[0031] In this application scenario, the system first acquires publishing rule information, column structure information, and field specification information for multiple publishing platforms, and then constructs a unified publishing knowledge graph based on this information. This unified publishing knowledge graph uses publishing platforms, columns, fields, and rule constraints as core entities, describing the commonalities and differences between the rules of different platforms through structured relationships. On this basis, a corresponding platform-perspective knowledge subgraph is generated for each publishing platform, enabling the rule constraints of each platform to be individually invoked and reasoned within a unified semantic framework.

[0032] Once content to be published enters the system, it undergoes semantic parsing to generate content element representations. Based on a platform-perspective knowledge subgraph, a decision-oriented GraphRAG structure is constructed. This structure generates structured community summary information through community partitioning and, combined with local subgraph retrieval and rule-based evidence unit association, infers and judges the feasibility of publishing the content across various platforms. In actual testing, some platforms determined the publication feasibility to be published, while others determined it to be unpublishable due to missing fields, mismatched categories, or unmet rule triggering conditions.

[0033] For platforms where content cannot be published, the system further performs reverse reasoning based on the platform's knowledge subgraph and decision-oriented GraphRAG structure to identify the rule constraints that cause the publication restriction and construct a minimal rule blocking subgraph. Based on this, the system generates a set of counterfactual adjustment actions and, under the premise of satisfying the minimum constraint conditions, generates a counterfactual adjustment path, ensuring that the content to be published meets the platform's rule requirements with as few content adjustments as possible. After adjustment, the system performs a publication feasibility determination again to confirm that the adjusted content meets the publication conditions and generates the corresponding publication decision result, completing the cross-platform automatic publication operation.

[0034] In practical applications, the method of this invention was compared with traditional manual rule verification methods. Test results show that, when processing the same amount of content to be published, this invention demonstrates significant differences in average preparation time, number of manual interventions, and efficiency in fixing unpublishable issues. Especially in cases of complex rules and significant platform differences, this invention can significantly reduce unnecessary content modifications through minimum rule blocking subgraphs and counterfactual adjustment paths, making the overall publishing process more stable and controllable.

[0035] To verify the feasibility of this invention in practice, it was applied to a multi-platform content centralized publishing system. Multiple batches of content to be published were processed, and the relevant data were statistically analyzed, yielding the following test results.

[0036] Table 1 Comparison of Implementation Effects of Cross-Platform Intelligent Publication Methods

[0037] As can be seen from the table above, this invention demonstrates significant advantages over traditional manual publishing methods in several key implementation metrics for cross-platform content publishing. Regarding publishing preparation efficiency, the average preparation time for a single piece of content using the traditional manual method is 38.6 minutes, while this time is reduced to 11.4 minutes using the method of this invention, a reduction of more than two-thirds. This result indicates that by using a unified publishing knowledge graph and a decision-oriented GraphRAG structure to centrally model and reason about platform rules, the time spent on repeatedly comparing rules and manually adjusting content can be significantly reduced, making the publishing process more efficient.

[0038] In terms of metrics related to manual intervention and problem remediation, this invention also demonstrates significant improvements. Traditionally, a single piece of content requires an average of 4.2 manual adjustments, while in this invention's method, this number is reduced to 1.1, indicating that most unpublishable issues can be automatically located and remediated during the system reasoning and counterfactual adjustment path generation stages. Regarding the efficiency of handling unpublishable issues, the traditional method has an average problem location time of 15.8 minutes and a remediation time of 22.4 minutes, while the method of this invention reduces the location and remediation times to 3.6 minutes and 6.9 minutes, respectively, demonstrating that the reverse reasoning mechanism based on the minimum rule blocking subgraph can more quickly identify key rule constraints and provide targeted adjustment solutions.

[0039] From the perspective of content adjustment quality and consistency, the average number of repeated field modifications under the traditional manual method is 6.7, while this number drops to 1.9 in the method of this invention. This indicates that by adjusting the path using counterfactual methods while meeting the minimum constraint conditions, large-scale, multi-round content modifications can be effectively avoided. Regarding the number of triggered rule constraints, the two methods are similar, at 8.3 and 8.1 respectively, indicating that this invention does not improve the pass rate by reducing rule validation, but rather improves processing efficiency and adjustment accuracy while ensuring complete rule validation.

[0040] In terms of final publication results, this invention demonstrates significant improvements in key performance indicators such as successful publication rate, single-batch cross-platform publication completion rate, and cross-platform consistency verification pass rate. The successful publication rate increased from 86.5% using the traditional method to 98.2%, the single-batch cross-platform publication completion rate increased from 79.4% to 96.8%, and the cross-platform consistency verification pass rate increased from 82.1% to 97.5%. These data indicate that this invention can more stably complete publication tasks in complex multi-platform environments, improving overall publication quality while ensuring rule consistency, thus validating the practical application value of this method in cross-platform intelligent publication scenarios.

[0041] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A cross-platform intelligent publishing method based on knowledge graphs, characterized in that, Includes the following steps: Obtain publication rule information, column structure information, and field specification information from multiple publication platforms, parse them, and construct a unified publication knowledge graph; Based on a unified publication knowledge graph, corresponding platform-perspective knowledge subgraphs are generated for different publication platforms. Based on the knowledge subgraph from the platform perspective, a decision-oriented GraphRAG structure is constructed for publication decisions; The system acquires content to be published, performs semantic parsing, and, based on a decision-oriented GraphRAG structure, performs cross-platform publication feasibility reasoning in the platform's knowledge subgraph to generate publication feasibility results. When the feasibility results indicate that the content to be published is not publishable on at least one publishing platform, reverse reasoning is performed based on the platform-perspective knowledge subgraph and the decision-oriented GraphRAG structure to obtain the counterfactual adjustment result. Based on the feasibility results and counterfactual adjustment results, a cross-platform publication decision is generated, and the publication operation is executed according to the publication decision results to obtain the publication execution results.

2. The cross-platform intelligent publishing method based on knowledge graphs according to claim 1, characterized in that, The process of obtaining, parsing, and constructing a unified publication knowledge graph from multiple publishing platforms, including acquiring and analyzing publication rule information, column structure information, and field specification information from multiple platforms, specifically includes: Obtain the set of publishing platforms participating in cross-platform publishing, wherein the set of publishing platforms includes multiple different publishing platforms; For each publishing platform in the set of publishing platforms, obtain the corresponding publishing rule information, column structure information and field specification information to form the platform's original information set. The platform's original information set is parsed and processed, and an entity set is constructed based on the publishing platform entity, column entity, field entity, and rule constraint entity. Based on the entity set, construct the rule constraint relationships between the entities in the entity set to form a relationship set; The entity sets and rule constraint relationship sets constructed by each publishing platform are uniformly merged to form a unified entity set and a unified rule constraint relationship set; A unified publication knowledge graph is constructed based on a unified set of entities and a unified set of rule constraint relationships.

3. The cross-platform intelligent publishing method based on knowledge graphs according to claim 1, characterized in that, The specific steps of generating corresponding platform-perspective knowledge subgraphs based on a unified publication knowledge graph for different publication platforms include: In the unified publication knowledge graph, retrieve the column entities, field entities, and rule constraint entities associated with the current publication platform to form a candidate entity set from the platform's perspective; Within a unified set of rule constraints, rule constraints corresponding to each entity in the candidate entity set from the platform's perspective are selected to form a set of relationships from the platform's perspective. Based on the candidate entity set and the relationship set from the platform perspective, a knowledge subgraph from the platform perspective is constructed. For the current publishing platform, establish a column mapping relationship between the unified column entity and the column structure information corresponding to the publishing platform in the platform's perspective knowledge subgraph, and write the column mapping relationship into the platform's perspective knowledge subgraph; For the current publishing platform, in the platform's perspective knowledge subgraph, establish a field mapping relationship between the unified field entity and the field specification information corresponding to the publishing platform, and write the field mapping relationship into the platform's perspective knowledge subgraph; For the current publishing platform, in the platform's perspective knowledge subgraph, the publishing rule information corresponding to the publishing platform is transformed into rule constraint relationships, and the rule constraint relationships are bound to the platform's perspective knowledge subgraph; After completing the platform perspective entity filtering, rule constraint relationship association, and column mapping relationship and field mapping relationship binding for each publishing platform in the publishing platform set, a platform perspective knowledge subgraph set is formed.

4. The cross-platform intelligent publishing method based on knowledge graphs according to claim 1, characterized in that, The construction of a decision-oriented GraphRAG structure based on platform-perspective knowledge subgraphs for publication decisions specifically includes: For each platform-view knowledge subgraph in the platform-view knowledge subgraph set, based on the entity nodes and rule constraints contained in the platform-view knowledge subgraph, the platform-view knowledge subgraph is divided into multiple community subgraphs. For each community subgraph, the column entities, field entities, and rule constraint entities contained in the community subgraph are collected to generate corresponding structured community summary information, forming a community summary set; For the knowledge subgraph from the platform perspective, perform local subgraph retrieval within the knowledge subgraph from the platform perspective to select the local subgraphs that are relevant to the publication decision. The rule constraint relationships contained in the local subgraph are associated with the corresponding rule evidence units to form a rule evidence association set; A decision-oriented GraphRAG structure is constructed based on community summary sets, local subgraphs, and rule-based evidence association sets. For each platform-view knowledge subgraph in the platform-view knowledge subgraph set, a decision-oriented GraphRAG structure related to the corresponding publishing platform is formed, and the decision-oriented GraphRAG structures are aggregated to form a decision-oriented GraphRAG structure set.

5. The cross-platform intelligent publishing method based on knowledge graphs according to claim 1, characterized in that, The process of obtaining the content to be published, performing semantic parsing, and, based on a decision-oriented GraphRAG structure, performing cross-platform publication feasibility reasoning within the platform's knowledge subgraph to generate publication feasibility results specifically includes: Obtain the content to be published, parse and process the content to be published, extract text information, attribute information and material information, and construct a representation of content elements; Based on the content element representation, the column entities and field entities that match the content element representation are retrieved in the unified publication knowledge graph to form a candidate publication element set. For each publishing platform, select a decision-oriented GraphRAG structure that corresponds to that platform; Based on the decision-oriented GraphRAG structure, the set of candidate publication elements is mapped to the entity nodes in the corresponding local subgraph of the decision-oriented GraphRAG structure, and the rule constraint relationship is identified in the local subgraph. Based on rule constraints, the content element representation is matched and judged with the column constraints, field constraints and rule triggering conditions corresponding to the current publishing platform to generate a publishing feasibility judgment result. Generate a corresponding publication feasibility assessment result for each publication platform in the collection of publication platforms.

6. The cross-platform intelligent publishing method based on knowledge graphs according to claim 1, characterized in that, When the feasibility results indicate that the content to be published is not publishable on at least one publishing platform, reverse reasoning is performed based on the platform-perspective knowledge subgraph and the decision-oriented GraphRAG structure to obtain the counterfactual adjustment result, specifically including: When the feasibility results indicate that the content to be published is not publishable on at least one publishing platform, the corresponding unpublishable platforms are determined from the set of publishing platforms, forming a set of unpublishable platforms. For each publishing platform in the set of non-publishable platforms, select a platform-perspective knowledge subgraph and a decision-oriented GraphRAG structure corresponding to the publishing platform; Perform reverse reasoning on the rule constraints that are triggered in the current publishing platform to form a set of rule constraints that result in the inability to publish. Within the set of rule constraints, the minimum rule blocking subgraph is identified by progressively eliminating rule constraint relationships that do not affect the non-publication determination result. For the minimum rule blocking subgraph, based on the rule constraint relationships contained in the minimum rule blocking subgraph, content adjustment actions are generated and aggregated to form a set of counterfactual adjustment actions; Under the condition of minimizing constraints, a combination of adjustment actions is selected from the set of counterfactual adjustment actions to form a counterfactual adjustment path; The minimum rule blocking subgraph and the counterfactual adjustment path are used together as the counterfactual adjustment results of the corresponding publishing platform to form a set of counterfactual adjustment results.

7. The cross-platform intelligent publishing method based on knowledge graphs according to claim 6, characterized in that, The process of identifying the minimum rule-blocking subgraph by progressively eliminating rule constraints that do not affect the non-publication determination result from the rule constraint set specifically includes: For the set of rule constraints that cause publication to be unpublishable, an initial rule subgraph is constructed in the platform-perspective knowledge subgraph corresponding to the current publishing platform; Using the initial rule subgraph as the processing object, a verification rule subgraph is formed based on the rule constraint relationships that constitute the initial rule subgraph; For the verification rule subgraph, a publication feasibility determination process is adopted to perform a publication feasibility determination on the current publication platform and obtain the publication feasibility determination result; The publication feasibility assessment results are processed to obtain the minimum rule blocking subgraph.

8. The cross-platform intelligent publishing method based on knowledge graphs according to claim 6, characterized in that, The step of selecting a combination of adjustment actions from the set of counterfactual adjustment actions to form a counterfactual adjustment path, under the condition of satisfying the minimization constraint, specifically includes: For the generated set of counterfactual adjustment actions, the set of counterfactual adjustment actions is determined; Based on the set of counterfactual adjustment actions, a set of candidate adjustment paths is constructed, where each candidate adjustment path consists of several content adjustment actions arranged in sequence. For each candidate adjustment path, a publication feasibility determination process is performed on the content to be published after applying the candidate adjustment path in the current publishing platform to obtain the corresponding publication feasibility determination result. In the candidate adjustment path set, the publication feasibility judgment results obtained for each candidate adjustment path are filtered to obtain the feasible adjustment path set. Under the condition of minimizing constraints, the candidate adjustment path with the fewest content adjustment actions is selected from the set of feasible adjustment paths as the counterfactual adjustment path.

9. The cross-platform intelligent publishing method based on knowledge graphs according to claim 1, characterized in that, The process of generating a cross-platform publication decision based on the publication feasibility results and counterfactual adjustment results, and then executing the publication operation according to the publication decision results, to obtain the publication execution results specifically includes: Based on the cross-platform publication feasibility results and counterfactual adjustment results set, a comprehensive judgment is made on each publication platform in the publication platform set to determine the target publication platform set; Based on the column mapping relationship in the platform perspective knowledge subgraph corresponding to each target publishing platform, the target column is determined. For the target publishing platform, when the corresponding publishing feasibility assessment result is that it can be published, the content adjustment action set is determined to be an empty action set; When the corresponding publication feasibility assessment result is that it cannot be published and there is a counterfactual adjustment result corresponding to the target publication platform, the counterfactual adjustment path contained in the counterfactual adjustment result is determined as the set of content adjustment actions of the target publication platform. Based on the target publishing platform, target section, and set of content adjustment actions, generate corresponding publishing decision results; Based on the publication decision, the content to be published is adjusted accordingly according to the set of content adjustment actions, and the corresponding publication execution results are generated.

Images