A method for adaptive arrangement of propaganda content for a propagation target

Abstract

The present application relates to the field of artificial intelligence, and particularly relates to a method for adaptive arrangement of propaganda content for a propagation target, which comprises the following steps: propagation data fusion construction, propagation relationship semantic modeling, propagation target driven semantic extraction, propagation state dynamic prediction, multi-target propaganda content arrangement optimization and propagation feedback adaptive optimization. The present application unifies the complex propagation relationship among the propaganda content, the propagation target, the propagation channel and the user through the propagation heterogeneous graph, the adaptive selection of the meta path and the collaborative modeling of the double-domain hypergraph, improves the matching accuracy between the propaganda content and the propagation target; according to the future propagation state, the present application dynamically generates the propaganda content arrangement strategy through the propagation state prediction, the multi-target content arrangement optimization and the propagation feedback adaptive updating mechanism, and improves the propaganda content propagation effect, the propagation resource utilization efficiency and the propagation target adaptability.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, specifically to an adaptive arrangement method for propaganda content oriented towards communication objectives. Background Technology

[0002] With the rapid development of internet media, mobile social platforms, and intelligent recommendation technologies, promotional content has been widely applied in various fields such as government publicity, brand promotion, product marketing, public services, and media communication. Existing methods for arranging promotional content focus on a single object within the content, user, or communication channel, making it difficult to uniformly model the complex and multi-type relationships between content, communication goals, communication channels, and users. They also underutilize high-order semantic relationships that exist during the communication process, resulting in limited ability to represent communication relationships and low matching accuracy between promotional content and communication goals. Consequently, they struggle to achieve good communication results in complex communication scenarios. Furthermore, existing methods typically formulate communication strategies based on the current communication state, lacking the ability to predict future trends and dynamic optimization mechanisms driven by communication feedback. This makes it difficult to adjust promotional content arrangement strategies in a timely manner, leading to decreased communication effectiveness, low efficiency in utilizing communication resources, and insufficient adaptability of promotional content to different communication goals. Consequently, they fail to meet the application needs for intelligent and adaptive arrangement of promotional content in complex communication scenarios. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an adaptive content arrangement method for dissemination objectives. This invention utilizes heterogeneous dissemination graphs, adaptive meta-path selection, and dual-domain hypergraph collaborative modeling to uniformly represent the complex dissemination relationships among dissemination content, objectives, channels, and users. It fully mines high-order semantic associations between dissemination entities, improving the accuracy of dissemination relationship representation and the semantic modeling capability of dissemination objectives. This enhances the matching accuracy between dissemination content and objectives, solving the problems of insufficient dissemination relationship modeling capability and poor matching effect between dissemination content and objectives in existing technologies. Furthermore, through dissemination state prediction, multi-objective content arrangement optimization, and a dissemination feedback adaptive update mechanism, it dynamically generates dissemination content arrangement strategies based on future dissemination states and continuously optimizes dissemination relationship characteristics and evaluation model parameters based on dissemination feedback. This enables dissemination content to adjust its arrangement strategy in real time according to changes in the dissemination process, improving the dissemination effect, dissemination resource utilization efficiency, and dissemination objective adaptability. This solves the problems of insufficient dissemination state prediction and dynamic optimization capabilities, and inadequate adaptive dissemination content arrangement strategies in existing technologies.

[0004] The technical solution adopted by this invention is as follows: This invention provides an adaptive arrangement method for promotional content oriented towards communication goals, the method comprising the following steps: Step S1: Collect publicity data, including historical publicity content data, user interaction behavior data, dissemination channel data and dissemination target data. Extract content theme features, semantic features and dissemination attribute features from publicity content data, extract behavioral preference features from user interaction behavior data, extract channel attribute features from dissemination channel data, and extract dissemination target features from dissemination target data. Then, perform correlation and fusion to construct a multi-source dissemination dataset. Step S2: Construct a propagation heterogeneity graph based on the multi-source propagation dataset, filter effective propagation semantic paths based on the meta-path selection mechanism of Monte Carlo tree search, and generate propagation relationship feature representations using the propagation heterogeneity graph; Step S3: Calculate the association relationship between propagation entities based on the propagation relationship feature representation, construct the content relationship hypergraph and the target association hypergraph, and extract the high-order semantic features driven by the propagation target through hypergraph convolution and attention aggregation mechanism; Step S4: Based on the high-order semantic features driven by the propagation target, construct a dynamic propagation relationship graph, input the dynamic propagation relationship graph into the time series prediction model, predict the propagation state within the future time window, and obtain the prediction result of the future propagation state. Step S5: Based on the prediction results of future propagation status, construct a multi-objective evaluation model and calculate the optimal orchestration strategy; Step S6: Generate a target-matched promotional content plan based on the optimal arrangement strategy, update the dissemination relationship characteristics based on actual dissemination feedback data, and optimize the parameters of the multi-objective evaluation model.

[0005] Furthermore, step S2 specifically includes the following steps: Step S21: Perform entity parsing and relation parsing on the multi-source propagation dataset to identify entities such as propaganda content, users, propagation channels, and propagation targets, construct an entity set, and construct a propagation relation set based on the relationships between the entities; Step S22: Construct a heterogeneous propagation graph based on the entity set and the propagation relation set, map different types of entities to different types of nodes, map propagation relations to heterogeneous edges of the corresponding type, and generate initial attribute vectors for each node and edge; Step S23: Construct a candidate meta-path set based on the propagation heterogeneous graph, encode the candidate meta-paths in the candidate meta-path set, and use a meta-path selection mechanism based on Monte Carlo tree search to iteratively search the candidate meta-paths to form an effective propagation semantic path set; Step S24: Based on the set of effective propagation semantic paths, perform path constraint message propagation on the heterogeneous graph, aggregate node information along the effective propagation semantic paths, use a heterogeneous graph message passing mechanism based on meta-path constraints to perform node information propagation on the heterogeneous graph, perform cross-type feature aggregation based on the propagation relationship between different types of nodes, and perform fusion encoding based on the association relationship between different node representation vectors to generate propagation relationship feature representation. The fusion encoding process adopts the following method: ; in, This represents the characteristics of the propagation relationship. Indicates the first Node-like propagation representation, Indicates the fusion weight of node types. Indicates the effective propagation semantic path for the first Node-like propagation contribution coefficient Indicates the number of node types.

[0006] Furthermore, step S23 specifically includes the following steps: Step S231: Based on the node type, heterogeneous edge type, and propagation target characteristics in the heterogeneous propagation graph, traverse the node connection relationships in the heterogeneous propagation graph to obtain the node transfer relationships between different types of nodes. Generate a node type sequence that satisfies the node type transfer constraints based on the node transfer relationships. Combine the association relationship between the propagation target entity and the publicity content entity to perform target association filtering on the node type sequence. Combine the heterogeneous relationships corresponding to the node type sequences that satisfy the propagation target characteristics to form a candidate metapath set. Step S232: Perform path state encoding on each candidate path in the candidate path set. Based on the node type sequence, edge type sequence, path length information, and propagation target features contained in the candidate path, generate the corresponding path state vector, represented as: ; in, This represents the path state vector corresponding to the p-th candidate meta-path. This indicates the node type characteristics included in the candidate metapath. This indicates the edge type feature corresponding to the candidate metapath. This represents the path length feature of candidate meta-elements. This represents the association features between candidate metapaths and propagation targets; A meta-path search tree is constructed based on the path state vector. The candidate meta-path expansion process is mapped to search actions, and the selection of the next node type and the next propagation relationship are taken as path expansion actions. Step S233: Using candidate meta-paths as the search object, the candidate meta-paths are iteratively expanded, simulated, and evaluated for benefits through the Monte Carlo tree search mechanism; During path simulation, based on the information propagation process of the candidate path, the degree of node representation enhancement, the degree of propagation entity association enhancement, and the degree of propagation target matching are calculated, and a path benefit evaluation function is constructed: ; in, This represents the path revenue evaluation value corresponding to the p-th candidate meta-path. Indicates the use of the first The degree of enhancement of node representation after information propagation of candidate meta-paths Indicates the use of the first The degree of enhancement of the association features between entities after the propagation of candidate meta-paths. This indicates the degree of enhancement in the matching between the propagating entity and the propagation target after adopting the i-th candidate meta-path. , , These represent node representation enhancement weights, association enhancement weights, and target matching weights, respectively. The degree of node representation enhancement is calculated based on the changes in node representation before and after path propagation: ; in, Indicates the path propagation before the first Each node's feature representation This indicates the first step after propagation using candidate metapaths. Each node's feature representation; The degree of enhanced entity association is represented as follows: ; in, This represents a function for calculating the feature similarity between propagating entities. Indicates the characteristics of entities propagating before the path propagation. This indicates the entity features that propagate after path propagation. To propagate the set of edges associated with entities; The degree of target matching in propagation is represented as ; in, This indicates that propagation semantic features are obtained by aggregating candidate meta-paths. Representation of the characteristics of the propagation target; Update the selection probability of each candidate metapath node during the Monte Carlo tree search process based on the path benefit evaluation value; Step S234: Based on the cumulative revenue evaluation results of each candidate path during the Monte Carlo tree search process, sort the candidate paths and generate a dynamic screening threshold based on the cumulative revenue distribution, expressed as: ; in, Indicates the path filtering threshold. This represents the average value of the earnings evaluation of all candidate metapaths. This represents the standard deviation of the earnings evaluation value. This represents the adaptive adjustment coefficient for the propagation scenario, which is dynamically determined based on the complexity of the propagation scenario. Candidate meta-paths whose path benefit evaluation value is greater than the dynamic screening threshold are selected as effective propagation semantic paths, forming a set of effective propagation semantic paths, which are used for subsequent propagation of information constraint propagation and feature fusion of heterogeneous graphs.

[0007] Furthermore, step S3 specifically includes the following steps: Step S31: Map the semantic state of the propagation entity relationship according to the propagation relationship feature representation to obtain the propagation entity relationship embedding representation. Calculate the feature similarity between any two propagation entities according to the propagation entity relationship embedding representation to generate the propagation entity association matrix. Based on the association strength between entities in the propagation entity association matrix, use a clustering algorithm to divide the entity set. Establish a cross-domain association mapping between the content relationship hypergraph and the target association hypergraph to form a dual-domain hypergraph structure. Step S32: Based on the content relationship hypergraph and the target association hypergraph, perform hypergraph convolution propagation according to the association relationship between nodes and hyperedges. Node features are aggregated in the bidirectional propagation process of "node-hyperedge-node" to obtain content relationship feature representation and target association feature representation. Step S33: Based on the content relationship feature representation and the target association feature representation, construct a dual-domain hypergraph association mechanism. By calculating the semantic consistency between the content relationship feature representation and the target association feature representation, establish semantic matching constraints between the promotional content and the dissemination target. Calculate the feature distance between the content relationship feature representation and the target association feature representation using the following formula: ; in, Indicates the semantic consistency constraint index of a two-domain hypergraph. Indicates the number of entity samples propagated. Indicates the first Content relationship feature representation corresponding to each propagation entity Indicates the first The target association feature representation corresponding to each propagation entity; Step S34: Based on the content relationship feature representation and the target association feature representation, a two-layer attention aggregation mechanism consisting of node-level attention and hyper-edge-level attention is introduced to generate high-order semantic features driven by the propagation target.

[0008] The beneficial effects achieved by the present invention using the above solution are as follows: (1) This invention uses heterogeneous propagation graphs, adaptive meta-path selection, and dual-domain hypergraph collaborative modeling to uniformly represent the complex propagation relationships between propaganda content, propagation targets, propagation channels, and users. It can fully explore the high-order semantic association information between propagation entities, improve the accuracy of propagation relationship representation and the semantic modeling ability of propagation targets, thereby improving the matching accuracy between propaganda content and propagation targets. This solves the problems of insufficient propagation relationship modeling ability and poor matching effect between propaganda content and propagation targets in the prior art. (2) By predicting the state of dissemination, optimizing the arrangement of multi-target content, and adapting to the feedback of dissemination, the dissemination content arrangement strategy is dynamically generated based on the future state of dissemination. The dissemination relationship characteristics and evaluation model parameters are continuously optimized in combination with the feedback of dissemination, so that the dissemination content can adjust the arrangement strategy in real time according to the changes in the dissemination process, thereby improving the dissemination effect of the dissemination content, the efficiency of dissemination resource utilization, and the adaptability of dissemination targets. This solves the problems of lack of dissemination state prediction and dynamic optimization capabilities and insufficient adaptive capabilities of dissemination content arrangement strategies in the existing technologies. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating an adaptive arrangement method for promotional content oriented towards communication goals, provided by the present invention.

[0010] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0011] The technical solutions proposed in this application will be further described below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments described in this application are only for explaining and illustrating the technical solutions, and are not intended to limit all implementation forms of this application. Other implementation methods obtained by those skilled in the art based on the technical content disclosed in this application without creative effort should all be covered within the protection scope of this application.

[0012] Example 1, see Figure 1 This invention provides an adaptive arrangement method for promotional content oriented towards communication goals, the method comprising the following steps: Step S1: Collect publicity data, including historical publicity content data, user interaction behavior data, dissemination channel data and dissemination target data. Extract content theme features, semantic features and dissemination attribute features from publicity content data, extract behavioral preference features from user interaction behavior data, extract channel attribute features from dissemination channel data, and extract dissemination target features from dissemination target data. Then, perform correlation and fusion to construct a multi-source dissemination dataset. Step S2: Construct a propagation heterogeneity graph based on the multi-source propagation dataset, filter effective propagation semantic paths based on the meta-path selection mechanism of Monte Carlo tree search, and generate propagation relationship feature representations using the propagation heterogeneity graph; Step S3: Calculate the association relationship between propagation entities based on the propagation relationship feature representation, construct the content relationship hypergraph and the target association hypergraph, and extract the high-order semantic features driven by the propagation target through hypergraph convolution and attention aggregation mechanism; Step S4: Based on the high-order semantic features driven by the propagation target, construct a dynamic propagation relationship graph, input the dynamic propagation relationship graph into the time series prediction model, predict the propagation state within the future time window, and obtain the prediction result of the future propagation state. Step S5: Based on the prediction results of future propagation status, construct a multi-objective evaluation model and calculate the optimal orchestration strategy; Step S6: Generate a target-matched promotional content plan based on the optimal arrangement strategy, update the dissemination relationship characteristics based on actual dissemination feedback data, and optimize the parameters of the multi-objective evaluation model.

[0013] Example 2, this example is based on the above example, step S2 specifically includes the following steps: Step S21: Perform entity parsing and relation parsing on the multi-source propagation dataset to identify entities such as propaganda content, users, propagation channels, and propagation targets, construct an entity set, and construct a propagation relation set based on the relationships between the entities; Step S22: Construct a heterogeneous propagation graph based on the entity set and the propagation relation set, map different types of entities to different types of nodes, map propagation relations to heterogeneous edges of the corresponding type, and generate initial attribute vectors for each node and edge; Step S23: Construct a candidate meta-path set based on the propagation heterogeneous graph, encode the candidate meta-paths in the candidate meta-path set, and use a meta-path selection mechanism based on Monte Carlo tree search to iteratively search the candidate meta-paths to form an effective propagation semantic path set; Step S24: Based on the set of effective propagation semantic paths, perform path constraint message propagation on the heterogeneous graph, aggregate node information along the effective propagation semantic paths, use a heterogeneous graph message passing mechanism based on meta-path constraints to perform node information propagation on the heterogeneous graph, perform cross-type feature aggregation based on the propagation relationship between different types of nodes, and perform fusion encoding based on the association relationship between different node representation vectors to generate propagation relationship feature representation. The fusion encoding process adopts the following method: ; in, This represents the characteristics of the propagation relationship. Indicates the first Node-like propagation representation, Indicates the fusion weight of node types. Indicates the effective propagation semantic path for the first Node-like propagation contribution coefficient Indicates the number of node types.

[0014] Example 3, this example is based on the above example, step S23 specifically includes the following steps: Step S231: Based on the node type, heterogeneous edge type, and propagation target characteristics in the heterogeneous propagation graph, traverse the node connection relationships in the heterogeneous propagation graph to obtain the node transfer relationships between different types of nodes. Generate a node type sequence that satisfies the node type transfer constraints based on the node transfer relationships. Combine the association relationship between the propagation target entity and the publicity content entity to perform target association filtering on the node type sequence. Combine the heterogeneous relationships corresponding to the node type sequences that satisfy the propagation target characteristics to form a candidate metapath set. Step S232: Perform path state encoding on each candidate path in the candidate path set. Based on the node type sequence, edge type sequence, path length information, and propagation target features contained in the candidate path, generate the corresponding path state vector, represented as: ; in, This represents the path state vector corresponding to the p-th candidate meta-path. This indicates the node type characteristics included in the candidate metapath. This indicates the edge type feature corresponding to the candidate metapath. This represents the path length feature of candidate meta-elements. This represents the association features between candidate metapaths and propagation targets; A meta-path search tree is constructed based on the path state vector. The candidate meta-path expansion process is mapped to search actions, and the selection of the next node type and the next propagation relationship are taken as path expansion actions. Step S233: Using candidate meta-paths as the search object, the candidate meta-paths are iteratively expanded, simulated, and evaluated for benefits through the Monte Carlo tree search mechanism; During path simulation, based on the information propagation process of the candidate path, the degree of node representation enhancement, the degree of propagation entity association enhancement, and the degree of propagation target matching are calculated, and a path benefit evaluation function is constructed: ; in, Indicates the first The path revenue evaluation value corresponding to each candidate meta-path. Indicates the use of the first The degree of enhancement of node representation after information propagation of candidate meta-paths Indicates the use of the first The degree of enhancement of the association features between entities after the propagation of candidate meta-paths. Indicates the use of the first The degree of enhancement in matching between the propagating entity and the propagation target after each candidate meta-path. , , These represent node representation enhancement weights, association enhancement weights, and target matching weights, respectively. The degree of node representation enhancement is calculated based on the changes in node representation before and after path propagation: ; in, Indicates the path propagation before the first Each node's feature representation This indicates the first step after propagation using candidate metapaths. Each node's feature representation; The degree of enhanced entity association is represented as follows: ; in, This represents a function for calculating the feature similarity between propagating entities. Indicates the characteristics of entities propagating before the path propagation. This indicates the entity features that propagate after path propagation. To propagate the set of edges associated with entities; The degree of target matching in propagation is represented as ; in, This indicates that propagation semantic features are obtained by aggregating candidate meta-paths. Representation of the characteristics of the propagation target; Update the selection probability of each candidate metapath node during the Monte Carlo tree search process based on the path benefit evaluation value; Step S234: Based on the cumulative revenue evaluation results of each candidate path during the Monte Carlo tree search process, sort the candidate paths and generate a dynamic screening threshold based on the cumulative revenue distribution, expressed as: ; in, Indicates the path filtering threshold. This represents the average value of the earnings evaluation of all candidate metapaths. This represents the standard deviation of the earnings evaluation value. This represents the adaptive adjustment coefficient for the propagation scenario, which is dynamically determined based on the complexity of the propagation scenario. Candidate meta-paths whose path benefit evaluation value is greater than the dynamic screening threshold are selected as effective propagation semantic paths, forming a set of effective propagation semantic paths, which are used for subsequent propagation of information constraint propagation and feature fusion of heterogeneous graphs.

[0015] In this embodiment, the heterogeneous propagation graph constructed in step S22 contains four types of nodes, namely, promotional content nodes. User nodes Dissemination channels and nodes and the target node for propagation Different types of nodes are connected through content association edges, user interaction edges, channel propagation edges, and target matching edges; Based on the node type information and heterogeneous edge type information in the heterogeneous graph, the connection relationships between nodes are traversed.

[0016] From the perspective of promotional content nodes Starting from the heterogeneous edge connection relationship, obtain the sequence of reachable node types, including: ; ; ; in, This indicates the propagation path that spreads to the target object through user interaction relationships. This refers to the communication path that influences the target audience through communication channels. This indicates a communication path that combines user interaction and communication channels to influence the target audience. Based on the characteristics of the propagation target, the above node type sequences are filtered, and the current propagation target characteristics are represented as follows: ; Calculate the matching degree between the propagation semantic features and the propagation target features corresponding to each candidate node type sequence, and remove node type sequences with matching degrees below a preset threshold to obtain a candidate metapath set. ; In this embodiment, the set of candidate metapaths is obtained as follows: ; ; ; ; State encoding is performed on the candidate metapaths to determine their status. For example, a path state vector is generated based on the node type, edge type, path length, and propagation target association features it contains: ; in, This indicates the node type encoding result. This indicates the encoding result of heterogeneous edge types. Indicates path length feature, This indicates the degree of correlation between the path and the dissemination target; Construct a Monte Carlo tree search structure based on the path state vectors described above; The root node represents the initial promotional content node; A tree node represents the sequence of currently selected node types; The tree edge indicates the next propagation relationship selection action; The Monte Carlo tree search algorithm is used to search for candidate metapaths.

[0017] Each round of the search process includes four stages: path selection, path expansion, path simulation, and revenue feedback. The specific process is as follows: Path selection phase: Calculate the node selection probability based on the cumulative reward value and number of visits to the current search tree node. ; in, Indicates the first Cumulative revenue from each path Indicates the first Number of visits per path This indicates the current number of times the parent node has been accessed. Indicates the exploration parameters; The next expansion path is selected based on the above selection probability. Path expansion phase: Based on the node type sequence corresponding to the current node, continue to select the next node type and propagation relationship; Current path: ; Based on the heterogeneous edge relationship extension, we obtain: ; New candidate path nodes are formed; Path simulation phase: For the expanded candidate meta-path, simulate the information propagation process on the path.

[0018] Calculate the degree of change in node representations before and after path propagation: ; in, Indicates the characteristics of nodes before propagation. This represents the node features after propagation through the candidate metapath; Calculate the degree of entity association enhancement during propagation: ; Represents the cosine similarity function; The target matching degree is calculated based on the similarity between the path semantic features and the propagation target features: ; Calculate the path revenue based on the above three indicators: ; set up ; After multiple rounds of Monte Carlo tree search iterations, the cumulative reward of each candidate metapath is calculated, yielding the following: Table 1 CUT 0.62 0.58 0.71 0.64 CDT 0.55 0.61 0.63 0.60 CUDT 0.76 0.73 0.85 0.79 therefore ; Preferred As a high-value semantic communication path; A dynamic selection threshold is generated based on the cumulative return distribution after multiple rounds of Monte Carlo tree search: ; in, This represents the average return of the candidate paths; Indicates the standard deviation of returns; This indicates parameters that are adjusted based on the complexity of the propagation scenario; When the candidate metapath payoff satisfies: When the time comes, the corresponding path will be added to the set of valid propagation semantic paths; The final result is: .

[0019] Example 4: This example is based on the above examples. Step S3 specifically includes the following steps: Step S31: Map the semantic state of the propagation entity relationship according to the propagation relationship feature representation to obtain the propagation entity relationship embedding representation. Calculate the feature similarity between any two propagation entities according to the propagation entity relationship embedding representation to generate the propagation entity association matrix. Based on the association strength between entities in the propagation entity association matrix, use a clustering algorithm to divide the entity set. Establish a cross-domain association mapping between the content relationship hypergraph and the target association hypergraph to form a dual-domain hypergraph structure. Step S32: Based on the content relationship hypergraph and the target association hypergraph, perform hypergraph convolution propagation according to the association relationship between nodes and hyperedges. Node features are aggregated in the bidirectional propagation process of "node-hyperedge-node" to obtain content relationship feature representation and target association feature representation. Step S33: Based on the content relationship feature representation and the target association feature representation, construct a dual-domain hypergraph association mechanism. By calculating the semantic consistency between the content relationship feature representation and the target association feature representation, establish semantic matching constraints between the promotional content and the dissemination target. Calculate the feature distance between the content relationship feature representation and the target association feature representation using the following formula: ; in, Indicates the semantic consistency constraint index of a two-domain hypergraph. Indicates the number of entity samples propagated. Indicates the first Content relationship feature representation corresponding to each propagation entity Indicates the first The target association feature representation corresponding to each propagation entity; Step S34: Based on the content relationship feature representation and the target association feature representation, a two-layer attention aggregation mechanism consisting of node-level attention and hyper-edge-level attention is introduced to generate high-order semantic features driven by the propagation target.

[0020] Example 5, this example is based on the above examples, step S31 specifically includes the following: Based on the propagation relationship feature representation, the similarity of the feature vectors corresponding to each propagation entity in the propagation heterogeneity graph is calculated to obtain the association strength between propagation entities; Construct a propagation entity association matrix based on the association strength, read the matrix row vector corresponding to each propagation entity in the propagation entity association matrix, and divide the propagation entities with completely identical matrix row vectors into the same entity set. Each entity set is mapped to a hyperedge in the hypergraph, and a content relation hypergraph and a target association hypergraph are constructed respectively. The content relation hypergraph describes the information relationship between different promotional content entities based on the relationship between promotional content entities, and the target association hypergraph describes the relationship between promotional content entities and promotional target entities based on the relationship between promotional content entities, user entities, and dissemination target entities.

[0021] Example 6, this example is based on the above examples, step S32 specifically includes the following: Based on the connection relationships between nodes and hyperedges in the content relation hypergraph and the target association hypergraph, node-hyperedge association matrices are constructed respectively. Based on the node-hyperedge association matrices, node features are aggregated into hyperedge features. The node features corresponding to the propagating entities within the same hyperedge are fused to generate hyperedge association features. The hyperedge association features are fed back to the associated nodes, and the node features participating in the same propagation relationship are updated to obtain the content relation feature representation corresponding to the content relation hypergraph and the target association feature representation corresponding to the target association hypergraph.

[0022] Example 7: Based on the above examples, step S34 introduces a two-layer attention aggregation mechanism consisting of node-level attention and hyper-edge-level attention to generate high-order semantic features driven by the propagation target. Specifically, it includes the following: Based on the content relationship feature representation and the target association feature representation, the feature contribution weights between nodes and between hyperedges are calculated respectively. The aggregation ratio of different propagating entity features within the same hyperedge is adjusted according to the node contribution weight to obtain node-level enhanced features. The influence of different hyperedges on the target association feature update process is adjusted according to the hyperedge contribution weight to obtain hyperedge-level enhanced features. The node-level enhanced features and hyperedge-level enhanced features are fused to generate high-order semantic features driven by the propagation target.

[0023] Example 8: This example is based on the above examples. Step S5 specifically includes the following steps: Step S51: Using the prediction results of future dissemination status as dynamic constraints, construct a set of evaluation indicators for the arrangement of promotional content, including dissemination effect, user participation, dissemination cost, and dissemination goal matching degree; Step S52: Based on the set of evaluation indicators for promotional content arrangement, encode the type of promotional content, content combination method, dissemination channel, and release time to generate candidate promotional content arrangement individuals. With the optimization objectives of maximizing dissemination effect, maximizing user participation, minimizing dissemination cost, and maximizing target matching degree, construct a multi-objective content arrangement optimization model oriented towards dissemination goals: ; in, This indicates the arrangement plan for the promotional content. , , , These represent the evaluation scores for dissemination effectiveness, user engagement, dissemination cost, and dissemination goal alignment, respectively. This indicates the weight of the corresponding evaluation indicator; Step S53: Based on the conflict relationship between dissemination effect, user participation, dissemination cost, and dissemination goal matching degree, divide the individual candidate promotional content arrangement into multiple collaborative optimization subgroups; Step S54: During the iteration process of each collaborative optimization subgroup, the candidate schemes are screened according to the non-dominated relationship between the candidate propaganda content arrangement schemes and the target space distribution to obtain the non-dominated solution set; Step S55: Based on the non-dominated solution set, comprehensively evaluate the candidate publicity content arrangement schemes, use a weighted evaluation function, and select the scheme with the highest comprehensive evaluation value as the final publicity content arrangement strategy.

[0024] It should be noted that when using sequential descriptions such as "first" and "second" in this application, they are only used to distinguish different technical features, processing objects, or execution steps, and do not indicate a specific sequential or hierarchical relationship between related technical features, processing objects, or execution steps. Unless otherwise expressly defined, open-ended expressions such as "including" and "comprising" used in this application indicate that the listed technical elements are not all limited, and may also include other technical elements not explicitly listed, or other related structures necessary to achieve the technical solution of this application.

[0025] The technical solution of this application has been described above with reference to specific embodiments. However, the above embodiments are only used to explain the technical concept of this application and do not constitute a limitation on the scope of protection of this application. Based on an understanding of the technical solution of this application, any equivalent substitutions, adjustments, or improvements made to the relevant technical features without departing from the technical purpose and principles of this application should be included within the scope of protection defined by the claims of this application.

[0026] It should be noted that the structural form shown in the accompanying drawings is only one specific implementation of the technical solution of this application. In actual application, it can be adapted to the application scenario, equipment conditions, and technical requirements. Therefore, without changing the core technical idea of ​​this application, reasonable modifications, combinations, or optimizations made by those skilled in the art based on the disclosure of this application are all within the scope of the technical solution of this application.

Claims

1. An adaptive arrangement method for promotional content oriented towards communication objectives, characterized in that: The method includes the following steps: Step S1: Collect publicity data, extract content theme features, semantic features and dissemination attribute features from publicity content data, extract behavioral preference features from user interaction behavior data, extract channel attribute features from dissemination channel data, extract dissemination target features from dissemination target data, and perform correlation and fusion to construct a multi-source dissemination dataset; Step S2: Construct a propagation heterogeneity graph based on the multi-source propagation dataset, filter effective propagation semantic paths based on the meta-path selection mechanism of Monte Carlo tree search, and generate propagation relationship feature representations using the propagation heterogeneity graph; Step S3: Calculate the association relationship between propagation entities based on the propagation relationship feature representation, construct the content relationship hypergraph and the target association hypergraph, and extract the high-order semantic features driven by the propagation target through hypergraph convolution and attention aggregation mechanism; Step S4: Based on the high-order semantic features driven by the propagation target, construct a dynamic propagation relationship graph, input the dynamic propagation relationship graph into the time series prediction model, predict the propagation state within the future time window, and obtain the prediction result of the future propagation state. Step S5: Based on the prediction results of future propagation status, construct a multi-objective evaluation model and calculate the optimal orchestration strategy; Step S6: Generate a target-matched promotional content plan based on the optimal arrangement strategy, update the dissemination relationship characteristics based on actual dissemination feedback data, and optimize the parameters of the multi-objective evaluation model.

2. The adaptive arrangement method for propaganda content oriented towards communication goals according to claim 1, characterized in that: Step S2 specifically includes the following steps: Step S21: Perform entity parsing and relation parsing on the multi-source propagation dataset to identify entities such as propaganda content, users, propagation channels, and propagation targets, construct an entity set, and construct a propagation relation set based on the relationships between the entities; Step S22: Construct a heterogeneous propagation graph based on the entity set and the propagation relation set, map different types of entities to different types of nodes, map propagation relations to heterogeneous edges of the corresponding type, and generate initial attribute vectors for each node and edge; Step S23: Construct a candidate meta-path set based on the propagation heterogeneous graph, encode the candidate meta-paths in the candidate meta-path set, and use a meta-path selection mechanism based on Monte Carlo tree search to iteratively search the candidate meta-paths to form an effective propagation semantic path set; Step S24: Based on the set of effective propagation semantic paths, perform path constraint message propagation on the heterogeneous graph, aggregate node information along the effective propagation semantic paths, use a heterogeneous graph message passing mechanism based on meta-path constraints to perform node information propagation on the heterogeneous graph, perform cross-type feature aggregation based on the propagation relationship between different types of nodes, and perform fusion encoding based on the association relationship between different node representation vectors to generate propagation relationship feature representation. The fusion encoding process adopts the following method: ； in, This represents the characteristics of the propagation relationship. Indicates the first Node-like propagation representation, Indicates the fusion weight of node types. Indicates the effective propagation semantic path for the first Node-like propagation contribution coefficient Indicates the number of node types.

3. The adaptive arrangement method for propaganda content oriented towards communication goals according to claim 1, characterized in that: Step S3 specifically includes the following steps: Step S31: Map the semantic state of the propagation entity relationship according to the propagation relationship feature representation to obtain the propagation entity relationship embedding representation. Calculate the feature similarity between any two propagation entities according to the propagation entity relationship embedding representation to generate the propagation entity association matrix. Based on the association strength between entities in the propagation entity association matrix, use a clustering algorithm to divide the entity set. Establish a cross-domain association mapping between the content relationship hypergraph and the target association hypergraph to form a dual-domain hypergraph structure. Step S32: Based on the content relationship hypergraph and the target association hypergraph, perform hypergraph convolution propagation according to the association relationship between nodes and hyperedges. Node features are aggregated in the bidirectional propagation process of "node-hyperedge-node" to obtain content relationship feature representation and target association feature representation. Step S33: Based on the content relationship feature representation and the target association feature representation, construct a dual-domain hypergraph association mechanism. By calculating the semantic consistency between the content relationship feature representation and the target association feature representation, establish semantic matching constraints between the promotional content and the dissemination target. Calculate the feature distance between the content relationship feature representation and the target association feature representation using the following formula: ； in, Indicates the semantic consistency constraint index of a two-domain hypergraph. Indicates the number of entity samples propagated. Indicates the first Content relationship feature representation corresponding to each propagation entity Indicates the first The target association feature representation corresponding to each propagation entity; Step S34: Based on the content relationship feature representation and the target association feature representation, a two-layer attention aggregation mechanism consisting of node-level attention and hyper-edge-level attention is introduced to generate high-order semantic features driven by the propagation target.

Images