Heterogeneous network influence maximization method and system based on meta-path semantic perception

Abstract

The present invention discloses a method for maximizing the influence of a heterogeneous network based on meta-path semantic perception, comprising the following steps: constructing a heterogeneous network graph based on heterogeneous network data; obtaining the meta-path of the heterogeneous network graph, simplifying and integrating it, and generating a valid path set; setting node path weights based on a specific heterogeneous network graph, determining the node sampling probability, and constructing a reverse sampling strategy; adopting the reverse sampling strategy to sample the heterogeneous network graph, and generating a valid reverse reachable set in combination with the determination of the valid path; based on the valid reverse reachable set, calculating the node marginal influence propagation gain, iteratively screening with the marginal gain as an indicator, obtaining a target seed set, and achieving influence maximization. The present invention constructs a reverse sampling strategy, combines the node semantics and path semantics in the heterogeneous network in the reverse sampling process, takes semantic perception into account, and helps to more accurately identify and select key nodes, thereby achieving the maximization of the influence of the heterogeneous network.

Description

Technical Field

[0001] The present invention relates to the field of information dissemination technology, and in particular to a method for maximizing the influence of heterogeneous networks based on meta-path semantic perception. Background Art

[0002] With the rapid development of the internet and social media, social networks have become essential platforms for communication and information dissemination in our daily lives. Maximizing information dissemination and product promotion within social networks has become a hot topic of research, leading to the development of influence maximization. Its core focus is identifying key nodes within complex social networks and leveraging these nodes to maximize the spread of information or behavior. Currently, influence maximization has widespread applications in a variety of fields, including viral marketing, information dissemination, and disease prevention and control.

[0003] Traditional social network analysis typically considers networks to be homogeneous, consisting of nodes and edges of the same type. However, real-life social networks are often heterogeneous, consisting of nodes and edges of different types. For example, nodes include users, posts, and comments, while edges include likes, reposts, and top-ups. Compared to homogeneous networks, heterogeneous networks possess richer structural and semantic information, and can more accurately reflect the complex relationships and behaviors within social networks.

[0004] However, existing research on heterogeneous network influence ignores the importance of semantic differences in meta-paths for influence mining, and therefore cannot accurately identify key nodes, making it difficult to accurately conduct network influence analysis. Summary of the Invention

[0005] The purpose of the present invention is to solve the problem that the existing technology does not consider the semantic weights of different meta-paths, resulting in inaccurate generated seed sets with the greatest influence. It provides a method and system for maximizing the influence of heterogeneous networks based on meta-path semantic perception. By constructing a reverse sampling strategy, it can more accurately evaluate the influence differences of different nodes on the source node based on node semantics and path semantics, which helps to accurately identify key nodes and maximize network influence.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] A method for maximizing the influence of heterogeneous networks based on meta-path semantic perception includes the following steps:

[0008] Construct heterogeneous network graphs based on heterogeneous network data;

[0009] Obtain the meta-paths of the heterogeneous network graph, simplify and integrate them, and generate a set of valid paths;

[0010] According to the specific heterogeneous network graph, the node path weight is set, the node sampling probability is determined, and the reverse sampling strategy is constructed;

[0011] A reverse sampling strategy is used to sample heterogeneous network graphs, and combined with the determination of effective paths, an effective reverse reachable set is generated;

[0012] Based on the effective reverse reachable set, the marginal influence propagation gain of the node is calculated, and iterative screening is performed with the marginal gain as the indicator to obtain the target seed set and maximize the influence.

[0013] Furthermore, a heterogeneous network graph is constructed based on the heterogeneous network data, specifically:

[0014] Define nodes and edges based on the specific scenarios of heterogeneous networks. In social network scenarios, node types include at least two of the following: users, posts, tags, and transactions; edge types include at least two of the following: posts, votes, associated tags, transfers, and follows.

[0015] The heterogeneous network graph expression is ,in, V Represents a node set; Q Represents an edge set; the node type mapping function is , the edge type mapping function is , A Represents a collection of node types, R Represents a set of edge types; the number of node types , the number of edge types .

[0016] Furthermore, the meta-paths of the heterogeneous network graph are obtained and simplified and integrated to generate a valid path set, specifically:

[0017] Get all the meta-paths of the heterogeneous network and set the effective path core length to l , for lengths exceeding l The meta path only retains the head and tail l / 2 length part to get the valid path set.

[0018] Furthermore, according to the specific heterogeneous network graph, the node path weights are set, the node sampling probability is determined, and a reverse sampling strategy is constructed, specifically:

[0019] For heterogeneous networks , the reverse sampling strategy is expressed as ;in m represents the path weight vector, is a dimensional vector, the weight Node-based The generated reverse reachable set;n represents the sampling probability vector, is a dimensional vector, representing the probability of a node being sampled in a heterogeneous network; n Satisfaction: For ,have and ;

[0020] Path weight vector m The role of is to assign weights to different nodes when calculating the influence coverage function to reflect the semantic importance of the nodes; the sampling probability vector n The role of is to decide which nodes to start from to build the reverse reachable set.

[0021] Furthermore, the path weight vector m and the sampling probability vector n The generation of is based on the semantic information of heterogeneous networks and a predefined meta-path weight strategy. The semantic information includes node types and path types.

[0022] Path weight vector m The generation basis is based on the predefined meta-path weight strategy. The meta-path in the social network is user U-post P-user U, which means that user 1 affects user 2 through post P. The path weight of this UPU meta-path is assigned; the sampling probability vector n The generation basis is based on the importance of node types, giving priority to target node types, assigning probability values ​​to each node in the node set, and allocating higher sampling probabilities to target nodes based on the node type distribution.

[0023] Furthermore, a reverse sampling strategy is used to sample the heterogeneous network graph, and combined with the determination of valid paths, a valid reverse reachable set is generated, specifically:

[0024] Heterogeneous network graph G Any target node in , the effective reverse reachable set is the subgraph Reaching the node via a valid path v A node set where the subgraph From the heterogeneous network graph obtained by sampling;

[0025] When the node u Appears in the valid reverse reachable set In the middle, it represents the node u Activate Node v The possibility of; after sufficient sampling, a set consisting of multiple valid reverse reachable sets is obtained RR , different seed sets S The combined influence of the seed set S With collectionRR The intersection of; After sufficient sampling, a set consisting of multiple valid reverse reachable sets is obtained RR ,Then in each round of seed set selection process, the seed nodes of this round are selected by marginal gain sorting.

[0026] Furthermore, a reverse sampling strategy is used to sample the heterogeneous network graph, and combined with the determination of valid paths, a valid reverse reachable set is generated, including:

[0027] Based on reverse sampling strategy The sampling probability vector in n , from the heterogeneous network graph G Node sampling is performed in the ;

[0028] In the sampling diagram In the example, the node can be reached through a valid path. v The node set of , where the nodes v is the root node; when generating a valid reverse reachable set, the node to be added and the node v The valid path type between nodes and the valid path weight corresponding to the node are assigned;

[0029] According to the reverse sampling strategy The path weight vector in m Compute the influence weighted cover function for all valid reverse reachable sets.

[0030] Furthermore, the influence weighted coverage function is used to calculate the seed set S exist The average weighted cover in the valid reverse reachable set is defined as:

[0031] ;

[0032] Where, w i Indicates the i Valid Reverse Reachable Set R i The weight of x i Indicates the i Valid Reverse Reachable Set R i With seed set S The intersection of hour ,otherwise ;

[0033] For a given reverse sampling strategy , any seed set and the number of reverse reachable sets , its influence weighted coverage function satisfies:

[0034] ;

[0035] set up For seed collection S The expected impact under influence spread, if is an effective reverse sampling strategy, then there exists a constant ,satisfy:

[0036] ;

[0037] Where, Indicates expected impact; Represents a random variable, reflecting the seed set S The scope of influence spread, that is, the number of nodes that are eventually activated; Indicates the preset coefficient; represents the single effective reverse reachable set cover estimator, that is, the weighted cover indicator value based on the random effective reverse reachable set; represents the expected value of a single valid reverse reachable set cover estimator.

[0038] Furthermore, based on the effective reverse reachable set, by calculating the node marginal influence propagation gain, iterative screening is performed using marginal gain as an indicator to obtain the target seed set and maximize the influence, specifically:

[0039] generate After a valid reverse reachable set, the greedy strategy is used to complete the k In the selection of seed nodes, in each round of seed set selection, the seed nodes of this round are selected by marginal gain sorting until the k The number of seed nodes required;

[0040] In the seed set selection phase, nodes are selected iteratively in each round to form k Bit seed set S ; In each iteration, the selection is selected The one that will have the greatest impact later Nodes with marginal gains ,in represents the final seed set; ;

[0041] The seed set selection process is regarded as a method to solve the weighted maximum cover problem. RR The size of the collection is increased by continuously trying RR The size of of Approximate solution.

[0042] A heterogeneous network influence maximization system based on meta-path semantic perception, applying any of the above-mentioned heterogeneous network influence maximization methods based on meta-path semantic perception, comprising:

[0043] A graph construction unit, used for constructing a heterogeneous network graph based on heterogeneous network data;

[0044] The path generation unit is used to obtain the meta-paths of the heterogeneous network graph, simplify and integrate them, and generate a valid path set;

[0045] A strategy construction unit is used to set node path weights, determine node sampling probabilities, and construct a reverse sampling strategy based on a specific heterogeneous network graph;

[0046] The sampling unit is used to sample the heterogeneous network graph using a reverse sampling strategy and generate a valid reverse reachable set based on the determination of valid paths;

[0047] The screening unit is used to calculate the node marginal influence propagation gain based on the effective reverse reachable set, and iteratively screen with the marginal gain as the indicator to obtain the target seed set and maximize the influence.

[0048] Compared with the existing technology, the present invention constructs a reverse sampling strategy, combines the node semantics and path semantics in the heterogeneous network during the reverse sampling process, takes into account semantic perception, and helps to more accurately identify and select key nodes, thereby maximizing the influence of the heterogeneous network. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Flowchart of the heterogeneous network influence maximization method based on meta-path semantic perception.

[0050] Figure 2 Schematic diagram of the heterogeneous network and the set path sampling strategy function.

[0051] Figure 3 Schematic diagram of sampling graph, reverse reachable set and path score. DETAILED DESCRIPTION

[0052] The following further illustrates the heterogeneous network influence maximization method and system based on meta-path semantic perception of the present invention with reference to the accompanying drawings and specific embodiments.

[0053] See also Figure 1 The present invention discloses a method for maximizing the influence of heterogeneous networks based on meta-path semantic perception, comprising the following steps:

[0054] S1: Construct a heterogeneous network graph based on heterogeneous network data;

[0055] S2: Obtain the meta-paths of the heterogeneous network graph, simplify and integrate them, and generate a set of valid paths;

[0056] S3: According to the specific heterogeneous network graph, the node path weight is set, the node sampling probability is determined, and the reverse sampling strategy is constructed;

[0057] S4: Use the reverse sampling strategy to sample the heterogeneous network graph and generate a valid reverse reachable set based on the determination of valid paths;

[0058] S5: Based on the effective reverse reachable set, by calculating the node marginal influence propagation gain, iterative screening is performed using marginal gain as an indicator to obtain the target seed set and maximize the influence.

[0059] The present invention can be applied to social networks, where a variety of node and relationship types exist, all of which can be used to enrich the network's semantics. This is illustrated using a specific example: Steemit is a blockchain-based social platform where users interact through actions such as posting, commenting, voting, and transacting. These diverse entities and relationships naturally form a heterogeneous network. The node types include user U, post P, tag T, and transaction TX. The edge types include user posting UP, user voting UP, post-associated tag PT, user transfer U-TX-U, and user following UU.

[0060] Once the target seed set is found, it can be used to efficiently maximize influence. For example, if the target seed set represents a set of user nodes, posts or advertisements can be posted through these user nodes to efficiently spread these posts or advertisements to the widest possible range, maximizing influence.

[0061] This invention establishes a valid path set based on a specific heterogeneous network and introduces a reverse sampling strategy. When creating a reverse reachable set, the path weight experienced by each node upon joining the reverse reachable set (RRset) is recorded. This mechanism enables a more accurate assessment of the influence of different nodes on the source node based on node and path semantics when constructing the RRset. This allows for more precise identification of key nodes, helping to maximize their influence.

[0062] Step S1: constructing a heterogeneous network graph in a heterogeneous network graph according to heterogeneous network data.

[0063] In this embodiment, the definitions of nodes and edges are set according to the specific scenarios of the heterogeneous network: for example, when studying the influence of users in a social network, users can be used as nodes and the interactions between users (such as following, liking, etc.) as edges to construct a heterogeneous network graph; when studying the influence of users' posts, posts can be used as nodes and the following, commenting, and other behaviors between users can be used as edges to construct a heterogeneous network graph.

[0064] The heterogeneous network graph expression is ,in, V Represents a node set; Q Represents an edge set; the node type mapping function is , the edge type mapping function is , A Represents a collection of node types, R Represents a set of edge types; the number of node types , the number of edge types .

[0065] Step S2: Obtain the meta-paths of the heterogeneous network graph, simplify and integrate them, and generate a valid path set.

[0066] After obtaining all valid meta-path types in the heterogeneous network, simplify and integrate them to obtain a valid path set. Specifically, the core length of the valid path is preset. l , that is, for lengths exceeding l The path of only retaining its head and tail approximations l / 2 The complete part of the length is used to obtain the valid path set.

[0067] Step S3: According to the specific heterogeneous network graph, the node path weights are set, the node sampling probability is determined, and a reverse sampling strategy is constructed.

[0068] For heterogeneous networks , the reverse sampling strategy is expressed as .in m represents the weight vector, is a dimensional vector, the weight Node-based The generated reverse reachable set RRsets; n represents the sampling probability vector, is a dimensional vector, representing the probability of a node being sampled in a heterogeneous network; n Satisfaction: For ,have and Path weight vector mThe role of is to assign weights to different nodes when calculating the influence coverage function, reflecting the semantic importance of the nodes. Sampling probability vector n The role of is to decide which nodes to start from to build the reverse reachable set.

[0069] The generation of weight vectors and sampling probability vectors is based on the semantic information of the heterogeneous network (node ​​type, path type) and a predefined meta-path weighting strategy. For example, m is generated based on predefined meta-path weights (for example, in a social network, the path weight of the meta-path (user U - post P - user U) can be assigned 0.4, indicating that user 1 influences user 2 through post P). n is generated based on the importance of node type, prioritizing target node types and assigning a probability value to each node in the node set. Based on the node type distribution, target nodes are given a higher sampling probability.

[0070] Step S4: Use the reverse sampling strategy to sample the heterogeneous network graph, and combine the determination of valid paths to generate a valid reverse reachable set, specifically:

[0071] Heterogeneous network graph G Any target node in , the valid reverse reachable set (VRR) is a subgraph Reaching the node via a valid path v A node set where the subgraph From the heterogeneous network graph obtained by sampling in .

[0072] When the node u Appears in the valid reverse reachable set In the middle, it represents the node u Activate Node v After sufficient sampling, we can obtain a set of multiple valid reverse reachable sets. RR , different seed sets S The combined influence of the seed set S With collection RR Therefore, the main process of the heterogeneous reverse influence sampling algorithm is: first perform a sufficient amount of sampling, and then obtain a set consisting of multiple valid reverse reachable sets. RR ,Then in each round of seed set selection process, the seed nodes of this round are selected by marginal gain sorting.

[0073] Step S4: adopt the reverse sampling strategy to sample the heterogeneous network graph, combine the determination of valid paths, and generate a valid reverse reachable set, including:

[0074] S41: Based on reverse sampling strategy The sampling probability vector inn , from the heterogeneous network graph G Node sampling is performed in the ;

[0075] S42: In the sampling diagram In the example, the node can be reached through a valid path. v The node set of , where the nodes v is the root node. The effective reverse reachable set generation algorithm is as follows Figure 2 As shown, unlike the classic reverse reachable set, the algorithm records the nodes to be added and the nodes when generating the valid reverse reachable set. v The valid path type between nodes and the valid path weight corresponding to the node are assigned;

[0076] S43: According to the reverse sampling strategy The path weight vector in m Compute the influence weighted cover function for all valid reverse reachable sets.

[0077] The influence weighted cover function is used to calculate the seed set S exist The average weighted cover in the valid reverse reachable set is defined as:

[0078] ;

[0079] Where, w i Indicates the i Valid Reverse Reachable Set R i The weight of x i is an indicator variable, x i Indicates the i Valid Reverse Reachable Set R i With seed set S The intersection of hour ,otherwise ; R i Indicates a sampling strategy based on The first set in the sequence of generated valid reverse reachable sets i indivual.

[0080] For a given reverse sampling strategy , any seed set and the number of reverse reachable sets , its influence weighted coverage function satisfies:

[0081] ;

[0082] set up For seed collection S The expected impact under the propagation model is is an effective reverse sampling strategy, then there exists a constant ,satisfy:

[0083] ;

[0084] Where, Indicates expected impact; Represents a random variable, reflecting the seed set S The scope of influence spread, that is, the number of nodes that are eventually activated; Indicates the preset coefficient; represents the single effective reverse reachable set cover estimator, that is, the weighted cover indicator value based on the random effective reverse reachable set; represents the expected value of a single valid reverse reachable set cover estimator.

[0085] In step S5, based on the effective reverse reachable set, the marginal influence propagation gain of the node is calculated, and iterative screening is performed using the marginal gain as an indicator to obtain the target seed set and maximize the influence.

[0086] Based on the effective reverse reachable set, the marginal influence propagation gain of the nodes is calculated and iterative screening is performed using the marginal gain as an indicator to obtain the target seed set and maximize the influence. Specifically:

[0087] When generated After valid reverse reachable sets, this embodiment uses a greedy strategy to complete the k In each round of seed set selection, the seed nodes of this round are selected by marginal gain sorting until the k The number of seed nodes required.

[0088] In the seed set selection phase, nodes are selected iteratively in each round to form k Bit seed set S In each iteration, the selection is selected The one that will have the greatest impact later Nodes with marginal gains ,in represents the final seed set; , Indicates selecting the node that can bring the greatest marginal gain of influence.

[0089] The seed set selection process can be regarded as a method to solve the weighted maximum cover problem. RR The growth of size, that is, the method of the present invention can be achieved by continuously trying to collect RR The size of of Approximate solution.

[0090] The main steps of the method of the present invention include an initialization stage, a sampling iterative optimization stage and a seed selection stage.

[0091] The following is a specific application example illustrating the proposed method. Using a real heterogeneous network dataset, the proposed method (HIN-based Reverse Influence Sampling, HRIS) is compared with CELF++ (Cost-Effective Lazy Forward++), a classic optimization algorithm for influence maximization, used to efficiently solve the seed selection problem for greedy strategies. The effectiveness and efficiency of the algorithms are comprehensively evaluated.

[0092] The academic network dataset (ACM dataset) was selected as a real-world heterogeneous network dataset. CELF++ was used as the benchmark algorithm to evaluate algorithm performance. This benchmark algorithm is applicable to both independent cascade models and linear threshold models. The experimental environment was based on a 64-bit Windows 10 Professional operating system, an Intel(R) Core(TM) i7-9750H CPU, 32GB of RAM, a GTX 1660 Ti graphics card, Python 3.8, and the PyCharm compiler.

[0093] In terms of algorithm parameters, the settings are as follows: For all methods based on Reverse Influence Sampling (RIS): set the approximation ratio parameter , which means that a 10% gap is allowed between the algorithm result and the optimal solution; setting ,in δ represents the failure probability parameter, z Indicates the number of network nodes (usually δ Set it to a smaller value, such as 1 / z , so the failure probability decreases as the network size increases; in experiments, it is usually fixed and δ ,like =0.1, δ =1 / n, to balance efficiency and accuracy).

[0094] In metapath2vec-IM, academic networks use the "APA" (author-article-author) path for path assignment. In heterogeneous academic networks, target and seed nodes are author nodes, and node semantics are not considered in influence propagation experiments.

[0095] In terms of propagation model, for independent cascade model, set the edge probability in the network ,in Representative Node v For the linear threshold (LT) model, the activation threshold of the nodes in the network is randomly assigned, and the edge weight is set to .

[0096] Experimental results show that under the independent cascade model, this algorithm has a wider influence spread range on real data sets and a shorter running time than CELF++, demonstrating good operating efficiency.

[0097] like Figure 3 As shown, the present invention also discloses a heterogeneous network influence maximization system based on meta-path semantic perception, comprising:

[0098] A graph construction unit is used to construct a heterogeneous network graph based on heterogeneous network data;

[0099] The path generation unit is used to obtain the meta-paths of the heterogeneous network graph, simplify and integrate them, and generate a valid path set;

[0100] A strategy construction unit is used to set node path weights, determine node sampling probabilities, and construct a reverse sampling strategy based on a specific heterogeneous network graph;

[0101] The sampling unit is used to sample the heterogeneous network graph using a reverse sampling strategy and generate a valid reverse reachable set based on the determination of valid paths;

[0102] The screening unit is used to calculate the marginal influence propagation gain of the nodes based on the effective reverse reachable set, and iteratively screen using the marginal gain as an indicator to obtain the target seed set and maximize the influence. The heterogeneous network influence maximization system based on meta-path semantics perception of the present invention can implement the heterogeneous network influence maximization method based on meta-path semantics perception of the present invention, and can perform any combination of the implementation steps of the method embodiment, and have the corresponding functions and beneficial effects of the method.

[0103] The present invention also discloses an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the steps described above in the method for maximizing the influence of a heterogeneous network based on meta-path semantics awareness. The electronic device of the present invention can execute the method for maximizing the influence of a heterogeneous network based on meta-path semantics awareness of the present invention, and can execute any combination of the steps of the method embodiments, thereby achieving the corresponding functions and beneficial effects of the method.

[0104] The present invention also discloses a storage medium. The computer-readable storage medium stores a computer program. When executed by a processor, the computer program implements any of the steps described above in the method for maximizing the influence of a heterogeneous network based on meta-path semantics awareness. The computer-readable storage medium of the present invention can execute the method for maximizing the influence of a heterogeneous network based on meta-path semantics awareness of the present invention, and can execute any combination of the steps of the method embodiments, thereby achieving the corresponding functions and beneficial effects of the method.

[0105] Although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It will also be understood that a detailed discussion of the actual implementation of each module is not necessary for understanding the present invention. More specifically, given the properties, functions, and internal relationships of the various functional modules in the system disclosed herein, the actual implementation of the module will be understood within the routine skills of an engineer. Therefore, a person skilled in the art will be able to implement the present invention set forth in the claims using ordinary techniques without undue experimentation. It will also be understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0106] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0107] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting, or processing it in another suitable manner as necessary, and then storing it in a computer memory.

[0108] Various components of the present invention may be implemented using hardware, software, firmware, or a combination thereof. In the aforementioned embodiments, multiple steps or methods may be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, any one of the following technologies known in the art, or a combination thereof, may be used: a discrete logic circuit having logic gates for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gates, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0109] The above description is a detailed description of the preferred embodiments of the present invention, but the embodiments are not intended to limit the scope of the patent application of the present invention. Any equivalent changes or modifications made under the technical spirit disclosed by the present invention should fall within the patent scope covered by the present invention.

Claims

1. A method for maximizing the influence of heterogeneous networks based on meta-path semantic perception, characterized by: The following steps are involved: Construct heterogeneous network graphs based on heterogeneous network data; Obtain the meta-paths of the heterogeneous network graph, simplify and integrate them, and generate a set of valid paths; According to the heterogeneous network graph, the node path weights are set, the node sampling probability is determined, and the reverse sampling strategy is constructed; A reverse sampling strategy is used to sample heterogeneous network graphs, and combined with the determination of effective paths, an effective reverse reachable set is generated; Based on the effective reverse reachable set, the marginal influence propagation gain of the nodes is calculated, and iterative screening is performed using the marginal gain as an indicator to obtain the target seed set and maximize the influence; According to the heterogeneous network data, a heterogeneous network graph is constructed, specifically: Define nodes and edges based on the specific scenarios of heterogeneous networks: In social network scenarios, node types include at least two of the following: users, posts, tags, and transactions; edge types include at least two of the following: posts, votes, associated tags, transfers, and follows; The heterogeneous network graph expression is ,in, V Represents a node set; Q Represents an edge set; the node type mapping function is , the edge type mapping function is , A Represents a collection of node types, R Represents a set of edge types; the number of node types , the number of edge types ; The reverse sampling strategy is used to sample the heterogeneous network graph, and combined with the determination of effective paths, an effective reverse reachable set is generated, specifically: Heterogeneous network graph G Any target node in , the valid reverse reachable set is the subgraph Reaching the node via a valid path v A node set where the subgraph From the heterogeneous network graph obtained by sampling; When the node u Appears in the valid reverse reachable set In the middle, it represents the node u Activate Node v The possibility of; after sufficient sampling, a set consisting of multiple valid reverse reachable sets is obtained RR , different seed sets S The combined influence of the seed set S With collection RR The intersection of; After sufficient sampling, a set consisting of multiple valid reverse reachable sets is obtained RR ,Then in each round of seed set selection process, the seed nodes of this round are selected by marginal gain sorting; Based on reverse sampling strategy The sampling probability vector in n , from the heterogeneous network graph G Node sampling is performed in the ; In the sampling diagram In the example, the node can be reached through a valid path. v The node set of , where the nodes v is the root node; when generating a valid reverse reachable set, the node to be added and the node v The valid path type between nodes and the valid path weight corresponding to the node are assigned; According to the reverse sampling strategy The path weight vector in m Calculate the influence weighted covering function for all valid reverse reachable sets; Based on the effective reverse reachable set, the marginal influence propagation gain of the nodes is calculated and iterative screening is performed using the marginal gain as an indicator to obtain the target seed set and maximize the influence. Specifically: generate After a valid reverse reachable set, the greedy strategy is used to complete the k In the selection of seed nodes, in each round of seed set selection, the seed nodes of this round are selected by marginal gain sorting until the k The number of seed nodes required; In the seed set selection phase, nodes are selected iteratively in each round to form k Bit seed set S ; In each iteration, the selection is selected The one that will have the greatest impact later Nodes with marginal gains ,in represents the final seed set; , Indicates selecting the node that can bring the greatest marginal gain of influence; The seed set selection process is regarded as a method to solve the weighted maximum cover problem. RR The size of the collection is increased by continuously trying RR The size of of Approximate solution; Indicates expected impact; Represents a random variable, reflecting the seed set S The scope of influence propagation, that is, the number of nodes that are finally activated.

2. The method for maximizing heterogeneous network influence based on meta-path semantic perception according to claim 1 is characterized in that: Obtain the meta-paths of the heterogeneous network graph, simplify and integrate them, and generate a valid path set, specifically: Get all the meta-paths of the heterogeneous network and set the effective path core length to l , for lengths exceeding l The meta path only retains the head and tail l / 2 length part to get the valid path set.

3. The method for maximizing heterogeneous network influence based on meta-path semantic perception according to claim 1 is characterized in that: According to the heterogeneous network graph, the node path weights are set, the node sampling probability is determined, and the reverse sampling strategy is constructed. Specifically: For heterogeneous networks , the reverse sampling strategy is expressed as ;in m represents the path weight vector, is a dimensional vector, the weight Node-based The generated reverse reachable set; n represents the sampling probability vector, is a dimensional vector, representing the probability of a node being sampled in a heterogeneous network; n Satisfaction: For ,have and ; Path weight vector m The role of is to assign weights to different nodes when calculating the influence coverage function to reflect the semantic importance of the nodes; the sampling probability vector n The role of is to decide which nodes to start from to build the reverse reachable set.

4. The method for maximizing heterogeneous network influence based on meta-path semantic perception according to claim 3 is characterized in that: Path weight vector m The generation of is based on a predefined meta-path weight strategy, sampling probability vector n The generation of is based on the semantic information of heterogeneous networks, which includes node types and path types; Sampling probability vector n The generation basis is based on the importance of node types, giving priority to target node types, assigning probability values ​​to each node in the node set, and allocating higher sampling probabilities to target nodes based on the node type distribution.

5. The method for maximizing heterogeneous network influence based on meta-path semantic perception according to claim 1 is characterized in that: The influence weighted cover function is used to calculate the seed set S exist The average weighted cover in the valid reverse reachable set is defined as: ； Where, w i Indicates the i Valid reverse reachable set R i The weight of x i Indicates the i Valid reverse reachable set R i With seed set S The intersection of hour ,otherwise ; For a given reverse sampling strategy , any seed set and the number of reverse reachable sets , its influence weighted coverage function satisfies: ； set up For seed collection S The expected impact under influence spread, if is an effective reverse sampling strategy, then there exists a constant ,satisfy: ； Where, Indicates the preset coefficient; represents the single effective reverse reachable set cover estimator, that is, the weighted cover indicator value based on the random effective reverse reachable set; represents the expected value of a single valid reverse reachable set cover estimator.

6. A heterogeneous network influence maximization system based on meta-path semantic perception, characterized by: The method for maximizing the influence of a heterogeneous network based on meta-path semantic perception according to any one of claims 1 to 5 is applied, comprising: A graph construction unit, used for constructing a heterogeneous network graph based on heterogeneous network data; The path generation unit is used to obtain the meta-paths of the heterogeneous network graph, simplify and integrate them, and generate a valid path set; The strategy construction unit is used to set node path weights, determine node sampling probabilities, and construct reverse sampling strategies based on the heterogeneous network graph; The sampling unit is used to sample the heterogeneous network graph using a reverse sampling strategy and generate a valid reverse reachable set based on the determination of valid paths; The screening unit is used to calculate the node marginal influence propagation gain based on the effective reverse reachable set, and iteratively screen with the marginal gain as the indicator to obtain the target seed set and maximize the influence.

Images