Topic-related influence maximization set search method considering self-activated nodes

Abstract

The present invention discloses a method for searching a set of topics related to self-activated nodes for maximum influence. The method includes: obtaining a network diagram, the network diagram including multiple nodes and the connection relationship between the multiple nodes, each node being associated with a corresponding interest topic distribution, the interest topic distribution including the degree of interest of the user corresponding to the node in each preset topic; obtaining a topic distribution of a target query, the topic distribution of the target query including the degree of association between the entity to be propagated and each preset topic; determining a self-activated node set based on the topic distribution of the target query and the network diagram; determining a seed node set based on the joint influence expansion of the nodes in the remaining set and the nodes in the self-activated node set, the remaining set being the node set in the network diagram excluding the self-activated node set, and taking the union of the seed node set and the self-activated node set as the search result. The present invention can improve the effect of information dissemination.

Description

Technical Field

[0001] The present invention relates to the field of information dissemination technology, and in particular to a method for searching a topic-related influence-maximizing set considering self-activated nodes. Background Art

[0002] Social networks are rapidly developing, with a vast number of users now using them to communicate and publish content. A seemingly insignificant piece of information on a social network can potentially have a massive impact through its dissemination. As a key issue in information dissemination, the influence maximization (IM) problem has attracted extensive research due to its potential commercial value. The IM problem aims to identify k users on a social network such that, after information propagates through the network, it influences as many users as possible. In real social networks, each user may have different interests, and the propagating entities may have different characteristics. Therefore, the probability of a user being influenced by different propagating entities may vary. Existing research on the influence maximization problem has mostly focused on the spread of influence from seed sets. In real life, when a company needs to promote its products, in addition to leveraging influential individuals on social networks, it often also places advertisements in other media, such as television, newspapers, and websites. If a person is sufficiently interested in this type of product, he or she may take the initiative to learn about it and be attracted by it, and then promote the product on social networks. At this time, he or she is similar to a seed node and plays the role of an influencer. However, the existing technology has not yet considered the method of maximizing the influence of such an influencer, resulting in insufficient actual influence expansion of the seed node set finally found and poor information dissemination effect.

[0003] Therefore, the existing technology still needs to be improved and enhanced. Summary of the Invention

[0004] In view of the above-mentioned defects of the prior art, the present invention provides a method for searching a topic-related influence maximization set considering self-activated nodes, aiming to solve the problem of poor actual information dissemination effect of the search results of the influence maximization problem in the prior art.

[0005] In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0006] A first aspect of the present invention provides a method for maximizing the influence of a set by considering the topic-related self-activated nodes, the method comprising:

[0007] Obtaining a network graph, the network graph including a plurality of nodes and connection relationships between the plurality of nodes, each node being associated with a corresponding interest topic distribution, the interest topic distribution including the interest level of the user corresponding to the node in each preset topic;

[0008] Obtaining a topic distribution of a target query, wherein the topic distribution of the target query includes a correlation between the entity to be propagated and each of the preset topics;

[0009] Determining a self-activated node set according to the subject distribution of the target query and the network graph;

[0010] A seed node set is determined based on the joint influence spread of the nodes in the remaining set and the nodes in the self-activated node set, where the remaining set is the node set in the network graph excluding the self-activated node set, and the union of the seed node set and the self-activated node set is used as the search result.

[0011] The method for maximizing the influence set related to themes of self-activated nodes, wherein determining the self-activated node set based on the theme distribution of the target query and the network graph, includes:

[0012] If the interest levels of all target topics in the interest topic distribution of the target node in the network graph are greater than a first preset threshold, the target node is added to the self-activated node set;

[0013] The target topic is a topic in the topic distribution of the target query whose relevance is greater than a second preset threshold.

[0014] The method for searching a set of topics that maximize influence related to self-activated nodes, wherein determining a seed node set based on the joint influence spread of nodes in the remaining set and nodes in the self-activated node set, includes:

[0015] Determining a joint influence spread of each node in the remaining set and the nodes in the self-activated node set, wherein the joint influence spread of the nodes in the remaining set and the nodes in the self-activated node set is the average number of reachable nodes of the nodes in the joint set in each snapshot graph, and the joint set is the union of the node in the remaining set and the self-activated node set;

[0016] Sorting the nodes in the remaining set based on the joint influence spread corresponding to each node in the remaining set;

[0017] The first node in the sorting result is removed from the remaining set and added to the seed node set;

[0018] Sorting the nodes in the remaining set based on the joint influence spread corresponding to each node in the remaining set;

[0019] When the first node after reordering is different from the first node before reordering, reorder again until the first node after reordering is the same as the first node before reordering;

[0020] When the first node after reordering is the same as the first node before reordering, the first node after reordering is removed from the remaining set and added to the seed node set;

[0021] The step of sorting the nodes in the remaining set based on the joint influence expansion degree corresponding to each node in the remaining set is re-executed until the number of nodes in the seed node set reaches a preset number.

[0022] The method for maximizing the influence set search related to themes that consider self-activated nodes, wherein the snapshot graph is obtained in the following manner:

[0023] Generate sample queries;

[0024] Calculating the activation probability of each connection edge in the network graph based on the sample query;

[0025] Deleting the connection edges in the network graph according to the activation probability to obtain the snapshot graph;

[0026] Among them, the given topic distribution is The activation probability of the edge connecting node u to node v is The calculation formula is: in γ i represents the relevance to topic i, Z represents the total number of topics, Indicates the interest level of the user corresponding to node v in topic i.

[0027] The method for searching a set of topics that maximize influence related to self-activated nodes, wherein determining a seed node set based on the joint influence spread of nodes in the remaining set and nodes in the self-activated node set, includes:

[0028] Arrange all nodes in the network graph in reverse order according to the out-degree of the nodes to obtain an arrangement result;

[0029] Initialize the joint influence expansion of all nodes in the arrangement result to 1;

[0030] Iteratively updating the joint influence expansion of the nodes in the arrangement result, and after each iteration, re-sorting all the nodes in the network graph based on the joint influence expansion until the first K nodes in the n-th sorting and the n-1-th sorting are the same;

[0031] Adding the first preset number of nodes other than the nodes in the self-activated node set in the nth sorting to the seed node set;

[0032] In each iteration, the following operations are performed:

[0033] Starting from the last node of the arrangement result, each node is selected as the first node in turn, and the joint influence expansion degree of each node is updated based on the probability of the connection edge between the selected first node and other nodes being activated.

[0034] The method for maximizing the influence set search related to topics that consider self-activated nodes, wherein the updating of the joint influence expansion of each node based on the activation probability of the connecting edges between the selected node and other nodes includes:

[0035] Starting from the first node of the arrangement result, each node is selected as the second node in turn, and the joint influence expansion degree of the selected first node and the selected second node is updated according to the following steps:

[0036] Obtaining an activation probability of a connection edge from the selected second node to the selected first node;

[0037] Updating the expected influence value of the selected second node to the sum of the product of the activation probability and the selected first node and the joint influence spread of the selected second node;

[0038] The expected influence value of the selected first node is updated to the product of the difference between 1 and the activation probability and the joint influence expansion degree of the selected first node.

[0039] The method for maximizing the influence set search related to themes that consider self-activated nodes, wherein the number of iterations for iteratively updating the joint influence expansion degree of the nodes in the arrangement result is not less than 10 times.

[0040] A second aspect of the present invention provides a device for maximizing the influence of a set based on the topic of a self-activated node, the device comprising:

[0041] A first information acquisition module is configured to acquire a network diagram, wherein the network diagram includes a plurality of nodes and connection relationships between the plurality of nodes, and each node is associated with a distribution of interest topics corresponding to the node, wherein the distribution of interest topics includes the degree of interest of the user corresponding to the node in each preset topic;

[0042] A second information acquisition module is configured to acquire a topic distribution of a target query, wherein the topic distribution of the target query includes a correlation between the entity to be propagated and each of the preset topics;

[0043] A self-activation node search module, configured to determine a self-activation node set based on the subject distribution of the target query and the network graph;

[0044] A seed node search module is used to determine a seed node set based on the joint influence expansion of the nodes in the remaining set and the nodes in the self-activated node set, wherein the remaining set is the node set in the network graph excluding the self-activated node set, and the union of the seed node set and the self-activated node set is used as the search result.

[0045] The third aspect of the present invention provides a terminal, which includes a processor and a computer-readable storage medium communicatively connected to the processor, the computer-readable storage medium is suitable for storing multiple instructions, and the processor is suitable for calling the instructions in the computer-readable storage medium to execute the steps of the method for maximizing the influence set related to the topic considering self-activated nodes as described above.

[0046] A fourth aspect of the present invention provides a computer-readable storage medium storing one or more programs, which can be executed by one or more processors to implement the steps of the above-mentioned method for maximizing the influence set related to the topic of self-activated nodes.

[0047] Compared with the prior art, the present invention provides a method for searching a topic-maximized influence set considering self-activated nodes. In the method for searching a topic-maximized influence set considering self-activated nodes provided by the present invention, the self-activated nodes are first determined based on the interest level of each node in the network graph in the topic and the topic distribution of the query, and then the seed nodes are determined based on the self-activated nodes. Finally, the union of the two is taken as the search result of the influence-maximized set, which fully considers the nodes in the social network that may spontaneously promote entities, thereby improving the information dissemination effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flowchart of an embodiment of the present invention's method for maximizing the influence of a set of search results based on topic-related self-activated nodes;

[0049] Figure 2 A schematic diagram of an algorithm for determining a set of self-activated nodes in an embodiment of a method for searching a set of maximizing the influence of topics related to self-activated nodes provided by the present invention;

[0050] Figure 3 A schematic diagram of an algorithm for using the first joint influence expansion evaluation method in an embodiment of the influence maximization set search method considering the topic-related self-activated nodes provided by the present invention;

[0051] Figure 4 A schematic diagram of an algorithm for using the second joint influence expansion evaluation method in an embodiment of the influence maximization set search method considering the topic-related self-activated nodes provided by the present invention;

[0052] Figure 5 Experimental verification of the subject-related influence maximization set search method considering self-activated nodes provided by the present invention Figure 1 ;

[0053] Figure 6 Experimental verification of the subject-related influence maximization set search method considering self-activated nodes provided by the present invention Figure 2 ;

[0054] Figure 7 Experimental verification of the subject-related influence maximization set search method considering self-activated nodes provided by the present invention Figure 3 ;

[0055] Figure 8 Experimental verification of the subject-related influence maximization set search method considering self-activated nodes provided by the present invention Figure 4 ;

[0056] Figure 9 Experimental verification of the subject-related influence maximization set search method considering self-activated nodes provided by the present invention Figure 5 ;

[0057] Figure 10 A schematic diagram of the structure of an embodiment of a device for maximizing the influence of a set based on themes related to self-activated nodes provided by the present invention;

[0058] Figure 11 This is a schematic diagram of the principles of an embodiment of a terminal provided by the present invention. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0060] The subject-related influence maximization set search method considering self-activation nodes provided by the present invention can be applied to terminals with computing capabilities. The terminal can execute the subject-related influence maximization set search method considering self-activation nodes provided by the present invention, and output the search results for the target query. The search results are node sets. The terminals can be but are not limited to various computers, mobile terminals, smart home appliances, wearable devices, etc.

[0061] Example 1

[0062] like Figure 1 As shown, in one embodiment of the method for maximizing the influence set search considering the topic related influence of the self-activated node, the method includes the steps of:

[0063] S100. Obtain a network diagram, wherein the network diagram includes multiple nodes and connection relationships between the multiple nodes, and each node is associated with a corresponding interest topic distribution, wherein the interest topic distribution includes the interest level of the user corresponding to the node in each preset topic.

[0064] The network graph is a directed graph, which is obtained based on a social network. Specifically, a topic-related social network is modeled as a directed graph G = (V, E), where V and E represent the set of nodes and the set of directed edges in the network, respectively. Each edge e = (u, v) ∈ E means that node u can directly propagate information to v, for example, when node v follows node u in the social network. Each node v ∈ V is associated with a topic distribution representing the user's interest. in represents the degree of interest of node v in topic i, and Z represents the total number of topics in network G. These topics can describe the interests of users and the characteristics of communication entities. The distribution of each user's interest topics can be extracted from the text content posted by the user on the social network, or from the content that the user likes and collects on the social network.

[0065] S200: Obtain a topic distribution of a target query, where the topic distribution of the target query includes a correlation between an entity to be propagated and each of the preset topics.

[0066] The query is an example of the influence maximization problem. The query gives the topic distribution describing the entity to be propagated and a constant k, which represents the number of seed nodes to be found, so as to maximize the expected number of active nodes after the propagation based on the found seed nodes. In this invention, the query is modeled as in represents the topic distribution of the query, γ i Indicates the extent to which topic i can represent the dissemination entity, that is, the degree of association between the entity to be disseminated and topic i, that is, k represents the number of seed nodes required by the query.

[0067] In the present invention, the IC model is used to describe the information propagation process. Each node in the social network will be in one of the two states, active or inactive, in any discrete time period during the information propagation process. In the initial state, only the seed node is active, and the other nodes are inactive. At this time, each seed node gets an opportunity to activate its own outgoing neighbors with a certain probability. The activation probability here is related to the topic distribution of the propagation entity and the node. In the present invention, whether a node can be activated by a propagation entity depends largely on the degree of interest of the node in the various topics related to the propagation entity, because the higher the user's interest in the various topics associated with the propagation entity, the more likely it is to accept the propagation entity. Given a topic distribution of The probability that node u activates its outgoing neighbor v is Defined as: The probability that a node u activates its outgoing neighbor v is also called the activation probability of the edge connecting node u to v.

[0068] If an outgoing neighbor v is successfully activated, it becomes active and has one chance to try to activate its outgoing neighbors in the same way in the next time period. Nodes that have never had an active incoming neighbor during the entire propagation process remain inactive. If no new active nodes are generated within a discrete time period, the propagation process ends.

[0069] Please refer again Figure 1 , the method provided in this embodiment further includes the steps of:

[0070] S300: Determine a self-activated node set according to the topic distribution of the target query and the network graph.

[0071] Given a query The present invention introduces the concept of influence spread of the seed set S considering the self-activated nodes (hereinafter referred to as influence spread), which is denoted as To describe the expected number of nodes directly or indirectly activated by seed nodes or self-activated nodes during the influence propagation process:

[0072] Among them S A represents the set of self-activated nodes, S represents the set of seed nodes, and I Q (S∪S A ) represents the node set activated by the nodes in the union of the seed node set and the self-activated node set, and E represents the expected value. The problem to be solved by the present invention is: given a topic-related social network G = (V, E) and a query Aims to find a seed set S of size k * , making Maximize, that is:

[0073] Given a query, a graph G = (V, E) with a defined probability for each edge can be generated, i.e., an instance of the traditional influence maximization problem can be determined. This means that the problem to be solved by the present invention maintains the NP-hard and submodular properties of the IM problem.

[0074] In the method provided by the present invention, self-activated nodes are first determined. Specifically, if a node has sufficient interest in the entity to be propagated, it is called a self-activated node. The interest threshold is λ. Given a query, if a node's interest in topics related to the query reaches λ, the node becomes a self-activated node. That is, determining the set of self-activated nodes based on the topic distribution of the target query and the network graph includes:

[0075] If the interest levels of all target topics in the interest topic distribution of the target node in the network graph are greater than a first preset threshold, the target node is added to the self-activated node set;

[0076] The target topic is a topic in the topic distribution of the target query whose relevance is greater than a second preset threshold.

[0077] The first preset threshold is the aforementioned λ, and the second preset threshold can be set to 0.

[0078] Figure 2 Algorithm 3-1 shown in describes how to obtain the self-activated node set based on this condition.

[0079] S400. Determine a seed node set based on the joint influence spread of the nodes in the remaining set and the nodes in the self-activated node set, where the remaining set is the node set in the network graph excluding the self-activated node set, and take the union of the seed node set and the self-activated node set as the search result.

[0080] In order to efficiently output node search results based on arbitrary queries, the present invention proposes two node joint influence expansion evaluation methods under the premise of considering self-activated nodes.

[0081] In one implementation, Figure 3 As shown in Algorithm 3-2, the step of determining the seed node set based on the combined influence spread of the nodes in the remaining set and the nodes in the self-activated node set includes:

[0082] Determining a joint influence spread of each node in the remaining set and the nodes in the self-activated node set, wherein the joint influence spread of the nodes in the remaining set and the nodes in the self-activated node set is the average number of reachable nodes of the nodes in the joint set in each snapshot graph, and the joint set is the union of the node in the remaining set and the self-activated node set;

[0083] Sorting the nodes in the remaining set based on the joint influence spread corresponding to each node in the remaining set;

[0084] The first node in the sorting result is removed from the remaining set and added to the seed node set;

[0085] Sorting the nodes in the remaining set based on the joint influence spread corresponding to each node in the remaining set;

[0086] When the first node after reordering is different from the first node before reordering, reorder again until the first node after reordering is the same as the first node before reordering;

[0087] When the first node after reordering is the same as the first node before reordering, the first node after reordering is removed from the remaining set and added to the seed node set;

[0088] The step of sorting the nodes in the remaining set based on the joint influence expansion degree corresponding to each node in the remaining set is re-executed until the number of nodes in the seed node set reaches a preset number.

[0089] The snapshot image is obtained as follows:

[0090] Generate sample queries;

[0091] Calculating the activation probability of each connection edge in the network graph based on the sample query;

[0092] Deleting the connection edges in the network graph according to the activation probability to obtain the snapshot graph;

[0093] Among them, the given topic distribution is The activation probability of the edge connecting node u to node v is The calculation formula is: in γ i represents the relevance to topic i, Z represents the total number of topics, Indicates the interest level of the user corresponding to node v in topic i.

[0094] Specifically, R simulations are performed on the social network in advance to obtain R snapshots. In each simulation, for each edge (u, v) ∈ E, if u successfully activates v, (u, v) is retained, otherwise it is deleted. The probability of each edge being retained or deleted is determined based on the activation probability of the edge. Through these simulations, R snapshots of the propagation process can be obtained. When evaluating the influence extension of any node set, these snapshots can be directly used without having to re-run the simulation for each evaluation. In the process of generating search results based on queries, first, according to the query, the self-activated node set S is obtained. A , then from V\S A When searching for the first seed node, it is necessary to evaluate V\S A Each node u and S A The joint influence spread u.spread of the nodes is then calculated, and these nodes are sorted in reverse order according to the spread value and recorded in L. The first node x in L is then the first seed node, which is added to S and removed from L. The remaining k-1 seed nodes are then searched for. When searching for these nodes, the first step is to update the marginal influence of the first node y in L relative to S, and then re-sort L. If y is still the first node, then it is the next seed node; otherwise, the marginal influence of the first node in L relative to S is continuously updated and re-sorted until a seed node that meets the criteria is found.

[0095] In another implementation, Figure 4 As shown in Algorithm 3-3, the step of determining the seed node set based on the combined influence spread of the nodes in the remaining set and the nodes in the self-activated node set includes:

[0096] Arrange all nodes in the network graph in reverse order according to the out-degree of the nodes to obtain an arrangement result;

[0097] Initialize the joint influence expansion of all nodes in the arrangement result to 1;

[0098] Iteratively updating the joint influence expansion of the nodes in the arrangement result, and after each iteration, re-sorting all the nodes in the network graph based on the joint influence expansion until the first K nodes in the n-th sorting and the n-1-th sorting are the same;

[0099] Adding the first preset number of nodes other than the nodes in the self-activated node set in the nth sorting to the seed node set;

[0100] In each iteration, the following operations are performed:

[0101] Starting from the last node of the arrangement result, each node is selected as the first node in turn, and the joint influence expansion degree of each node is updated based on the probability of the connection edge between the selected first node and other nodes being activated.

[0102] The updating of the joint influence expansion of each node based on the activation probability of the connection edge between the selected node and other nodes includes:

[0103] Starting from the first node of the arrangement result, each node is selected as the second node in turn, and the joint influence expansion degree of the selected first node and the selected second node is updated according to the following steps:

[0104] Obtaining an activation probability of a connection edge from the selected second node to the selected first node;

[0105] Updating the expected influence value of the selected second node to the sum of the product of the activation probability and the selected first node and the joint influence spread of the selected second node;

[0106] The expected influence value of the selected first node is updated to the product of the difference between 1 and the activation probability and the joint influence expansion degree of the selected first node.

[0107] When there is no connection edge from the selected second node to the selected first node, the activation probability of the connection edge from the selected second node to the selected first node is defined as 0.

[0108] In Algorithm 3-3, r is the initial arrangement of all nodes - arranged in reverse order of the out-degree of each node. is the i-th node in the array r, M r is the influence expansion based on the arrangement r, where M r (u) is equivalent to the expected number of nodes activated by node u. Repeat the calculation of M for each node r value and according to the new M r The value reorders all nodes until two consecutive permutations r (t-1) and r (t) The first k nodes of (t) The first K in S do not exist A The nodes in are output as the seed set for a given query. In practical applications, K can directly be the set number k of nodes in the seed set. In this implementation, the self-activated node S is used to find the seed node. A Exclude to ensure that the seed node and S A The joint influence of the nodes in the arrangement result is as large as possible. In the present invention, the number of iterations for iteratively updating the joint influence expansion degree of the nodes in the arrangement result is not less than 10 times, that is, iterative calculation Mr The number of iterations is at least 10 to avoid the reordering process ending prematurely and thus improve the effectiveness of the algorithm to a certain extent.

[0109] In order to verify the effectiveness of the method provided by the present invention, experiments were conducted.

[0110] The experimental setup is as follows:

[0111] 1. Dataset

[0112] This paper uses three data sets for experimental research. The three data sets are literature citation networks, namely:

[0113] Arnetminer: Extracted from existing research literature, including 10 topics.

[0114] arXiv-HepPh: obtained from the snap website. Since this network has no data on related topics, the present invention simulates 30 topics and randomly assigns 1-5 topics to each node.

[0115] DBLP: Derive from existing research literature and contains 100 topics.

[0116] More detailed statistics of these three datasets are shown in Table 1.

[0117] Table 1 Some statistics of the dataset used

[0118]

[0119]

[0120] Figure 5 The distribution of the number of topics associated with nodes in each dataset is shown. It can be seen that in Arnetminer, most nodes are associated with only one topic, while in DBLP, the vast majority (over 93%) of nodes are associated with no more than 10 topics. In arXiv-HepPh, topics are assigned to each node in a uniformly random manner, resulting in a nearly identical number of nodes per topic.

[0121] 2. Query

[0122] To evaluate the algorithm's performance on any given query, we varied the number of topics q included in the query from 1 to 3, since in real life, a product is typically associated with only a few topics. To account for the varying sizes of the datasets, we varied the number of seed nodes k from 10 to 50 in the Arnetminer dataset and from 10 to 100 in the arXiv-HepPh and DBLP datasets. For each query q, we randomly generated 50 queries, and used their average influence spread and average run time to evaluate the algorithm's performance on the query.

[0123] 3. Comparison of algorithms

[0124] Several widely used classic influence maximization algorithms are used as comparison algorithms in the experiment:

[0125] Degree: Select the k nodes with the largest degrees as seed nodes.

[0126] Static Greedy: Generate R snapshots in advance and use these snapshots to evaluate the influence spread of a given set of nodes.

[0127] RIS: Uses reverse reachable sets to evaluate the influence spread of a given set of nodes.

[0128] 4. Parameter settings

[0129] In Algorithm 3-2, R is set to 100, meaning 100 snapshots are generated. λ is set to 0.8 to ensure that an individual is self-activated only if they have sufficient interest in all topics associated with a product. The number of simulations performed to estimate the influence spread of any given set of nodes is set to 1000.

[0130] The present invention uses Python to implement each algorithm and runs on a computer with Ubuntu 16.04 system, 62G memory and a CPU model of 2.2GHz Intel(R) Xeon(R) E5-2650 v4.

[0131] Experimental results analysis

[0132] (1) Comparison of effectiveness

[0133] Figure 6The results show how the influence spread of each algorithm in each dataset changes with the increase of seed set size k when the number of topics with non-zero weights in the query is q = 2. In the Arnetminer and arXiv-HepPh datasets, all five algorithms were run in the experiment, while in the DBLP dataset, because it is too large for the SG and RIS algorithms, only the other three algorithms were run. It can be found that the SG-CELF-SA and IMRank-SA algorithms that consider self-activated nodes significantly achieve higher influence spread. The specific improvement of the influence spread of the algorithm proposed in this invention in each dataset is shown in Table 2. In the Arnetminer and arXiv-HepPh datasets, the method provided by this invention achieves a significant improvement in influence spread, while in the DBLP dataset, the improvement is not as obvious. One possible reason is that in the DBLP dataset, the higher the degree of the node, the greater the influence on the entire network, because the more cited papers are, the more well-known they are in the entire citation network. Therefore, even with simple algorithms, it is relatively easy to find nodes with greater influence, especially the Degree algorithm.

[0134] Table 2 Improvement of the proposed algorithm on the influence expansion of the comparison algorithm

[0135]

[0136] (2) Efficiency comparison

[0137] Figure 7 The figure shows how the running time of each algorithm in each dataset changes with the increase of the seed set size k when the number of topics with non-zero weights in the query is q = 2. Except for the Degree algorithm, which takes much shorter time, the ratio of the running time of the algorithm proposed in the present invention to that of the SG and RIS algorithms is shown in Table 3. It can be clearly seen that the algorithm proposed in the present invention takes much less time than the SG and RIS algorithms. In addition, when k increases, the running time required for the method proposed in the present invention remains basically unchanged, which is better than the SG algorithm and similar to the RIS algorithm. In the DBLP dataset, due to its large scale, the RIS and SG algorithms are no longer feasible, while Algorithm 3-3 of the present invention can still be completed in a few seconds, and Algorithm 3-2 can also be completed in a few minutes.

[0138] Table 3 Ratio between the proposed algorithm and the comparison algorithm

[0139]

[0140] (3) Influence of self-activation nodes

[0141] In order to study the contribution of self-activated nodes to the overall influence expansion, the experiment compared the proportion of the influence expansion of self-activated nodes and seed nodes to the overall influence expansion. The results are as follows: Figure 8 As shown. Given the self-activation threshold λ and the query condition, since the self-activation node set S A is fixed. As the size of the seed set k increases, S A Specifically, as k increases from 10 to 100, in Algorithm 3-2, S A The influence expansion share of S decreased from 38% to 12%; in Algorithm 3-1, this proportion decreased from 36% to 11%. A The role of should not be ignored, especially when the k required by the query is relatively small.

[0142] (4) Influence of interest threshold λ

[0143] In order to study the effect of the self-activation threshold, based on the arXiv-HepPh dataset, we increase λ from 0.6 to 1 and observe the changes in the influence extension and running time of the two algorithms proposed in this paper. The results are as follows: Figure 9 As shown in Figure 2, the influence scalability of both algorithms decreases as λ increases, as the self-activation conditions become increasingly stringent, resulting in fewer and fewer self-activated nodes. In particular, when λ increases to 1, the problem degenerates into the traditional TIM problem, where there are no self-activated nodes. On the other hand, the runtime of both algorithms remains nearly constant as λ increases.

[0144] In summary, this embodiment provides a method for searching a topic-related influence-maximizing set taking into account self-activated nodes. First, the self-activated nodes are determined based on the interest level of each node in the network graph in the topic and the topic distribution of the query. Then, the seed nodes are determined based on the self-activated nodes. Finally, the union of the two is taken as the search result of the influence-maximizing set. This method fully considers the nodes in the social network that may spontaneously promote entities, thereby improving the information dissemination effect.

[0145] It should be understood that, although the steps in the flowcharts provided in the accompanying drawings of the present invention are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps may be performed in other orders. Moreover, at least a portion of the steps in the flowcharts may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily performed at the same time, but may be performed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but may be performed in rotation or alternation with other steps or at least a portion of sub-steps or stages of other steps.

[0146] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0147] Example 2

[0148] Based on the above embodiments, the present invention also provides a device for searching a set of topics that considers the influence maximization of self-activated nodes. Figure 10 As shown, the device for maximizing the influence set related to themes of self-activated nodes includes:

[0149] A first information acquisition module is configured to acquire a network diagram, wherein the network diagram includes a plurality of nodes and connection relationships between the plurality of nodes, and each node is associated with a corresponding interest topic distribution, wherein the interest topic distribution includes the interest level of the user corresponding to the node in each preset topic, as specifically described in the first embodiment;

[0150] A second information acquisition module is configured to acquire a topic distribution of a target query, wherein the topic distribution of the target query includes a correlation between the entity to be propagated and each of the preset topics, as specifically described in the first embodiment;

[0151] A self-activation node search module, configured to determine a self-activation node set based on the subject distribution of the target query and the network graph, as specifically described in the first embodiment;

[0152] A seed node search module is used to determine a seed node set based on the joint influence expansion of the nodes in the remaining set and the nodes in the self-activated node set, wherein the remaining set is the node set in the network graph excluding the self-activated node set, and the union of the seed node set and the self-activated node set is used as the search result, as specifically described in Example 1.

[0153] Example 3

[0154] Based on the above embodiments, the present invention also provides a terminal, such as Figure 11 As shown, the terminal includes a processor 10 and a memory 20. Figure 11 Only some of the components of the terminal are shown, but it should be understood that implementation of all of the shown components is not required, and more or fewer components may be implemented instead.

[0155] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the terminal. Furthermore, the memory 20 may also include both an internal storage unit of the terminal and an external storage device. The memory 20 is used to store application software and various types of data installed on the terminal. The memory 20 may also be used to temporarily store data that has been output or is about to be output. In one embodiment, the memory 20 stores a program 30 for maximizing the influence of a set related to a topic that considers self-activation nodes. The program 30 for maximizing the influence of a set related to a topic that considers self-activation nodes can be executed by the processor 10, thereby realizing the method for maximizing the influence of a set related to a topic that considers self-activation nodes in the present application.

[0156] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other chip, configured to execute program code or process data stored in the memory 20, such as executing the target-based multimodal named entity recognition method.

[0157] In one embodiment, when the processor 10 executes the influence maximization set search program 30 in the memory 20 that considers the topic-related influence of the self-activated node, the following steps are implemented:

[0158] Obtaining a network graph, the network graph including a plurality of nodes and connection relationships between the plurality of nodes, each node being associated with a corresponding interest topic distribution, the interest topic distribution including the interest level of the user corresponding to the node in each preset topic;

[0159] Obtaining a topic distribution of a target query, wherein the topic distribution of the target query includes a correlation between the entity to be propagated and each of the preset topics;

[0160] Determining a self-activated node set according to the subject distribution of the target query and the network graph;

[0161] A seed node set is determined based on the joint influence spread of the nodes in the remaining set and the nodes in the self-activated node set, where the remaining set is the node set in the network graph excluding the self-activated node set, and the union of the seed node set and the self-activated node set is used as the search result.

[0162] Example 4

[0163] The present invention also provides a computer-readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the influence maximization set search method considering the topic-related influence of self-activated nodes as described above.

[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for maximizing the influence of a topic-related set considering self-activated nodes, characterized in that: The method comprises: Obtaining a network graph, the network graph including a plurality of nodes and connection relationships between the plurality of nodes, each node being associated with a corresponding interest topic distribution, the interest topic distribution including the interest level of the user corresponding to the node in each preset topic; Obtaining a topic distribution of a target query, wherein the topic distribution of the target query includes a correlation between the entity to be propagated and each of the preset topics; Determining a self-activated node set according to the subject distribution of the target query and the network graph; Determining a seed node set based on the joint influence spread of nodes in a remaining set and nodes in the self-activated node set, wherein the remaining set is a node set in the network graph excluding the self-activated node set, and taking the union of the seed node set and the self-activated node set as a search result; The step of determining the seed node set based on the combined influence spread of the nodes in the remaining set and the nodes in the self-activated node set includes: Determining a joint influence spread of each node in the remaining set and the nodes in the self-activated node set, wherein the joint influence spread of the nodes in the remaining set and the nodes in the self-activated node set is the average number of reachable nodes of the nodes in the joint set in each snapshot graph, and the joint set is the union of the node in the remaining set and the self-activated node set; Sorting the nodes in the remaining set based on the joint influence spread corresponding to each node in the remaining set; The first node in the sorting result is removed from the remaining set and added to the seed node set; Sorting the nodes in the remaining set based on the joint influence spread corresponding to each node in the remaining set; When the first node after reordering is different from the first node before reordering, reorder again until the first node after reordering is the same as the first node before reordering; When the first node after reordering is the same as the first node before reordering, the first node after reordering is removed from the remaining set and added to the seed node set; Re-performing the step of sorting the nodes in the remaining set based on the joint influence expansion degree corresponding to each node in the remaining set until the number of nodes in the seed node set reaches a preset number; The snapshot image is obtained as follows: Generate sample queries; Calculating the activation probability of each connection edge in the network graph based on the sample query; Deleting the connection edges in the network graph according to the activation probability to obtain the snapshot graph; Among them, the given topic distribution is The activation probability of the edge connecting node u to node v is The calculation formula is: in γ i represents the relevance to topic i, Z represents the total number of topics, Indicates the interest level of the user corresponding to node v in topic i.

2. The method for maximizing the influence set search based on the topic of self-activated nodes according to claim 1, characterized in that: The determining of the self-activated node set according to the topic distribution of the target query and the network graph includes: If the interest levels of all target topics in the interest topic distribution of the target node in the network graph are greater than a first preset threshold, the target node is added to the self-activated node set; The target topic is a topic in the topic distribution of the target query whose relevance is greater than a second preset threshold.

3. The method for maximizing the influence of a set based on the topic of self-activated nodes according to claim 1, characterized in that: The step of determining the seed node set based on the combined influence spread of the nodes in the remaining set and the nodes in the self-activated node set further includes: Arrange all nodes in the network graph in reverse order according to the out-degree of the nodes to obtain an arrangement result; Initialize the joint influence expansion of all nodes in the arrangement result to 1; Iteratively updating the joint influence expansion of the nodes in the arrangement result, and after each iteration, re-sorting all the nodes in the network graph based on the joint influence expansion until the first K nodes in the n-th sorting and the n-1-th sorting are the same; Adding the first preset number of nodes other than the nodes in the self-activated node set in the nth sorting to the seed node set; In each iteration, the following operations are performed: Starting from the last node of the arrangement result, each node is selected as the first node in turn, and the joint influence expansion degree of each node is updated based on the probability of the connection edge between the selected first node and other nodes being activated.

4. The method for maximizing the influence set search based on the topic of self-activated nodes according to claim 3, characterized in that: The updating of the joint influence expansion degree of each node based on the activation probability of the connection edge between the selected first node and other nodes also includes: Starting from the first node of the arrangement result, each node is selected as the second node in turn, and the joint influence expansion degree of the selected first node and the selected second node is updated according to the following steps: Obtaining an activation probability of a connection edge from the selected second node to the selected first node; Updating the expected influence value of the selected second node to the sum of the product of the activation probability and the selected first node and the joint influence spread of the selected second node; The expected influence value of the selected first node is updated to the product of the difference between 1 and the activation probability and the joint influence expansion degree of the selected first node.

5. The method for maximizing the influence set search based on the topic of self-activated nodes according to claim 3, characterized in that: The number of iterations for iteratively updating the joint influence expansion degree of the nodes in the arrangement result is not less than 10 times.

6. A device for searching a set of topics related to self-activated nodes with a maximum influence, wherein the device is applied to the method for searching a set of topics related to self-activated nodes with a maximum influence according to any one of claims 1 to 5, and is characterized in that: The device comprises: A first information acquisition module is configured to acquire a network diagram, wherein the network diagram includes a plurality of nodes and connection relationships between the plurality of nodes, and each node is associated with a distribution of interest topics corresponding to the node, wherein the distribution of interest topics includes the degree of interest of the user corresponding to the node in each preset topic; A second information acquisition module is configured to acquire a topic distribution of a target query, wherein the topic distribution of the target query includes a correlation between the entity to be propagated and each of the preset topics; A self-activation node search module, configured to determine a self-activation node set based on the subject distribution of the target query and the network graph; A seed node search module is used to determine a seed node set based on the joint influence expansion of the nodes in the remaining set and the nodes in the self-activated node set, wherein the remaining set is the node set in the network graph excluding the self-activated node set, and the union of the seed node set and the self-activated node set is used as the search result.

7. A terminal, characterized in that: The terminal includes: a processor, a computer-readable storage medium communicatively connected to the processor, the computer-readable storage medium being suitable for storing a plurality of instructions, and the processor being suitable for calling the instructions in the computer-readable storage medium to execute the steps of the method for maximizing the influence set search of the subject-related self-activated nodes according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the method for maximizing the influence set related to the subject of the self-activated node as described in any one of claims 1 to 5.

Images