Social network influence maximization method based on node reading effect and delay

Abstract

The application discloses a social network influence maximization method based on node reading effect and delay, comprising the following steps: S1, obtaining historical transmission data of multiple articles, and constructing a historical propagation network with time sequence for each article; S2, constructing a propagation network portrait according to the historical propagation network of all articles, and calculating the reading delay of nodes on the historical propagation network; S3, predicting the reading duration of a node after receiving an article, the reading delay, and the probability of propagating the article to other nodes; S4, for a certain node, forming a predicted propagation tree with the node as a root node; and S5, screening a plurality of initial propagation nodes based on the propagation tree. The article realizes influence maximization propagation based on node reading effect and delay.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of social networks, in particular to a social network influence maximization method based on node reading effect and delay. BACKGROUND

[0002] With the development of Internet technology, social networks play an increasingly important role in society, and influence maximization is an important part of social network analysis, which is widely used in marketing strategies, targeted advertising, public opinion prediction and control.

[0003] There are several historical propagation data of articles with labels, each article can have one or more labels such as military, finance, etc. The historical data of propagation on social networks includes the nodes passed through, the time passed through, the node reading time, and the association between the nodes shared by other nodes, etc.

[0004] Based on these data, how to maximize the propagation effect of a new article on the social network within a specified time range, i.e. how to determine the initial propagation node set so that the article can achieve the maximum propagation effect, has become a problem to be solved. SUMMARY

[0005] To solve the above problems, the present application provides a social network influence maximization method based on node reading effect and delay, which takes the historical propagation data of multiple articles as input, generates a node performance table and an edge propagation table of the social network, then generates a propagation network portrait based on the two tables, predicts the reading time, reading delay and propagation probability between nodes of a new article based on the propagation network portrait, and forms a predicted propagation tree of the article on the social network, and finally selects several initial propagation nodes using heuristic methods to achieve the influence maximization of the article.

[0006] The present application provides a social network influence maximization method based on node reading effect and delay, and the specific technical solutions are as follows:

[0007] S1: Obtain the historical transmission data of multiple articles, and construct a historical propagation network with time sequence for each article;

[0008] S2: Construct a propagation network portrait based on the historical propagation network of all articles, and calculate the reading delay of the nodes on the historical propagation network;

[0009] S3: Predict the reading time, reading delay and probability of propagating the article to other nodes of a node after receiving the article;

[0010] S4: For a certain node, find other nodes that can propagate in a cascading calculation mode, until the predicted propagation probability is less than a set threshold γ or the total propagation delay exceeds a preset specified value, forming a predicted propagation tree with the node as the root node;

[0011] The sum of the product of the reading duration and the probability of all nodes on the propagation tree is the predicted influence of the node;

[0012] S5: Based on the propagation tree, a plurality of initial propagation nodes are screened out.

[0013] Further, in step S1, based on the node number, reading time, reading duration, propagation node, receiving node and propagation time, the historical propagation network is constructed;

[0014] Among them, the starting point of the historical propagation network is the article publisher, and the first layer node is the initial propagation node set.

[0015] Further, in step S2, the construction of the propagation network portrait is as follows:

[0016] S201: For any node, use the node behavior data table to represent the historical behavior data on the node, and convert it into a node performance table;

[0017] S202: For any two nodes forming an edge, use the edge propagation data table to represent the historical propagation data on the edge, and convert it into an edge propagation force table;

[0018] S203: According to the nodes and edges in the historical propagation network of all articles, the propagation network portrait is constructed, and the reading delay of the nodes on the historical propagation network is calculated according to the propagation network portrait.

[0019] Further, the node behavior data table includes reading time, article label and reading duration three fields;

[0020] The reading time is a pre-divided corresponding time period;

[0021] The reading duration is the reading time length of the corresponding node at a certain reading time for a certain article label;

[0022] The node performance table includes reading time number, article label label and reading force, which correspond to the three fields in the node behavior data table respectively.

[0023] Further, the edge propagation data table includes propagation time, article label and propagation times three fields;

[0024] The edge propagation force table includes a propagation time point number, an article label number and a propagation force, which respectively correspond to three fields in the edge propagation data table.

[0025] Further, the reading delay is calculated as a numerical difference between a reading time point of the current node for a certain label article and a receiving time point of the label article.

[0026] When the reading delay is calculated, the node reads by default within a preset time after receiving the article, and if articles of the same label type are received from multiple previous nodes within the preset time, the data received at the latest time point is used as the reference.

[0027] Further, the reading duration of a node after receiving an article is predicted, and the prediction is calculated as follows:

[0028] If the node has reading records, the reading duration is predicted and calculated as follows:

[0029] p = (w1 * p1 + w2 * p2 + w3 * p3) / 3 * λ

[0030] w1, w2, and w3 represent weight coefficients, and w1 < w2 < w3, w1 + w2 + w3 = 1;

[0031] p1 represents the average reading duration calculated based on all reading records of the node;

[0032] p2 represents the average reading duration calculated based on reading records of the same label of the node;

[0033] p3 represents the average reading duration calculated based on reading records of articles of the same label and the same time point arriving at the node;

[0034] If the node has no reading records, the reading duration is predicted and calculated as follows:

[0035] p = p' * λ

[0036] p' represents the average value of the historical reading duration of all nodes in the historical propagation network;

[0037] wherein λ is a disturbance coefficient, which is a random number between 0.8 and 1.2.

[0038] Further, the reading delay of a node after receiving an article is predicted, and the prediction is calculated as follows:

[0039] If the node has reading records, the reading delay is predicted and calculated as follows:

[0040] d = (w1 * d1 + w2 * d2 + w3 * d3) / 3 * λ

[0041] w1, w2, and w3 represent weights, and w1 <w2<w3,w1+w2+w3=1;

[0042] d1 represents the average reading latency calculated based on all reading records on this node;

[0043] d2 represents the average reading latency calculated based on reading records of the same type of label on this node;

[0044] d3 represents the average reading latency calculated based on the reading records of articles with the same tag that arrived at the same time on this node;

[0045] If this node has no reading records, the predicted reading delay is calculated as follows:

[0046] d=d'*λ

[0047] d' is the average historical reading latency of all nodes on the historical propagation network;

[0048] λ represents a perturbation coefficient, which is a random number between 0.8 and 1.2.

[0049] Furthermore, the probability of a node propagating an article to other nodes is predicted as follows:

[0050] If there is no historical propagation data between node i and node j at any time, the probability that node i propagates the article to node j is predicted to be ρ, where ρ is a set value, which is a decimal less than 1 and close to 0.

[0051] If there is historical propagation data between node i and node j, the probability that node i will propagate data with label g to node j at time t is calculated as follows:

[0052] Calculate the total propagation force between node i and its neighboring nodes under the same label and at the same time, and divide it by the propagation force between node i and node j under the same label g and at the same time t. Multiply the resulting ratio by β, where β is a random number between 0.5 and 1.

[0053] Furthermore, in step S5, the initial propagation node selection process is as follows:

[0054] Let Q be the set of unselected nodes and S be the set of selected seed nodes.

[0055] S501: Select the node q with the greatest expected influence from Q;

[0056] S502: Add q to set S;

[0057] S503: Remove q and the propagation tree rooted at q from Q;

[0058] S504: If the number of nodes in S is less than k, and there are still elements in Q, then recalculate the expected influence and propagation tree of all nodes in Q, and return to S501.

[0059] The beneficial effects of this invention are as follows:

[0060] The historical propagation data of each article on social networks constitutes a historical propagation network. Using the historical propagation data of multiple articles as input, a node performance table and an edge propagation power table of the social network are generated, thereby generating a propagation network profile. By calculating and predicting the reading time, reading delay, and propagation probability between nodes of a new article, the predicted propagation tree corresponding to the article on the social network is obtained. Then, a heuristic method is used to select several initial propagation nodes to maximize the spread of the article based on the reading effect and delay of the nodes. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0062] Figure 2 This is a schematic diagram of the propagation network profile structure of the present invention. Detailed Implementation

[0063] The technical solutions in the embodiments of the present invention are clearly and completely described in the following description. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0064] In the description of the embodiments of the present invention, it should be noted that the indicated orientation or positional relationship is based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of the invention is conventionally placed during use, or the orientation or positional relationship in which those skilled in the art conventionally understand it during use. This is only for the convenience of describing the present invention and simplifying the description, and is not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of the present invention. Furthermore, the terms "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0065] In the description of the embodiments of the present invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set" and "connection" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0066] Example 1

[0067] Embodiment 1 of the present invention discloses a method for maximizing social network influence based on node reading effect and latency, such as... Figure 1 As shown, the specific steps are as follows:

[0068] S1: Obtain historical transmission data for multiple articles and construct a time-series historical propagation network G = (N, E) for each article;

[0069] Where N is the set of nodes, E is the set of directed edges, and each node in N is n = (s, t, d), where s is the node number, t is the reading time, and d is the reading duration;

[0070] In E, each edge e = (s, r, t), where s is the propagating node, r is the receiving node, and t is the propagation time;

[0071] In this embodiment, the starting point of the historical propagation network is the article publisher, and the first layer of nodes is the initial propagation node set.

[0072] S2: Based on the historical propagation network of all articles, construct a propagation network profile (GP) and calculate the reading latency of nodes on the historical propagation network.

[0073] In this embodiment, the construction of the propagation network profile is carried out as follows:

[0074] S201: For any node, use a node behavior data table to represent the historical behavior data of that node, and convert it into a node expressiveness table.

[0075] The node behavior data table includes three fields: reading time, article tags, and reading duration. The reading duration is used to indicate the user's level of attention to the article. In practice, if there are e-commerce links in the article, the user's order behavior can also be indirectly expressed by the reading duration.

[0076] The reading time refers to a pre-defined time period.

[0077] In this embodiment, the 24 hours of a day are divided into 24 time periods. 0:00 to 1:00 is the first time period, 1:00 to 2:00 is the second time period, and so on, until 23:00 to 24:00 is the 24th time period.

[0078] If the reading time of a node falls within a certain time period, then the value of the reading time is the value of that time period.

[0079] The reading time is the length of time spent reading a specific article tag at a specific reading moment, corresponding to the reading node.

[0080] The reading time refers to the length of time a node takes to read a specific tag at a given moment, calculated in seconds. The reading time is cumulative. For example, at a certain moment, for a specific tag (such as beer), if two articles have reading times of 25 seconds and 80 seconds respectively, then the total reading time for this tag at that moment is 105 seconds.

[0081] The node performance table includes reading time number, article tag number, and reading ability, which correspond to the three fields in the node behavior data table, respectively.

[0082] These three fields are the calculable numeric values ​​of three fields in the node behavior data table; for example, the dissemination time of 1:00 to 2:00 corresponds to the dissemination time number 2, the article tag value of culture corresponds to the article tag number 1, and reading ability corresponds to the reading time in the node behavior data table;

[0083] Let p(i, t, g) represent the historical reading power (time) of node i for article with label g at time t.

[0084] As shown above, the specific structure of the node behavior data table is shown in Table 1 below:

[0085] Table 1: Node Behavior Data Table

[0086]

[0087]

[0088] The specific structure of the node performance table is shown in Table 2 below:

[0089] Table 2: Node Performance Table

[0090] Node Reading time number Article label number Reading power 2 2 1 40 2 2 5 40 5 8 3 110 4 8 6 56

[0091] S202: For any edge formed by two nodes, use an edge propagation data table to represent the historical propagation data on this edge, and convert it into an edge propagation force table.

[0092] The side propagation data table includes three fields: propagation time, article tag, and propagation count.

[0093] The rules for determining the propagation time are similar to those for the reading time in the node behavior data table. The article tag is consistent with the node behavior data table. The article tag is the tag of the article propagated by this side. If there are multiple tags for an article, there will be multiple data entries in the table. Each data entry is for only one tag, and the propagation count for each tag is incremented by 1.

[0094] The edge propagation power table includes propagation time number, article tag number, and propagation power, which correspond to the three fields in the edge propagation data table, respectively.

[0095] These three fields are the computable numerical values ​​of the three fields in the edge propagation data table. The propagation time number and article tag number have the same meaning as the reading time number and article tag number in the node expressiveness table. The value of propagation power is equal to the number of propagations.

[0096] Let pw(i,j,t,g) represent the propagation power of edge (i,j) on the article with label g at time t, that is, the historical data of node i propagating the article with label g to node j at time t.

[0097] As shown above, the specific structure of the edge propagation data table is shown in Table 3 below:

[0098] Table 3: Side Propagation Data Table

[0099] Edge Propagation time Article label Propagation times 1,2 1:00-2:00 Culture 1 1,2 1:00-2:00 Beer 2 3,5 7:00-8:00 Electric appliance 1 3,4 7:00-8:00 Baijiu 3

[0100] The specific structure of the edge propagation force table is shown in Table 4 below:

[0101] Table 4: Side Propagation Force Table

[0102] Edge Propagation time number Article label number Propagation power 1,2 2 1 1 1,2 2 5 2 3,5 8 3 1 3,4 8 6 3

[0103] S203: Construct the propagation network profile GP based on the nodes and edges in the historical propagation network of all articles;

[0104] like Figure 2 As shown, tf i，j Describes the edge propagation force table for edge (i, j), ef i The node expressiveness table for node i, the edge propagation force table for each edge, and the node expressiveness table for each node are all derived from the historical propagation data of all articles.

[0105] Based on the propagation network profile, calculate the reading delay of nodes on the historical propagation network;

[0106] In this embodiment, the reading delay is calculated as the numerical difference between the reading time of the current node for a certain tagged article and the receiving time of the tagged article;

[0107] When calculating reading latency, a node will read an article within two days of receiving it by default. If multiple preceding nodes send articles of the same tag type within two days, the most recent one will be used. That is, the article being read is the one that was sent from the preceding node with the most recent arrival time.

[0108] For example, (1) if node 5 reads an article with tag 2 at time 8 and finds that node 8 sent an article with tag 2 at time 4 (which is the time of the previous day), then this article is attributed to the sharing of node 8, and the reading delay is 28.

[0109] (2) Node 5 reads an article with tag 2 at time 8 and finds that it received an article with tag 2 from nodes 8 and 2 at times 6 and 7 (times 6 and 7 are the same time on the same day). Therefore, it attributes this article to the sharing of node 2 and the reading delay is 1.

[0110] The default node will share the article immediately after reading it, meaning the sharing time is the same as the reading time.

[0111] Let d(i, t, g) represent the reading delay of node i for article label g at time t.

[0112] As shown above, the node expressiveness table is expanded to include a reading delay attribute. The specific table structure is shown in Table 5 below.

[0113] Table 5: Node Performance Table Including Reading Delay

[0114] Node Reading time number Article label number Reading power Reading delay 2 2 1 40 2 2 2 5 40 1 5 8 3 110 0 4 8 6 56 8

[0115] S3: Predict the reading time, reading delay, and probability of a node spreading the article to other nodes after receiving it.

[0116] In this embodiment, the reading time of a node after receiving an article is predicted, and the specific calculation is as follows:

[0117] Based on historical reading data, predict the reading time of a node after receiving an article with a certain tag at a certain time.

[0118] If the node has a reading record, the predicted reading duration p is calculated as follows:

[0119] p = (w1*p1 + w2*p2 + w3*p3) / 3*λ

[0120] w1, w2, and w3 represent weight coefficients, and w1 <w2<w3,w1+w2+w3=1;

[0121] p1 represents the average reading time calculated based on all reading records on this node;

[0122] p2 represents the average reading time calculated based on reading records of the same type of tag on this node;

[0123] p3 represents the average reading time calculated based on the reading records of articles with the same tag and arriving at the same time on this node;

[0124] If the node has no reading records, the predicted reading duration p is calculated as follows:

[0125] p=p'*λ

[0126] p' represents the average historical reading time of all nodes on the historical propagation network;

[0127] Wherein, λ is a perturbation coefficient, which is a random number between 0.8 and 1.2.

[0128] In this embodiment, the reading delay of a node after receiving an article is predicted, and the specific calculation is as follows:

[0129] Based on historical delay data for article reading, predict the reading delay of a node after receiving an article with a certain tag at a certain time.

[0130] If the node has a reading record, the predicted reading delay is calculated as follows:

[0131] d = (w1*d1 + w2*d2 + w3*d3) / 3*λ

[0132] w1, w2, and w3 represent weights, and w1 <w2<w3,w1+w2+w3=1;

[0133] d1 represents the average reading latency calculated based on all reading records on this node;

[0134] d2 represents the average reading latency calculated based on reading records of the same type of label on this node;

[0135] d3 represents the average reading latency calculated based on the reading records of articles with the same tag that arrived at the same time on this node;

[0136] If this node has no reading records, the predicted reading delay is calculated as follows:

[0137] d=d'*λ

[0138] d' is the average historical reading latency of all nodes on the historical propagation network;

[0139] λ represents a perturbation coefficient, which is a random number between 0.8 and 1.2.

[0140] In this embodiment, the probability of a node propagating an article to other nodes is predicted as follows:

[0141] Let q(i, j, t, g) represent the probability that node i will propagate the article with label g to node j at time t.

[0142] If there is no historical data propagated between node i and node j at any time, i.e., pw(i,j,t,g)=0, then the probability q(i,j,t,g) of node i propagating the article to node j is predicted to be ρ, where ρ is a set value, which is a decimal less than 1 and close to 0.

[0143] If there is historical propagation data between node i and node j, i.e., pw(i, j, t, g) ≠ 0, the probability that node i will propagate data with label g to node j at time t is calculated as follows:

[0144] Calculate the total propagation force between node i and its neighboring nodes under the same label and time, and divide it by the propagation force pw(i,j,t,g) between node i and node j under the same label g and time t. Multiply the resulting ratio by β, and use this result to represent q(i,j,t,g), where β is a random number between 0.5 and 1.

[0145] Multiplying the above calculation result by an article quality coefficient may result in a value greater than 1 or less than 1.

[0146] If the propagation probability from node i to node j is greater than a certain threshold γ, and the total propagation delay does not exceed a specified value, then it means that the propagation can reach node j.

[0147] S4: For a given node, in a cascading calculation mode, search downwards for other nodes that can propagate, until the predicted propagation probability is less than the set threshold γ or the total propagation delay exceeds the preset value, thus forming a predicted propagation tree with that node as the root node.

[0148] The sum of the products of reading time and probability for all nodes on the propagation tree is the expected influence of that node.

[0149] S5: Based on the propagation tree, select several initial propagation nodes using a heuristic method that selects k nodes.

[0150] The specific process is as follows:

[0151] Let Q be the set of unselected nodes and S be the set of selected seed nodes.

[0152] S501: Select the node q with the greatest expected influence from Q;

[0153] S502: Add q to set S;

[0154] S503: Remove q and the propagation tree rooted at q from Q;

[0155] S504: If the number of nodes in S is less than k, and there are still elements in Q, then recalculate the expected influence and propagation tree of all nodes in Q, and return to S501.

[0156] The following example illustrates a dissemination network of 50 people, with each person representing a node. There are ten tags for each article, represented by numbers 1 to 10. Each article has a dissemination history, and each article has two tags. The following ten tables represent the historical dissemination data of ten articles; the reading times in the tables are converted into corresponding reading time numbers.

[0157] Historical dissemination data for Article 1 (articles tagged with 5 and 10).

[0158] Node number Reading time Reading time number Reading duration (seconds) 2 7:57 8 26 1 8:34 9 30 3 10:37 11 65 4 9:21 10 21 5 10:01 11 46 7 12:27 13 64 8 11:15 12 58 11 14:22 15 81 10 13:31 14 45 15 17:43 18 89 24 20:48 21 92 37 11:58 12 42 41 19:35 20 42

[0159] Historical dissemination data for Article 2 (articles tagged with 1 and 6).

[0160] Node number Reading time Reading time number Reading duration (seconds) 43 9:50 10 67 33 10:27 11 37 7 11:15 12 54 1 12:06 13 28 23 11:12 12 30 2 13:31 14 88 44 14:46 15 94 39 15:00 15 89 4 17:35 18 42 16 16:34 17 39 9 19:14 20 23 3 17:49 18 66 18 20:19 21 24

[0161] Historical dissemination data for Article 3 (Articles tagged with 2 and 7).

[0162] Node number Reading time Reading time number Reading duration (seconds) 4 11:39 12 45 35 12:37 13 81 7 16:18 17 41 20 16:18 17 55 40 17:29 18 23 3 15:56 16 83 19 0:18 1 22 43 23:33 24 33 24 6:18 7 59 25 1:22 2 80 16 9:24 10 48 26 15:48 16 74 9 12:37 13 86

[0163] Historical dissemination data for Article 4 (Articles tagged with 3 and 8).

[0164]

[0165]

[0166] Historical dissemination data for article 5 (articles tagged with 4 and 9).

[0167] Node number Reading time Reading time number Reading duration (seconds) 4 10:36 11 22 22 11:05 12 99 1 12:05 13 27 7 14:40 15 13 46 15:19 16 39 14 15:19 16 98 12 17:34 18 15 23 19:57 20 38 3 21:34 22 37 21 21:06 22 26 32 22:51 23 58 9 23:53 24 40 27 1:45 2 23

[0168] Historical dissemination data for Article 6 (Articles tagged with 1 and 5).

[0169] Node number Reading time Reading time number Reading duration (seconds) 13 15:37 16 21 41 17:11 18 45 25 17:44 18 81 2 19:24 20 62 9 21:21 22 91 47 21:24 22 14 40 23:19 25 33 24 0:05 1 61 34 1:21 2 92 11 3:20 4 32 7 4:47 5 33 20 6:14 7 56 13 15:37 16 21

[0170] Historical dissemination data for Article 7 (Articles tagged with 2 and 6).

[0171] Node number Reading time Reading time number Reading duration (seconds) 47 15:11 16 86 30 16:39 17 16 27 18:04 19 69 37 18:08 19 52 38 18:56 19 93 11 20:59 21 44 33 21:00 21 97 2 22:53 23 22 36 23:52 24 86 40 23:11 24 71 13 23:24 24 12 47 15:11 16 86 30 16:39 17 16

[0172] Historical dissemination data for Article 8 (Articles tagged with 3 and 7).

[0173]

[0174]

[0175] Historical dissemination data for Article 9 (Articles tagged with 4 and 8).

[0176] Node number Reading time Reading time number Reading duration (seconds) 3 10:15 11 32 23 10:39 11 86 14 11:29 12 25 12 12:17 13 73 38 14:18 15 86 34 15:35 16 86 9 17:03 18 65 39 17:22 18 46 41 17:15 18 90 49 19:12 20 85 12 20:42 21 34 3 10:15 11 32 23 10:39 11 86

[0177] Historical dissemination data for Article 10 (Articles tagged with 6 and 9).

[0178] Node number Reading time Reading time number Reading duration (seconds) 3 8:59 9 33 24 10:10 11 69 33 10:49 11 78 4 12:58 13 58 1 14:55 15 87 37 15:28 16 17 32 17:25 18 40 10 19:02 20 88 17 19:29 20 31 15 21:27 22 85 19 23:26 24 79 12 0:08 1 58 2 0:04 1 60

[0179] A node performance table was obtained based on the article's historical dissemination data (data with a performance of 0 are not listed in the table below).

[0180]

[0181]

[0182]

[0183]

[0184] A node reading delay table was obtained based on the article's historical dissemination data (data with a delay of 0 is not listed in the table below).

[0185]

[0186]

[0187]

[0188] The side propagation power table is obtained based on the historical propagation data of the article (data with propagation power of 0 are not listed in the table below).

[0189]

[0190]

[0191]

[0192]

[0193] Predict the reading time of the node after receiving the article. Due to the large amount of data, only the predicted reading time of the node numbered 3 for different types of articles at each time point is listed here, and data with a predicted reading time of 0 are not listed in the table below.

[0194]

[0195]

[0196]

[0197]

[0198] Predict the propagation probability of the article on the nodes (data with a propagation probability of 0 are not listed in the table below).

[0199]

[0200]

[0201]

[0202]

[0203] Predict the node's influence (data with an influence of 0 is not listed in the table below).

[0204]

[0205]

[0206]

[0207] Select the initial set of propagation nodes and the propagation time. The number of nodes is specified manually. Here are the options for the initial propagation node set to include 2 nodes (the combination of propagation time number and article tag number that cannot be selected or does not show a clear trend based on historical data is not listed in the table below).

[0208]

[0209]

[0210] This invention is not limited to the specific embodiments described above. The invention extends to any new feature or combination disclosed in this specification, as well as any new method or process step or combination disclosed herein.

Claims

1. A method for social network influence maximization based on node reading effect and delay, characterized in that, The method comprises the following steps: S1: obtaining historical transmission data of multiple articles, and constructing a historical propagation network with time sequence for each article; S2: constructing a propagation network portrait according to the historical propagation network of all articles, and calculating the reading delay of the nodes on the historical propagation network; S3: predicting the reading duration of a node after receiving an article, the reading delay, and the probability of propagating the article to other nodes; S4: for a certain node, in a cascading calculation mode, finding other propagable nodes downward until the predicted propagation probability is less than a set threshold γ or the total propagation delay exceeds a preset specified value, thereby forming a predicted propagation tree with the node as a root node; The sum of the product of the reading duration and the probability of all nodes on the propagation tree is the predicted influence of the node; S5: screening a plurality of initial propagation nodes based on the propagation tree.

2. The social network influence maximization method of claim 1, wherein, In step S1, the historical propagation network is constructed based on node number, reading time, reading duration, propagation node, receiving node, and propagation time; The starting point of the historical propagation network is an article publisher, and the first layer nodes are an initial propagation node set. 3.The social network influence maximization method of claim 1, wherein, In step S2, the construction of the propagation network portrait is as follows: S201: for any node, the historical behavior data on the node is represented by a node behavior data table and converted into a node performance table; S202: for an edge formed by any two nodes, the historical propagation data on the edge is represented by an edge propagation data table and converted into an edge propagation force table; S203: according to the nodes and edges in the historical propagation network of all articles, the propagation network portrait is constructed, and the reading delay of the nodes on the historical propagation network is calculated according to the propagation network portrait.

4. The social network influence maximization method of claim 3, wherein, The node behavior data table comprises three fields of reading time, article label, and reading duration; The reading time is a pre-divided corresponding time period; The reading duration is the reading time length of the corresponding node at a certain reading time for a certain article label; The node performance table comprises reading time number, article label number, and reading force, which correspond to the three fields in the node behavior data table, respectively.

5. The social network influence maximization method of claim 3, wherein, The edge propagation data table comprises three fields of propagation time, article label, and propagation times; The edge propagation force table comprises propagation time number, article label number, and propagation force, which correspond to the three fields in the edge propagation data table, respectively.

6. The social network influence maximization method of claim 1, wherein, The calculation of the reading delay is the numerical difference between the reading time of the current node for a certain label article and the receiving time of the label article; When calculating the reading delay, by default, the node reads within a preset time after receiving the article, and if multiple previous nodes transmit articles of the same label type within the preset time, the data received at the latest time is used as the reference.

7. The social network influence maximization method of claim 1, wherein, The prediction of the reading duration of a node after receiving an article is calculated as follows: If the node has a reading record, the predicted reading duration is calculated as follows: p = (w1 * p1 + w2 * p2 + w3 * p3) / 3 * λ w1, w2, and w3 represent weight coefficients, and w1 < w2 < w3, w1 + w2 + w3 = 1; p1 represents the average reading duration calculated based on all reading records on the node; p2 represents the average reading duration calculated based on reading records of the same label on the node; p3 represents the average reading duration calculated based on reading records of articles of the same label and arriving at the same time on the node; If the node has no reading record, the predicted reading duration is calculated as follows: p=p' * λ p' represents the average value of historical reading duration of all nodes on the historical propagation network; wherein λ is a perturbation coefficient, which is a random number between 0.8 and 1.

2. 8.The social network influence maximization method of claim 1, wherein, The reading delay of a node after receiving an article is predicted, and the specific calculation is as follows: If the node has reading records, the predicted reading delay is calculated as follows: d=(w1*d1+w2*d2+w3*d3) / 3*λ w1, w2, w3 represent weights, and w1 < w2 < w3, w1 + w2 + w3 = 1; d1 represents the average reading delay calculated based on all reading records on the node; d2 represents the average reading delay calculated based on reading records of the same label on the node; d3 represents the average reading delay calculated based on reading records of articles of the same label and arriving at the same time on the node; If the node has no reading record, the predicted reading delay is calculated as follows: d=d' * λ d' represents the average value of historical reading delay of all nodes on the historical propagation network; λ represents a perturbation coefficient, which is a random number between 0.8 and 1.

2. 9.The social network influence maximization method of claim 1, wherein, The probability of a node propagating an article to other nodes is predicted, and the specific process is as follows: If there is no historical propagation data between node i and node j at any time, the probability of node i propagating an article to node j is ρ, which is a set value and is taken as a decimal less than 1 and approaching 0; If there is historical propagation data between node i and node j, the probability of node i propagating data with label g to node j at time t is calculated as follows: The sum of the propagation force of the same label and the same time between the node i and the adjacent nodes is calculated, and the obtained proportion value is multiplied by β, and β is a random number between 0.5 and 1. 10.The social network influence maximization method of claim 1, wherein, In step S5, the selection of the initial propagation node, and the specific process is as follows: Let the set of unselected nodes be Q, and the set of selected seed nodes be S; S501: Select a node q with the largest predicted influence from Q; S502: Add q to set S; S503: Remove q and the propagation tree with q as the root node from Q; S504: If the number of nodes in S is less than k and there are still elements in Q, recalculate the predicted influence of all nodes in Q and the propagation tree, and return to S501.

Images