Symbolic social network-oriented positive information maximization propagation method, program, equipment and storage medium

Abstract

The invention provides a symbolic social network-oriented positive information maximum propagation method, a program, equipment and a storage medium. According to the method, link prediction is carried out by using a link prediction model based on a symbol potential factor, and the missing of a network link relationship is complemented. On the basis, seed node propagation information is selected by adopting a symbol-based PageRank algorithm, so that the positive information influence is maximized. According to the method, the problems of user relation missing and relation complexity in a current network are considered, the network structure can be improved through a link prediction technology, and the seed node with the most influence is selected through a seed node selection algorithm based on the complete network structure to spread information, so that the positive information influence of the information is maximized.

Description

Technical Field

[0001] The present invention belongs to the technical field of social network, and specifically relates to a method, program, device and storage medium for maximizing the dissemination of positive information in symbolic social network. Background Art

[0002] Influence maximization aims to select a small number of users in order to maximize the influence of information dissemination. There are two key issues: the first is how to select the most influential set of users at the beginning, which corresponds to the seed node selection problem; the second is how to make others influenced and forward the information, in fact, this is to choose a different information cascade model. Information dissemination can describe the spread dynamics of rumors, diseases, memes or marketing activities. The initial stage of dissemination is started by one or a group of nodes in the social network.

[0003] In order to describe the information diffusion process in online social networks, classic information propagation models include Kempe et al. [David Kempe, Jon M. Kleinberg, Tardos. Maximizing the spread of influence through a social network [C]. Knowledge Discovery and Data Mining, 2003: 137-146.] proposed the independent cascade (IC) model and the linear threshold (LT) model. Some researchers use epidemic models (SIS, SIR, SIRS, SEIR) to simulate the infection and recovery process of nodes in the network (Wang et al. [Wang X, Lu J, Wang Z, et al. Dynamics of discrete epidemic models on heterogeneous networks [J]. Physica A: Statistical Mechanics and its Applications, 2020, 539 (C): 267-282.]). By analogizing information to disease, Woo and Chen [Woo J, Chen H. Epidemic model for information diffusion in web forums: experiments in marketing exchange and political dialog [J]. Springerplus, 2016, 5 (1): 66-83.] used the SIR model to model the topic diffusion in online forums. Yun Chai et al. [Chai Y, Wang Y, Zhu LA Stochastic Information Diffusion Model in Complex Social Networks[J]. IEEE Access, 2019: 175897-175906.] proposed a random information diffusion model on complex networks. Based on the SIR model, this model takes into account the population disturbance and changes in social network connectivity in the social network, and introduces the birth rate and death rate of the population into the model, making the model closer to real life.Wu et al. [Wu X, Fu L, Long H, et al. Adaptive Diffusion of Sensitive Information in Online Social Networks [J]. IEEE Transactions on Knowledge and Data Engineering, 2021, 33 (8): 3020-3034.] used a constrained combined multi-arm bandit framework to achieve normal diffusion of non-sensitive information while minimizing the spread of sensitive information, where the size of the sensitive information diffusion is used as the reward of the bandit framework. With the rapid development of the field of neural networks, current work has begun to use deep learning for diffusion modeling. Gao et al. [Sheng G, Pang H, Gallinari P, et al. A Novel Embedding Method for Information Diffusion Prediction in Social Network Big Data [J]. IEEE Transactions on Industrial Informatics, 2017, 13 (4): 2097-2105.] proposed the use of an information embedding-based diffusion prediction (IEDP) model. Bahareh et al. [Fatemi B, Molaei S, Pan S, et al. GCN Fusion: An efficient graph convolutional network based model for information diffusion [J]. Expert Systems with Application, 2022 (Sep.): 202-215.] proposed an efficient information diffusion model based on graph convolutional networks, combining feature selection, node representation learning and information diffusion to propose an information diffusion model.

[0004] During the information propagation process, nodes have different importance (vitality) in the network. Nodes play a vital role in many processes on or related to complex networks, such as controllability, opinion formation, preventing catastrophic cascading failures, and identifying high-performing researchers. However, identifying important nodes is not an easy task. Sun et al. [Sun H, Du H, Huang J, et al. Leader-aware community detection in complex networks [J]. Knowledge and Information Systems, 2020, 62 (2): 639-668.] proposed a new leader-aware community detection algorithm that can find the community structure and the leader of each community. The leadership of each node is measured and each node is made to obey its local leader to form a dependency tree. Once all dependency trees are clear, the community structure emerges. Lu et al. [

[33] Lu P, Dong C, Guo Y. A Novel Method Based on Node's Correlation to Evaluate Important Nodes in Complex Networks [J]. Journal of Shanghai Jiaotong University (Science), 2022, 27 (5): 1-13.] proposed a new method based on node distribution and global influence in complex networks. The main idea is that the importance of linked nodes is not only related to their relative position in the network, but also to their correlation with each other. The nodes in the complex network are classified according to the distance matrix, and then the correlation coefficients between node pairs are calculated. From the perspective of the entire network, the global similarity centrality (GSC) is proposed based on the correlation and shortest distance between nodes. Dan-Dan et al. [Dan-Dan Lu. Leader-Based Community Detection Algorithmin Attributed Networks [J]. IEEE Access, 2021: 119666-119674.] proposed a leader-based method that combines topology and attribute information (TALB), using the attribute information between nodes in the network to construct an attribute similarity matrix, which is then combined with the network topology information to establish the dependency relationship between nodes. In reality, user relationships can be positive, negative, neutral, or unrelated. The above influence maximization studies have not considered the complexity of user relationships. Summary of the invention

[0005] The purpose of the present invention is to provide a method, program, device and storage medium for maximizing the dissemination of positive information in a symbolic social network.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A method for maximizing the dissemination of positive information in a symbolic social network, the specific steps are as follows:

[0008] Step 1: Social network structure detection: By analyzing social network application messages, the identity information of the communicating parties is extracted and input into the social network link relationship database;

[0009] This information is then input into the social network link relationship database to help build a complete network link relationship map, providing basic data support for subsequent link prediction, seed node selection and semantic propagation operations;

[0010] Step 2: Link prediction;

[0011] By adopting the symbolic latent factor model, the SLF vector of each node in the symbolic social network is learned to improve the structural information of the network; in the link prediction process, the positive link score, negative link score, neutral link score and no relationship score are calculated by calculating the symbolic latent factor vector of the node pair, so as to realize the prediction of the link between the nodes;

[0012] Step 3: Seed node selection;

[0013] By using the PageRank algorithm of the symbolic network, combined with the adjacency matrix and the SPR value of the node, the symbolic page ranking is calculated to measure the status and influence of each node in the network; after the SPR algorithm converges, the most influential individual node is selected as the seed node for information dissemination; in this way, the seed node selection device helps to improve the effect of information dissemination and optimize the information dissemination path in the social network;

[0014] Step 4: semantic propagation;

[0015] By updating the interest of susceptible individuals in the social network, their attitude towards the positive information being spread is influenced. According to different relationships, the interest of individuals is updated in different ways; after the interest is updated, the direction of information dissemination is influenced according to the interest of individuals.

[0016] Furthermore, the step 2 is specifically as follows:

[0017] Step 2.1: Set the initial interest level for each node in the social network;

[0018] Step 2.2: Use the missing link completion algorithm to predict links in the symbolic social network to improve the network structure information;

[0019] The most important thing about using the SLF model for link prediction is to learn the SLF vector of each node. The negative log-likelihood function is used as the objective function, and the SLF vector of the node is learned by minimizing the objective function; the definition is as follows:

[0020]

[0021] in f a (x) is an activation function: f a (x) = p o exp(x) / 1+p o (exp(x)-1);

[0022] The above formula contains four components, corresponding to positive, negative, neutral and no relationship respectively; in order to minimize the objective function, the coordinate descent method is used, in which other SLF vectors are fixed to update one SLF vector; therefore, the objective function becomes:

[0023]

[0024] In the above formula, and Used to indicate the successor neighbors with positive, negative, neutral, and no relationship with node u. and represents the predecessor neighbors with different social relationships with node u; the SLF vector of the node in this formula Taking partial derivatives respectively, we get:

[0025]

[0026] Furthermore, the step 3 is specifically as follows:

[0027] Step 3.1: Seed node selection;

[0028] If the information is in the initial stage of propagation (step = 0), the symbol-based PageRank algorithm is used to select influential nodes in the network as seed nodes to propagate information; if the information propagation stage is entered, individuals with an attitude of 1 in the current network are added to the set of infected individuals;

[0029] Step 3.2: Information dissemination.

[0030] Furthermore, the step 4 is specifically as follows:

[0031] Step 4.1: Traverse each node in the network;

[0032] If the node is susceptible to infection and is ready to receive information, add it to the set of infected individuals;

[0033] If the node is not in the set of infected individuals and is currently in its infection cycle, add it to the set;

[0034] If there are susceptible individuals in the current network but the current time is less than the start time of their infection cycle, add one unit of time to re-propagate information;

[0035] Step 4.2: Interest update;

[0036] Determine whether there are susceptible individuals in the infection cycle in the current network;

[0037] Calling the interest updating algorithm to update the interest of susceptible individuals;

[0038] Determine whether individuals with attitude 1 can return to attitude 0 with a certain probability;

[0039] Step 4.3: When all individuals in the current network are infectious or removed, information propagation ends.

[0040] A system for maximizing the dissemination of positive information in a symbolic social network, comprising a social network structure detection device, a link prediction device, a seed node selection device, and a semantic propagation device;

[0041] The social network structure detection device: extracts identity information of both parties in communication by analyzing social network application messages, and inputs the information into a social network link relationship database;

[0042] Link prediction device: initializes the interest of each node in the social network, performs link prediction on the symbolic social network through the missing link completion algorithm, and improves the network structure information;

[0043] The seed node selection device: This device uses the symbol-based PageRank algorithm to select seed nodes. The algorithm selects the most influential nodes to spread positive information and maximize the impact of information dissemination;

[0044] The semantic propagation device traverses each node in the network, increases the interest of individuals in the symbolic social network in the positive information by a certain value, and then continuously infects other nodes through seed nodes to achieve maximum propagation of positive information.

[0045] A computer device / equipment / system comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of a method for maximizing the dissemination of positive information in a symbolic social network.

[0046] A computer-readable storage medium stores a computer program / instruction, which, when executed by a processor, implements the steps of a method for maximizing the dissemination of positive information in a symbolic social network.

[0047] A computer program product, comprising a computer program / instruction, characterized in that: when the computer program / instruction is executed by a processor, the steps of a method for maximizing the dissemination of positive information in a symbolic social network are implemented.

[0048] The beneficial effects of the present invention are:

[0049] The present invention integrates key steps such as link prediction, semantic propagation, seed node selection and interest updating, and realizes the maximum propagation of positive information in symbolic social networks by continuously iterating information propagation and interest updating. The present invention takes into account the problems of missing user relationships and complex relationships in the current network, and can improve the network structure through link prediction technology. Based on the complete network structure, the most influential seed node is selected through the seed node selection algorithm to propagate information, thereby maximizing the positive information influence of the information. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a schematic diagram of the system of the present invention;

[0051] Figure 2 is a structural block diagram of the system described in the present invention;

[0052] Figure 3 It is the SLF model;

[0053] Figure 4 The impact of link prediction on PIMPSN (Epinions on the left and Slashdot on the right);

[0054] Figure 5 The impact of seed node selection algorithm on PIMPSN (Epinions on the left and Slashdot on the right);

[0055] Figure 6 for link prediction performance analysis;

[0056] Figure 7 is the influence of k1 and k2 on SLF. DETAILED DESCRIPTION

[0057] The present invention is further described below in conjunction with the accompanying drawings.

[0058] The working process of the system for maximizing the dissemination of positive information in a symbolic social network of the present invention comprises the following steps:

[0059] First, the present invention divides the information propagation process into:

[0060] Initial stage: At t = 0, the attitude values ​​of all individuals are 0, and the symbol-based PageRank algorithm is needed to select seed nodes. The algorithm selects k most influential nodes to spread positive information and maximize the impact of information dissemination.

[0061] Propagation stage: When t>0, select the susceptible individuals in the current symbolic social network, update their interest in the received information, and check whether the individuals whose attitude becomes 1 after the update still have a certain interest in this information. The update probability is p i,t For each individual v i For example, p i,t =1-x i,t If there are currently no susceptible individuals, then the time is incremented and the propagation function is called to propagate the information.

[0062] Termination phase: If all individuals are contagious or removed, information propagation ends.

[0063] Figure 1 A schematic diagram of the system of the present invention, comprising:

[0064] 1) Use the Signed Latent Factor (SLF) model to predict links and improve network structure information; 2) Use the Signed Page Rank Algorithm (SPR) based on symbols to select seed nodes; 3) Use the semantic propagation mechanism to intervene and maximize the spread of positive information. As shown in the figure. The semantic propagation mechanism means that when the network manager wants to maximize the spread of the published positive information, he first increases the interest of individuals in the symbolic social network in the positive information by a certain value, and then continuously infects other nodes through seed nodes to maximize the spread of positive information.

[0065] Figure 2 The structure block diagram of the system of the present invention is shown, comprising:

[0066] Social network structure detection device: By analyzing social network application messages, the identity information of the communicating parties is extracted and input into the social network link relationship database.

[0067] Link prediction device: initializes the interest degree of each node in the social network, performs link prediction on the symbolic social network through the missing link completion algorithm, and improves the network structure information.

[0068] Seed node selection device: This device uses the symbol-based PageRank algorithm to select seed nodes. The algorithm selects the most influential nodes to spread positive information and maximize the impact of information dissemination.

[0069] Semantic propagation device: traverses every node in the network, increases the interest of individuals in the symbolic social network in the positive information by a certain value, and then continuously infects other nodes through seed nodes to maximize the spread of positive information.

[0070] Figure 3 For the SLF model, specifically:

[0071] The current network has the problem of missing link relationships and complex user relationships. When the present invention wants to semantically propagate negative information, the first thing to do should be to improve the structural information of the network. The present invention uses a symbolic latent factor model to predict links in social networks. The prediction process is as follows Figure 3 Specifically, it can be described as first learning the SLF vector of each node from the given symbolic social network G, and then connecting the SLF vector node feature f of node u u , Then the node features of node u and node v are combined to form the node pair feature f of node pair (u, v) (u,v) , the positive and negative feedback between node pairs are calculated according to the symbolic latent factor vector of the node pairs, and then the positive link score, negative link score, neutral link score and no relationship score between the node pairs are calculated, and the largest value is selected as the link symbol between the node pairs.

[0072] The most important thing about using the SLF model for link prediction is to learn the SLF vector of each node. The present invention uses the negative log-likelihood function as the objective function and learns the SLF vector of the node by minimizing the objective function. The definition is as follows:

[0073]

[0074] in f a (x) is an activation function: f a (x) = p o exp(x) / 1+p o (exp(x)-1).

[0075] The above formula contains four components, corresponding to positive, negative, neutral and no relationship. In order to minimize the objective function, the coordinate descent method is used, in which other SLF vectors are fixed to update one SLF vector. Therefore, the objective function becomes:

[0076]

[0077] In the above formula, and Used to indicate the successor neighbors with positive, negative, neutral, and no relationship with node u. and represents the predecessor neighbors with different social relationships with node u. Taking partial derivatives respectively, we get:

[0078]

[0079] Calculate the link score of the node pair. The SLF model is used to calculate the score between the node pairs. It includes three steps:

[0080] Given a node pair (u,v), calculate and The inner product of calculate and The inner product of

[0081] Quantify the impact of positive and negative feedback on social relations, expressed as and Activation function f a (x) = p o exp(x) / 1+p o (exp(x)-1) constructs a nonlinear mapping of positive and negative feedback and their effects on the relationship. It normalizes the effect to the interval [p o ,1).

[0082] The positive, negative, neutral and no relationship scores between node pairs are defined as Such calculations are meaningful. Specifically, if user u admires user v very much and sees a small amount of bad things about user v (i.e. The possibility of (i) is very high, then the present invention tends to believe that there is a social relationship between user u and user v and that it is a positive social relationship. Other score calculations also have similar meanings.

[0083] In order to verify the effectiveness of the present invention, four real-world symbolic social networks are used in the experiment: WikiElec, WikiRfa, Slashdot and Epinions. As shown in Table 1:

[0084] Table 1

[0085]

[0086] 1. WikiElec and WikiRfa: WikiElec is a voting network for Wikipedia administrator elections, and WikiRfa is the latest version of Wikipedia. The two datasets contain positive, negative, and neutral votes (links), with average sparsity of 31.7% and 25.3%, respectively. It is worth noting that both WikiElec and WikiRfa contain more than 6% neutral votes - a proportion sufficient to test the impact of considering neutral links.

[0087] 2. Slashdot is a social network of technology-related news websites. Its average sparsity is 13.4%. Each user can mark another user as a friend (positive) or an enemy (negative).

[0088] 3. Epinions is a trust network for consumer review sites. Its average sparsity is 12.8%. Its members can decide to trust (positive) or distrust (negative) other users based on the quality of their reviews.

[0089] The present invention selects the following methods for comparison when conducting seed node selection experiments:

[0090] 1. The traditional PageRank method selects seed nodes.

[0091] 2. Randomly select a seed node.

[0092] The evaluation metric for maximizing the influence of positive information is the number of affected nodes in the network. When doing link prediction, this experiment uses the area under the curve (AUC) and micro-F1 score to evaluate the prediction performance. Larger AUC / micro-F1 values ​​indicate better performance, with an upper limit of 1.0. The prediction performance of positive, negative, and invalid social relationships is expressed as AUC@p, AUC@n, and AUC@non, respectively. The symbol prediction of the relationship between nodes uses micro-F1.

[0093] In the link prediction experiment, the dimension of the node representation is set to 64, k1=32, k2=32, and the activation function f a In the equation, p0=0.001, and the threshold value x for updating interest during the propagation of positive information is 0.7.

[0094] The present invention designs three experiments to verify the effectiveness of the proposed PIMPSN:

[0095] 1. Comparative experiment on maximizing influence. 1) By comparing PIMPSN and PIMPSN-SLF (without considering link prediction), it is verified that link prediction helps to improve the accuracy of finding seed nodes when maximizing influence propagation, thereby achieving better positive information propagation effect. 2) By comparing the impact of different seed node selection methods on influence propagation, the effectiveness of seed node selection methods is verified.

[0096] 2. Link prediction performance analysis experiment The purpose is to analyze the link prediction performance of the missing link completion algorithm proposed in the present invention.

[0097] 3. Parameter sensitivity analysis experiment: The purpose is to analyze the impact of the parameters involved in SLF on its link prediction effect.

[0098] Experiment 1: Impact maximization comparison experiment.

[0099] In this experiment, we first compare the effect of link prediction on maximizing the influence of positive information. Figure 4 The results in Figure 2 show the impact of using link prediction (PIMPSN) and not using link prediction (PIMPSN-SLF) on information propagation on the datasets Epinions and Slashdot, with the number of seed nodes varying from 20 to 100, with a step size of 20. It can be observed that when the number of seed nodes reaches 100, the number of nodes affected by PIMPSN is 13.5% higher than that of PIMPSN-SLF on the dataset Epinions, and 14.2% higher on the dataset Slashdot, which proves that link prediction before the start of positive information propagation can improve network structure information and significantly improve the effect of maximizing influence.

[0100] Next, the seed node selection algorithm used in the present invention is compared with other seed node selection algorithms, including random seed node selection (RANDOM) and traditional PageRank algorithm (PR), such as Figure 5 As shown. It can be observed that when the number of seed nodes is 100, on the Epinions dataset, the impact of SPR is 16.2% higher than PR and 31.5% higher than RANDOM, while on the Slashdot dataset, the impact of SPR is 21% higher than PR and 42.3% higher than RANDOM. The experimental results show that the seed node set selected by the SPR algorithm adopted in the present invention is better in maximizing influence, and as the number of selected seed nodes increases, the number of affected nodes in the network also increases. The experimental results show that in a symbolic social network, it is very necessary to consider the complex relationships between individuals to select seed nodes.

[0101] Experiment 2: Link prediction performance analysis experiment;

[0102] Figure 6 The prediction results of the missing link completion algorithm based on the SLF model adopted by the present invention on four data sets are shown. The horizontal axis represents different data sets, and the vertical axis is the value corresponding to different evaluation indicators. The higher the value, the better the prediction effect. It can be observed that the overall prediction effect on the data set Epinions is higher than that on other data sets, and the Epinions data set has more positive links than the other three. It can also be seen that on the data sets WikiElec and WikiRfa, the AUC@non index is higher than that of the other two data sets, because WikiElec and WikiRfa both have a certain proportion of neutral links, and the link prediction method of the present invention takes neutral links into account, which is conducive to improving the value of AUC@non.

[0103] Experiment 3: Parameter sensitivity analysis experiment;

[0104] The parameters involved in link prediction using the SLF model include the dimension k1 of the positive latent factor space, the dimension k2 of the negative latent factor space, and p0 in the activation function. The following analyzes the impact of these parameters one by one:

[0105] 1) The influence of k1 and k2;

[0106] Figure 7 This figure reflects the influence of parameters k1 and k2 on the link prediction performance of the SLF model on the dataset WikiElec. In this experiment, other parameters are fixed, n=100, p0=0.001, and the values ​​of k1 and k2 are changed in the set {8,16,32,64}. The figure is a three-dimensional surface graph, the horizontal axis is the change of k1 and k2 values, and the vertical axis corresponds to AUC@p, AUC@n, AUC@non and micro-F1 respectively. It can be seen that when k2=32, as k1 increases, the corresponding values ​​of AUC@p are 0.932, 0.941, 0.963 and 0.952 respectively, that is, with the fixed value of k2, as k1 increases, the overall performance of SLF increases first and then decreases. When k1 is small, SLF does not have enough representation ability, but when k1 is large, SLF tends to overfit. And compared with k2, k1 has a greater impact on performance. The reason is that symbolic social networks usually have more positive links than negative links, and k1 is closely related to positive links. When both k1 and k2 are 32, the prediction effect is the best.

[0107] 2) The influence of p0 in the activation function;

[0108] In order to investigate the influence of p0, this experiment fixed other parameters, k1=32, k2=32, and changed the value of p0 in the set {0.001, 0.005, 0.01, 0.05}. Table 2 shows the experimental results on the WikiElec dataset. It can be observed that the smaller p0 is, the better the SLF performance is.

[0109] Table 2

[0110]

[0111] Activation function f a (x) Construct a nonlinear mapping of positive feedback and negative feedback and their impact on the relationship. It normalizes the effect to the interval [p0,1). The larger p0 is, the larger the minimum value of the corresponding values ​​of positive feedback and negative feedback is, which will reduce the distinction between the relationships between nodes and lead to a decrease in SLF performance.

[0112] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for maximizing the dissemination of positive information in a symbolic social network, characterized by: The specific steps are as follows: Step 1: Social network structure detection: by analyzing social network application messages, extract the identity information of the communicating parties and input it into the social network link relationship database; This information is then input into the social network link relationship database to help build a complete network link relationship map, providing basic data support for subsequent link prediction, seed node selection and semantic propagation operations; Step 2: Link prediction; By adopting the symbolic latent factor model, the SLF vector of each node in the symbolic social network is learned to improve the structural information of the network; in the link prediction process, the positive link score, negative link score, neutral link score and no relationship score are calculated by calculating the symbolic latent factor vector of the node pair, so as to realize the prediction of the link between the nodes; Step 3: Seed node selection; By using the PageRank algorithm of the symbolic network, combined with the adjacency matrix and the SPR value of the node, the symbolic page ranking is calculated to measure the status and influence of each node in the network; after the SPR algorithm converges, the most influential individual node is selected as the seed node for information dissemination; in this way, the seed node selection device helps to improve the effect of information dissemination and optimize the information dissemination path in the social network; Step 4: semantic propagation; By updating the interest of susceptible individuals in the social network, their attitudes towards the positive information being spread can be influenced. According to different relationships, the interest of individuals is updated in different ways; after the interest is updated, the direction of information dissemination is affected according to the interest of individuals.

2. The method for maximizing the dissemination of positive information in a symbolic social network according to claim 1, characterized in that: The step 2 is specifically as follows: Step 2.1: Set the initial interest level for each node in the social network; Step 2.2: Use the missing link completion algorithm to predict links in the symbolic social network to improve the network structure information; The most important thing about using the SLF model for link prediction is to learn the SLF vector of each node. The negative log-likelihood function is used as the objective function, and the SLF vector of the node is learned by minimizing the objective function; the definition is as follows: in f a (x) is an activation function: f a (x) = p o exp(x) / 1+p o (exp(x)-1); The above formula contains four components, corresponding to positive, negative, neutral and no relationship respectively; in order to minimize the objective function, the coordinate descent method is used, in which other SLF vectors are fixed to update one SLF vector; therefore, the objective function becomes: In the above formula, and Used to indicate the successor neighbors with positive, negative, neutral, and no relationship with node u. and represents the predecessor neighbors with different social relationships with node u; the SLF vector of the node in this formula Taking partial derivatives respectively, we get:

3. The method for maximizing the dissemination of positive information in a symbolic social network according to claim 1, characterized in that: The step 3 is specifically as follows: Step 3.1: Seed node selection; If the information is in the initial stage of propagation (step = 0), the symbol-based PageRank algorithm is used to select influential nodes in the network as seed nodes to propagate information; if the information propagation stage is entered, individuals with an attitude of 1 in the current network are added to the set of infected individuals; Step 3.2: Information dissemination.

4. The method for maximizing the dissemination of positive information in a symbolic social network according to claim 1, characterized in that: The step 4 is specifically as follows: Step 4.1: Traverse each node in the network; If the node is susceptible to infection and is ready to receive information, add it to the set of infected individuals; If the node is not in the set of infected individuals and is currently in its infection cycle, add it to the set; If there are susceptible individuals in the current network but the current time is less than the start time of their infection cycle, add one unit of time to re-propagate information; Step 4.2: Interest update; Determine whether there are susceptible individuals in the infection cycle in the current network; Calling the interest updating algorithm to update the interest of susceptible individuals; Determine whether individuals with attitude 1 can return to attitude 0 with a certain probability; Step 4.3: When all individuals in the current network are infectious or removed, information propagation ends.

5. A system for maximizing the dissemination of positive information in symbolic social networks according to any one of claims 1 to 4, characterized in that: It includes a social network structure detection device, a link prediction device, a seed node selection device and a semantic propagation device; The social network structure detection device: extracts identity information of both parties in communication by analyzing social network application messages, and inputs the information into a social network link relationship database; Link prediction device: initializes the interest of each node in the social network, performs link prediction on the symbolic social network through the missing link completion algorithm, and improves the network structure information; The seed node selection device: This device uses the symbol-based PageRank algorithm to select seed nodes. The algorithm selects the most influential nodes to spread positive information and maximize the impact of information dissemination; The semantic propagation device traverses each node in the network, increases the interest of individuals in the symbolic social network in the positive information by a certain value, and then continuously infects other nodes through seed nodes to achieve maximum propagation of positive information.

6. A computer device / equipment / system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

8. A computer program product comprising a computer program / instructions, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images