Social network relationship prediction method based on Monte Carlo simulation network enhancement technology

Abstract

The invention relates to a social network relation prediction method based on a Monte Carlo simulation network enhancement technology. Compared with the prior art, the social network relation prediction method based on the Monte Carlo simulation network enhancement technology overcomes the defects that the prediction accuracy is reduced and the robustness is insufficient due to the fact that modeling is conducted directly based on an incomplete or noisy network structure. The method comprises the following steps: constructing a social network graph model; extracting structural features and attribute features between each pair of users; carrying out initial probability modeling; performing Monte Carlo enhanced sampling based on the initial connection probability; and obtaining a social network relationship prediction result. According to the method, the uncertainty of the topological structure caused by noise interference and measurement errors in a real network is effectively relieved through the Monte Carlo enhancement strategy, and the accuracy and robustness of social relation prediction are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of big data analysis, in particular to a social network relationship prediction method based on a Monte Carlo simulation network enhancement technology. BACKGROUND

[0002] With the rapid development of Internet technology and social platforms, social networks have become an important carrier of human behavior and information dissemination. Social relationship prediction, as a key problem in network science and data mining, has attracted widespread attention. The goal is to predict potential relationships that may be established between users based on existing network connection information, which has important application value in community division, personalized recommendation, anti-fraud identification, and propagation path analysis.

[0003] At present, most mainstream social relationship prediction methods are based on network topology structure information, using similarity measures between nodes (such as the number of common neighbors, Adamic-Adar index, resource allocation index, etc.) or through machine learning models (such as logistic regression, support vector machine, graph embedding method) to build prediction models. These methods indeed show good performance on static, fully observable networks, especially under ideal conditions with complete and less error data.

[0004] However, previous research has achieved remarkable results in certain specific situations, but the core premise condition is often based on the assumption of completely accurate network structure. In fact, social networks in the real world are generally affected by noise pollution, sampling bias, measurement error, node shielding, and information loss, making the obtained network topology structure have high uncertainty and incompleteness. These unavoidable disturbances can cause network connection information to deviate or even be wrong. Under this background, if the traditional relationship prediction method is directly based on the network data with errors for modeling, it will often amplify the uncertainty in the initial data, thereby affecting the stability and generalization ability of the model. SUMMARY

[0005] The purpose of the application is to solve the defects of the prior art that directly based on incomplete or noisy network structure for modeling, resulting in decreased prediction accuracy and insufficient robustness, and to provide a social network relationship prediction method based on Monte Carlo simulation network enhancement technology to solve the above problems.

[0006] In order to achieve the above purpose, the technical scheme of the application is as follows:

[0007] A social network relationship prediction method based on Monte Carlo simulation network enhancement technology, comprising the following steps:

[0008] 11) Construct a social network graph model;

[0009] 12) Extracting structural features and attribute features between each pair of users;

[0010] 13) Initial probability modeling;

[0011] 14) Monte Carlo enhanced sampling based on initial connection probability;

[0012] 15) Obtaining social network relationship prediction results: repeating the feature extraction and training process in the enhanced network, averaging the prediction results of multiple enhanced networks, obtaining edge probability estimation, and obtaining social network relationship prediction results.

[0013] The construction of the social network graph model includes the following steps:

[0014] Extracting a social network ,

[0015] representing a set of users of a social network, representing the i-th user, representing the number of users;

[0016] representing a set of relationships between any two users, is the i-th user and the j-th user whether there is a relationship between them; If represents the i-th user

[0017] and the j-th user have a relationship and are neighbors; if represents the i-th user and the j-th user have no relationship. The extraction of structural features and attribute features between each pair of users includes the following steps: 31) For any two users and , extract feature vector ;

[0018]

[0019] 32) The extracted features include: number of common neighbors ,

[0020]

[0021]

[0022] ​​​​​​​ , (1)

[0023] wherein, denotes a neighbor of a user , is a neighbor of a user ;

[0024] resource allocation indicator :

[0025] , (2)

[0026] wherein, is a neighbor of a user ;

[0027] Adamic-Adar indicator ,

[0028] ; (3)

[0029] node degree similarity ,

[0030] ; (4)

[0031] age similarity ,

[0032] ,

[0033] wherein, and represent the age of a user and , respectively;

[0034] interest similarity ,

[0035] (6)

[0036] wherein, and represent the set of interest of a user and , respectively;

[0037] geographical distance similarity ,

[0038] , (7)

[0039] wherein, and represent the geographical location of a user and , respectively;

[0040] 33) user and feature vector is defined as:

[0041] (8)

[0042] The initial probability modeling is performed by using a multi-layer perception model to model the existence of a relationship between users, including the following steps:

[0043] Using a multi-layer perception model, the feature vector is used as input, is used as output, is the existence of a relationship between the th user and the th user ;

[0044] The training formula is as follows:

[0045] (9)

[0046] wherein, , , , are learnable parameters, is a sigmoid activation function,

[0047] , ,

[0048] is the initial connection probability of the users and .

[0049] The Monte Carlo enhanced sampling based on the initial connection probability includes the following steps:

[0050] 51) Based on the initial connection probability of the users and , an enhanced network is generated based on Monte Carlo sampling,

[0051] wherein, is the enhanced social network, is the user set of the social network, is the edge set of the enhanced social network;

[0052] ,

[0053] , (10)

[0054] In formula (10) is the generated random number, After the network is enhanced users With the users Whether there is a relationship between them;

[0055] 52) Repeat step 51) to generate Enhanced Network ,

[0056] ,in represents the number of enhanced networks,

[0057] For the A set of relationships between two users in an enhanced network, For the The first users With the users Whether there is a relationship between them;

[0058] if Indicates in The first in the enhanced network users With the users There is a relationship and they are neighbors;

[0059] if Indicates in The first in the enhanced network users With the users It doesn't matter.

[0060] The acquisition of the social network relationship prediction result includes the following steps:

[0061] 61) For each enhanced network , according to the steps of extracting structural features and attribute features between each pair of users, re-extract the corresponding features ,

[0062] use As input, As output, is the eigenvector, For the a first user and a second user ;

[0063] training the following multi-layer perception model:

[0064] , (11)

[0065] wherein, represents vector splicing, , , , is a learning parameter, is a connection probability of a first user and a second user in an enhanced network, is a Sigmoid activation function;

[0066] 62) fusing multi-network prediction results:

[0067] the following formula (5) is used to predict an edge probability between a first user and a second user :

[0068] , (5) wherein,

[0069] is a number of enhanced networks.

[0070] A computer readable storage medium, the storage medium has a computer program stored thereon, when the computer program is executed by a processor, a social network relationship prediction method based on Monte Carlo simulation network enhancement technology can be realized.

[0071] A computer device, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, when the processor is executed, a social network relationship prediction method based on Monte Carlo simulation network enhancement technology can be realized.

[0072] Advantages

[0073] Compared with the prior art, the social network relationship prediction method based on Monte Carlo simulation network enhancement technology effectively alleviates the topological structure uncertainty caused by noise interference and measurement error in the real network through the Monte Carlo enhancement strategy, significantly improves the accuracy and robustness of social relationship prediction, and has strong practical application value.

[0074] In the relationship prediction modeling process, the application not only considers the local structure similarity between nodes (such as the number of common neighbors, the resource allocation index, the Adamic-Adar index, etc.), but also introduces the attribute characteristics at the individual level (such as node degree, age, interest, geographical location, etc.), realizes the organic integration of structure information and attribute information, and improves the expression ability of the model for different types of network relationships.

[0075] The application estimates the connection probability between users by a multi-layer perception model, and generates multiple enhanced networks based on the probability by Monte Carlo sampling. The feature extraction and training prediction process are repeated in these enhanced networks, and multiple prediction results are fused, which effectively reduces the prediction deviation caused by observation error, incomplete information or structural noise of the original network, and enhances the robustness and stability of the model.

[0076] The method has clear structure, simple implementation, good scalability and practicability. It can be adapted to various types of social network data, and can be integrated into existing recommendation systems and community identification system frameworks, and has wide application and promotion prospects. It is superior to existing traditional prediction methods in accuracy, robustness and practicability, and can more effectively cope with the modeling challenges brought by data uncertainty in real social networks. BRIEF DESCRIPTION OF DRAWINGS

[0077] Figure 1 The method sequence diagram of the application;

[0078] Figure 2 The algorithm flowchart of the application. DETAILED DESCRIPTION

[0079] In order to have a further understanding and understanding of the structural characteristics and effects achieved by the application, the preferred embodiments and the accompanying drawings are described as follows:

[0080] As shown in Figure 1 and Figure 2 The social network relationship prediction method based on Monte Carlo simulation network enhancement technology of the application realizes the paradigm shift from "deterministic graph" to "probability graph set" by introducing structure probability modeling and simulation enhancement mechanism, effectively alleviating the model deviation caused by measurement error, information missing and other problems. This design idea breaks the limitation of traditional single graph prediction framework, solves the problem of how to model the connection prediction process stably when the network structure is uncertain, and has strong theoretical innovation and method universality.

[0081] The steps include:

[0082] First, construct a social network graph model.

[0083] Extract social network ,

[0084] represent the user set of social network, represent the first user, represent the number of users;

[0085] represent the relationship set between any two users, for the user and the user whether there is a relationship;

[0086] If represent the user and the user have a relationship and are neighbors; if represent the user and the user have no relationship.

[0087] Second, extract the structural features and attribute features between each pair of users, realize the organic integration of structural information and attribute information, and thus improve the expression ability of the model for different types of network relationships.

[0088] (1) For any two users and , extract features to form vector .

[0089] (2) The extracted features include:

[0090] The number of common neighbors ,

[0091] , (1)

[0092] Among them, represent the neighbors of user , is the neighbor of user ;

[0093] Resource allocation index :

[0094] , (2)

[0095] where, is a neighbor of user ;

[0096] Adamic-Adar index ,

[0097] ; (3)

[0098] node degree similarity ,

[0099] ; (4)

[0100] age similarity ,

[0101] ,

[0102] where, and represent the ages of users and , respectively;

[0103] interest similarity ,

[0104] (6)

[0105] where, and represent the interest sets of users and , respectively;

[0106] geographical distance similarity ,

[0107] (7)

[0108] where, and represent the geographical locations of users and , respectively.

[0109] (3) The feature vector and of user are defined as:

[0110] (8)

[0111] Third, initial probability modeling is performed to model using a probabilistic graph when there is uncertainty in the network structure, effectively alleviating model bias caused by measurement errors, information missing, and other problems.

[0112] An initial probability model of whether there is a relationship between users is modeled using a multi-layer perception model, which includes the following steps:

[0113] A multi-layer perception model is used to take the feature vector as input, as output, for the first user and the first user whether there is a relationship between them;

[0114] The training formula is as follows:

[0115] , (9)

[0116] wherein, , , , are learnable parameters, is a sigmoid activation function,

[0117] , ,

[0118] is the initial connection probability of users and .

[0119] Step 4: Based on the initial connection probability, Monte Carlo enhanced sampling is performed, which breaks the limitations of the traditional single graph prediction framework and uses multiple sampling graphs to more realistically approximate the real social network.

[0120] (1) Based on the initial connection probability of users and , enhanced network is generated based on Monte Carlo sampling,

[0121] wherein, is the enhanced social network, is the user set of the social network, is the edge set of the enhanced social network;

[0122] ,

[0123] , (10)

[0124] In equation (10), is a random number generated, For each enhanced network , the relationship between the first user and the second user is determined.

[0125] (2) Repeat step (1) to generate enhanced networks ,

[0126] , where denotes the number of enhanced networks,

[0127] is the relationship set between the two users in the first enhanced network , is the relationship between the first user and the second user in the first enhanced network .

[0128] If , it means that the first user and the second user have a relationship in the first enhanced network and are neighbors.

[0129] If , it means that the first user and the second user have no relationship in the first enhanced network .

[0130] Step 5: Obtain the social network relationship prediction result: repeat the feature extraction and training process in the enhanced network, average the prediction results of multiple enhanced networks, estimate the edge probability, and obtain the social network relationship prediction result.

[0131] (1) For each enhanced network , according to the steps of extracting the structural features and attribute features between each pair of users, re-extract the corresponding features ,

[0132] Use as input, as output, as the feature vector, as the first user and the second user in the first enhanced network .​​​​​​​​ users Whether there is a relationship between them;

[0133] Train the following multilayer perceptron model:

[0134] , (11)

[0135] in, represents vector concatenation, 、 、 、 To learn the parameters, For users and in the first The connection probability in the enhanced network is is the Sigmoid activation function.

[0136] (2) Fusion of multi-network prediction results:

[0137] Use formula (5) to predict the users With the users The probability of the edge between :

[0138] , (5)

[0139] in, is the number of enhanced networks.

[0140] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A social network relationship prediction method based on Monte Carlo simulation network enhancement technology, characterized in that: The following steps are involved: 11) Construct a social network graph model; 12) Extracting structural features and attribute features between each pair of users; 13) Conduct initial probability modeling; 14) Monte Carlo enhancement sampling based on initial connection probability; 15) Obtaining social network relationship prediction results: Repeat the feature extraction and training process in the enhanced network, average the prediction results of multiple enhanced networks, obtain the edge probability estimate, and obtain the social network relationship prediction results.

2. The social network relationship prediction method based on Monte Carlo simulation network enhancement technology according to claim 1, characterized in that: The construction of the social network graph model includes the following steps: Extracting social networks , represents the set of users of a social network, Indicates the users, Indicates the number of users; represents the set of relationships between any two users. For the users With the users Whether there is a relationship between them; if Indicates the users With the users There is a relationship and they are neighbors; if Indicates the users With the users It doesn't matter.

3. The social network relationship prediction method based on Monte Carlo simulation network enhancement technology according to claim 1, characterized in that: Extracting the structural features and attribute features between each pair of users comprises the following steps: 31) For any two users and , extract features to form vectors ; 32) The extracted features include: Number of common neighbors , ， （1） in, Represents a user Neighbors, is a user Neighbors; Resource allocation indicators : ， （2） in, is a user Neighbors; Adamic-Adar indicator , ； （3） Node degree similarity , ； （4） Age similarity , ， in, and Representing users and age; Interest similarity , ， （6） in, and Representing users and A collection of hobbies and interests; Geographic distance similarity , ， （7） in, and Representing users and geographical location; 33) User and The eigenvector of Defined as: 。 4. The social network relationship prediction method based on Monte Carlo simulation network enhancement technology according to claim 1, characterized in that: The initial probability modeling is to use a multi-layer perceptron model to perform initial probability modeling on whether there is a relationship between users, which includes the following steps: Using a multi-layer perceptron model, the feature vector As input, As output, For the users With the users Whether there is a relationship between them; The training formula is as follows: ， （9） in, 、 、 、 is a learnable parameter, is the Sigmoid activation function, ， ， For users and The initial connection probability.

5. The social network relationship prediction method based on Monte Carlo simulation network enhancement technology according to claim 1, characterized in that: The Monte Carlo enhancement sampling based on the initial connection probability includes the following steps: 51) Leverage users and The initial connection probability , Generate enhanced network based on Monte Carlo sampling , in, For enhanced social networks, A collection of users of a social network. is the edge set of the enhanced social network; ， ， （10） In formula (10) is the generated random number, After network enhancement users With the users Whether there is a relationship between them; 52) Repeat step 51) to generate Enhanced Network , ,in represents the number of enhanced networks, For the A set of relationships between two users in an enhanced network, For the The first users With the users Whether there is a relationship between them; if Indicates in The first in the enhanced network users With the users There is a relationship and they are neighbors; if Indicates in The first in the enhanced network users With the users It doesn't matter.

6. The social network relationship prediction method based on Monte Carlo simulation network enhancement technology according to claim 1, characterized in that: The acquisition of the social network relationship prediction result includes the following steps: 61) For each enhanced network , according to the steps of extracting structural features and attribute features between each pair of users, re-extract the corresponding features , use As input, As output, is the eigenvector, For the The first users With the users Whether there is a relationship between them; Train the following multilayer perceptron model: ，（11） in, represents vector concatenation, 、 、 、 To learn the parameters, For users and in the first The connection probability in the enhanced network is is the Sigmoid activation function; 62) Fusion of multi-network prediction results: Use formula (5) to predict the users With the users The probability of the edge between : ，（5） in, is the number of enhanced networks.

7. A computer-readable storage medium, characterized in that The storage medium stores a computer program. When the computer program is executed by the processor, the social network relationship prediction method based on Monte Carlo simulation network enhancement technology described in any one of claims 1 to 6 can be implemented.

8. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the social network relationship prediction method based on Monte Carlo simulation network enhancement technology as described in any one of claims 1 to 6 can be implemented.

Images