Social network opinion reconstruction method and device based on graph movement smoothness

Abstract

The application provides a social network opinion reconstruction method and device based on graph movement smoothing, wherein the method comprises the following steps: determining a sampling user set according to a reverse greedy sampling algorithm based on a pre-constructed directed social network model; determining graph movement smoothing degrees corresponding to all users in the directed social network model; obtaining opinion observation values corresponding to each user in the sampling user set; and reconstructing social network opinions of the users in the directed social network model based on the graph movement smoothing degrees and the opinion observation values. The method is aimed at a directed social network structure, and the sampling user set is determined through the reverse greedy sampling algorithm. The method can not only effectively improve the accuracy of social network opinion reconstruction, overcome the defects that the existing user opinion reconstruction method is mainly aimed at a non-directed network structure and cannot be directly extended to a directed network structure and the reconstruction mean square error is large, and realize accurate reconstruction of social network opinions, but also effectively improve the running efficiency of the reconstruction process.

Description

A method and apparatus for reconstructing opinions in social networks based on graph movement smoothness. Technical Field

[0001] This invention relates to the field of Internet technology, and in particular to a method and apparatus for reconstructing opinions in social networks based on graph motion smoothing. Background Technology

[0002] With the development of mobile communication and internet technology, more and more social networking platforms have gradually become an indispensable part of people's daily lives. Platforms such as Weibo, Twitter, Instagram, TikTok, and LinkedIn have built directional social network structures based on the following relationships between users. This directional social network structure can further bridge the gap between different types of users; for example, ordinary users can directly follow opinion leaders, thus obtaining information more conveniently and directly.

[0003] On the one hand, social networks can be used to effectively disseminate important announcements and influence the public. For example, during the pandemic, social media platforms such as Weibo can be used to announce epidemic prevention policies. On the other hand, malicious users may use social network platforms to spread extreme or false information, thereby threatening public safety.

[0004] Therefore, statistically analyzing and monitoring user opinions on social networks is crucial for managing social network platforms. However, real-world social networks are often quite large, and different users express their opinions and viewpoints at varying frequencies and times. Consequently, achieving a comprehensive observation of the opinion values ​​of all users is relatively difficult.

[0005] To address this, existing technologies offer a model-based graph signal reconstruction method. This method primarily targets undirected network structures, utilizing the eigenvalue decomposition of their symmetric graph Laplacian matrix to define a "low-pass graph signal" located in the eigenspace, thereby reconstructing the graph signal. However, due to the variance of observation noise or biases in the reconstruction method itself, errors will occur between the reconstructed graph signal and the true graph signal.

[0006] To reduce the reconstruction error of the aforementioned model-based graph signal reconstruction methods, existing technologies have designed graph signal sampling methods and efficient algorithms for large-scale networks. Although these methods have achieved good results on undirected network structures, the graph Laplacian matrix of directed network structures is not symmetric, and its complex eigenvalue decomposition greatly restricts the definition of frequency and low-pass characteristics. Therefore, these methods cannot be directly extended to directed network structures.

[0007] Therefore, how to solve the problem that existing user opinion reconstruction methods are mainly designed for undirected network structures and cannot be directly extended to directed network structures, and that their reconstruction mean square error is relatively large, is an important issue that urgently needs to be addressed in the field of Internet technology. Summary of the Invention

[0008] This invention provides a method and apparatus for reconstructing opinions in social networks based on graph motion smoothing, which overcomes the shortcomings of existing user opinion reconstruction methods that are mainly designed for undirected network structures and cannot be directly extended to directed network structures, and whose reconstruction mean square error is relatively large, thereby achieving accurate reconstruction of opinions in social networks.

[0009] On one hand, the present invention provides a method for reconstructing opinions in a social network based on graph movement smoothness, comprising: determining a set of sampled users based on a pre-constructed directed social network model and a reverse greedy sampling algorithm; determining the graph movement smoothness corresponding to all users in the directed social network model and obtaining the opinion observations corresponding to each user in the sampled user set;

[0010] Based on the graph movement smoothness and the opinion observations, the social network opinions of users in the directed social network model are reconstructed.

[0011] Furthermore, the step of determining the sampled user set based on the pre-built directed social network model and the reverse greedy sampling algorithm includes: determining a second historical opinion reconstruction value considering observation noise based on the historical opinion reconstruction value; calculating the mean square error of the reconstruction based on the second historical opinion reconstruction value; constructing an opinion sampling problem with minimum mean square error based on the mean square error; and solving the opinion sampling problem using the reverse greedy sampling algorithm to obtain the sampled user set.

[0012] Further, the step of using the reverse greedy sampling algorithm to solve the opinion sampling problem and obtain the sampled user set includes: S111, initializing the current sampled user set as the complete set of all users; S112, using the objective function to calculate the mean square error after removing each user from the current sampled user set; S113, determining the user to be removed corresponding to the minimum mean square error; S114, removing the user to be removed from the current sampled user set, jumping to step S112, and repeating steps S112-S114 until the number of users to be removed reaches a preset threshold, thereby obtaining the sampled user set.

[0013] Furthermore, the objective function is as follows:

[0014] ;

[0015] in, , For hyperparameters, , The autocorrelation matrix of the reconstructed historical opinion values. For noise variance, For the current sample set The function, For only the first A vector with one element being 1 and all other elements being 0.

[0016] Furthermore, the user to be removed is determined by a proxy objective function, which is as follows:

[0017] ;

[0018] ;

[0019] in, It is a matrix The List, , , , It is a matrix The List, , It is a matrix The OK, This represents the noise variance.

[0020] Furthermore, after determining the minimum proxy objective function in each round, the parameter matrix is ​​adjusted using a pre-defined formula. , as well as The preset formula is updated as follows:

[0021] ;

[0022] ;

[0023] ;

[0024] in, The user corresponding to the minimum agent objective function.

[0025] Further, the step of reconstructing the social network opinions of users in the directed social network model based on the graph movement smoothness and the opinion observations includes: obtaining the reconstructed opinion values ​​of the social network opinions using the following formula:

[0026] ;

[0027] in, For the reconstruction value of opinions, For the sampling matrix, For hyperparameters, For random walk-graph Laplace matrix, These are observations of opinions.

[0028] Secondly, the present invention provides a social network opinion reconstruction device based on graph movement smoothness, comprising: a sampling user set determination module, used to determine the sampling user set based on a pre-constructed directed social network model and according to a reverse greedy sampling algorithm; a graph movement smoothness determination module, used to determine the graph movement smoothness of the true opinion values ​​corresponding to all users in the directed social network model; an opinion observation value acquisition module, used to acquire the opinion observation values ​​of each user in the sampling user set; and a social network opinion reconstruction module, used to reconstruct the social network opinions of users in the directed social network model based on the graph movement smoothness and the opinion observation values.

[0029] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the social network opinion reconstruction method based on graph movement smoothness as described above.

[0030] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the social network opinion reconstruction method based on graph movement smoothness as described above.

[0031] This invention provides a method for reconstructing social network opinions based on graph motion smoothing. Based on a pre-constructed directed social network model, it determines the sampled user set using a reverse greedy sampling algorithm, determines the graph motion smoothness corresponding to all users in the directed social network model, and obtains the opinion observations corresponding to each user in the sampled user set. Based on the graph motion smoothness and opinion observations, it reconstructs the social network opinions of users in the directed social network model. This method, targeting directed social network structures, uses a reverse greedy sampling algorithm to determine the sampled user set. It not only effectively improves the accuracy of social network opinion reconstruction and overcomes the shortcomings of existing user opinion reconstruction methods, which are mainly designed for undirected network structures and cannot be directly extended to directed network structures, but also suffer from large mean square errors in reconstruction. This achieves accurate reconstruction of social network opinions and effectively improves the operational efficiency of the reconstruction process. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0033] Figure 1 is a flowchart illustrating the social network opinion reconstruction method based on graph movement smoothness provided by the present invention.

[0034] Figure 2 is a flowchart illustrating the reverse greedy sampling algorithm provided by this invention;

[0035] Figure 3 is a schematic diagram of the social network opinion reconstruction device based on graph movement smoothness provided by the present invention;

[0036] Figure 4 is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0038] Figure 1 shows a flowchart illustrating the social network opinion reconstruction method based on graph movement smoothness provided by the present invention. As shown in Figure 1, the method includes:

[0039] S110 determines the set of sampled users based on a pre-built directed social network model and a reverse greedy sampling algorithm.

[0040] It should be noted that the social network opinion reconstruction method based on graph movement smoothness provided by this invention is for directed social network structures.

[0041] It is understood that the directed social network model in this embodiment can be specifically modeled as a directed graph. Each node in the directed graph This represents a user in a social network, and the social network under consideration has a total of [number missing]. One user, that is In a directed graph, directed edges represent the influence relationships between users in a social network.

[0042] In particular, for directed edges Its representative user Able to influence users Since the social network under consideration is a directed graph structure, users Able to influence users This does not represent the user's It can also affect users .

[0043] The directed graph structure of a social network can use an adjacency matrix. To characterize, for elements in the adjacency matrix: if Then elements Representing users For users The magnitude of the impact; if Then elements .

[0044] Among them, users The in-degree can be defined as That is, except for users All other users besides the user The sum of the effects. Similarly, users can be defined. The out-degree is That is, users The sum of the effects on all other users.

[0045] For users Define his neighbor as That is, all things that can affect users A collection of user opinions.

[0046] It should also be noted that the directed social network model provided in this embodiment has strong connectivity, that is, the above-mentioned directed graph... It satisfies strong connectivity. Specifically, for a directed graph... Any two different users All of them have bidirectional reachable paths, and are directed graphs. No self-loop, that is .

[0047] As discussed above, each user in a social network influences and is also influenced by other users. If a real-world social network does not satisfy strong connectivity, each strongly connected component of the social network can be analyzed separately.

[0048] Based on the construction of a directed social network model, the sampling set is determined according to the reverse greedy sampling algorithm.

[0049] It's important to understand that reconstructing opinions from a social network requires sampling a number of users within the network and observing their opinion values. These sampled opinions are then used to reconstruct the opinions of all users on the network. The method used to sample these users significantly impacts the accuracy of the reconstructed opinions.

[0050] In view of this, this embodiment proposes a reverse greedy sampling algorithm to determine the set of sampling users. Specifically, based on the historical opinion reconstruction values ​​of users in the social network, considering the observation noise when observing user opinion values ​​during the reconstruction process, a second historical opinion reconstruction value corresponding to the historical opinion reconstruction value is determined, and the mean square error of reconstruction is calculated based on the second historical opinion reconstruction value.

[0051] Furthermore, based on the reconstructed mean square error, a minimum mean square error opinion sampling problem is constructed. This problem is then solved using a reverse greedy sampling algorithm to obtain the set of sampled users for reconstruction. The users in this sampled user set are the users in the aforementioned directed social network.

[0052] The reconstructed historical opinion value of users in the social network can be obtained by reconstructing historical opinion values ​​through existing user opinion reconstruction technology, or it can be obtained by reconstructing historical opinion values ​​through the reconstruction method provided by this invention. No specific limitation is made here.

[0053] S120, determine the graph movement smoothness for all users in the directed social network model.

[0054] It can be understood that, based on the directed social network model pre-built in step S110, graph movement smoothness is further used to characterize the smoothness of the aforementioned opinion observations.

[0055] Specifically, users' genuine opinion values ​​can be modeled as a directed graph. Signals on the node ,in, Representing users The actual opinion value. The user's opinion value is a real-valued scalar.

[0056] Taking Weibo discussion topics as an example, in a certain topic discussion scenario on Weibo, the user's opinion value reflects the user's emotional value towards the topic. A larger opinion value indicates that the user has a more positive attitude towards the topic being discussed, while a smaller opinion value indicates that the user has a more negative attitude towards the topic being discussed.

[0057] It's important to understand that in directed social networks, users' genuine opinion values ​​exhibit "convergence," meaning that users who are neighbors tend to have similar genuine opinion values. If two users are neighbors, and the stronger the influence relationship between them, the more similar their opinions will be. This convergence of user opinions on social networks essentially reflects the convergence of genuine opinion values. For directed graph structures It exhibits local similarity, or in other words, genuine opinion values. It is smooth.

[0058] In this embodiment, graph shift variation (GSV), a term used in graph signal processing research, is employed to characterize the smoothness of the true opinion values ​​of all users in a directed social network model. Graph shift variation is defined as follows:

[0059] (1)

[0060] in,

[0061] ;

[0062] ;

[0063] ;

[0064] In the definition of graph movement smoothness above, It is a normalized adjacency matrix, and It is a random walk graph Laplacian matrix. It is an identity matrix.

[0065] As can be seen from the above, the smoothness of graph movement measures the true opinion value. And the true opinion value after moving one step The differences between them. Actual opinion value. The smoothness of graph movement is essentially about the matrix. The quadratic form. Because It is semi-positive definite, therefore, the true opinion value graph movement smoothness It must be non-negative.

[0066] S130, obtain the opinion observation value corresponding to each user in the sampled user set.

[0067] Based on the determination of the sampling user set according to the reverse greedy sampling algorithm in step S110, the opinion observation value corresponding to each user in the sampling user set is further obtained. Specifically, this can be obtained using open-source sentiment analysis tools.

[0068] For example, in one specific embodiment, NLTK and SnowNLP are used to measure and extract user opinion observations.

[0069] S140 reconstructs users' social network opinions in a directed social network model based on graph movement smoothness and opinion observations.

[0070] It is understandable that, based on the graph movement smoothness corresponding to the opinion observations calculated in step S130, the social network opinions of users in the directed social network model are further reconstructed according to the graph movement smoothness.

[0071] Specifically, sampling is directed to social networks. ( ( ) users, and obtain their opinion observations. Let the set of sampled users be Obviously, the goal is to achieve this through sampling. Reconstructing the entire social network from individual opinion observations User's opinion It's an undetermined issue.

[0072] Because opinions in social networks tend to converge, they exhibit relatively low graph smoothness. Therefore, this embodiment of the invention utilizes graph movement smoothness as a regularization term to construct an opinion value reconstruction problem under the condition of sampling a subset of users and observing their opinions, as follows:

[0073] (2)

[0074] in, Let be the sampling matrix, which satisfies the following condition for the first... One sampling user have ,the remaining All elements in the array are 0. These are observations of opinions. For temporary variables to be optimized, To reconstruct hyperparameters, It is a random walk-graph Laplace matrix.

[0075] In the aforementioned issue of reconstructing opinion values, the first item... Reflects the temporary variables corresponding to the sampled users Compared with observed values The squared error between them, the second term This reflects temporary variables Smoothness. Parameters Smoothness regularization term controlled The weights in this optimization problem, namely the opinion value reconstruction problem.

[0076] The aforementioned opinion value reconstruction problem concerns the optimization variables. The objective function of the opinion value reconstruction problem is a quadratic convex function. To solve for the optimal reconstructed opinion, we let the objective function of the opinion value reconstruction problem be related to the optimization variables. By taking the derivative and setting it to zero, the optimal opinion reconstruction value can be obtained. The conditions to be met:

[0077] (3)

[0078] When the directed graph structure under consideration satisfies strong connectivity, the optimal opinion reconstruction value obtained from the above equation (3) is... There is a unique solution:

[0079] (4)

[0080] in, For the reconstruction value of opinions, For the sampling matrix, For hyperparameters, For random walk-graph Laplace matrix, These are observations of opinions.

[0081] It should be noted that the opinion reconstruction value can be obtained by efficiently solving equation (3) directly using the conjugate gradient method (CG).

[0082] In this embodiment, based on a pre-constructed directed social network model, a sample user set is determined using a reverse greedy sampling algorithm. The graph movement smoothness corresponding to all users in the directed social network model is also determined, and the opinion observations corresponding to each user in the sample user set are obtained. Based on the graph movement smoothness and opinion observations, the social network opinions of users in the directed social network model are reconstructed. This method, targeting directed social network structures, uses a reverse greedy sampling algorithm to determine the sample user set. This not only effectively improves the accuracy of social network opinion reconstruction and overcomes the shortcomings of existing user opinion reconstruction methods, which are mainly designed for undirected network structures and cannot be directly extended to directed network structures, but also suffer from large mean square errors in reconstruction. This achieves accurate reconstruction of social network opinions and effectively improves the operational efficiency of the reconstruction process.

[0083] Based on the above embodiments, further, based on a pre-built directed social network model, the set of sampled users is determined according to the reverse greedy sampling algorithm, including: determining a second historical opinion reconstruction value considering observation noise based on the historical opinion reconstruction value; calculating the mean square error of the reconstruction based on the second historical opinion reconstruction value; constructing an opinion sampling problem with minimum mean square error based on the mean square error; and solving the opinion sampling problem using the reverse greedy sampling algorithm to obtain the set of sampled users.

[0084] Understandably, in real-world opinion reconstruction scenarios, the measurement of user opinion values ​​often involves observation noise, which results in a certain degree of accuracy and mean square error in the opinion observations obtained using open-source sentiment analysis tools.

[0085] Observation noise of user opinion observations In this embodiment, it is assumed that it is unbiased, that is, its mean satisfies Furthermore, the observation noise for each user is uncorrelated and has the same variance. .

[0086] When opinion observations contain noise In this case, a second historical opinion reconstruction value, taking into account observation noise, can be obtained based on the historical opinion reconstruction value. The historical opinion reconstruction value can be an opinion reconstruction value calculated using existing user opinion reconstruction methods, or an opinion reconstruction value obtained using the reconstruction method provided in this embodiment; no specific limitation is made here.

[0087] Taking the historical opinion reconstruction value in equation (4) as an example, when the opinion observation value is noisy, the opinion reconstruction value in equation (4) becomes the second historical opinion reconstruction value, as shown in the following equation:

[0088] (5)

[0089] in, It is a diagonal matrix, also known as a diagonal sampling matrix. If ,but satisfy .

[0090] Furthermore, based on the reconstructed value of the second historical opinion described in equation (5), the mean square error of the reconstruction can be calculated. Specifically, the formula for calculating the mean square error is as follows:

[0091] (6)

[0092] in, , This is the autocorrelation matrix of the true opinion values. It can usually be estimated based on users' historical opinion data. For matrices The trace operation.

[0093] Based on the calculated mean square error of the reconstruction, the following opinion sampling problem is constructed to minimize the mean square error:

[0094] (7)

[0095] ;

[0096] The objective function is the reconstructed mean square error of the opinion values, and the constraint condition refers to the total number of samples. There are 10 users, and the optimization variable is the diagonal sampling matrix. It contains information about the users being sampled.

[0097] In the aforementioned minimum mean square error opinion sampling problem (7), for the optimization variable That is, the diagonal sampling matrix has constraints. as well as Therefore, this problem is essentially a combinatorial optimization problem, and all feasible solutions (diagonal sampling matrices) satisfying the conditions are: One. And for larger-scale social networks ( Directly enumerating all possible sample sets would be too probable and impractical.

[0098] In view of this, this embodiment proposes a reverse greedy sampling algorithm to solve the above-mentioned opinion sampling problem more efficiently. Specifically, Figure 2 shows a flowchart of the reverse greedy sampling algorithm provided by the present invention, which is used to further explain the reverse greedy sampling algorithm in step S110 of Figure 1.

[0099] As shown in Figure 2, the opinion sampling problem is solved using the reverse greedy sampling algorithm, resulting in a sampled user set, including:

[0100] S111, Initialize the current set of sampled users to the complete set of all users;

[0101] S112, use the objective function to calculate the mean squared error after removing each user from the current sampled user set;

[0102] S113, Determine the user to be removed corresponding to the minimum mean square error;

[0103] S114, remove the user to be removed from the current sampled user set, jump to step S112, repeat steps S112-S114 until the number of users to be removed reaches the preset threshold, and obtain the sampled user set.

[0104] The preset threshold can be set according to actual conditions and is not specifically limited here. For example, in a specific embodiment, the preset threshold is: .

[0105] Specifically, for ease of analysis, the diagonal sampling matrix will be used. Marked as the set of sampled users function ,if Then it satisfies To simplify the notation, let... Therefore, the function in the opinion sampling problem (7) becomes

[0106] (8)

[0107] in, The corresponding constraints It can be simplified to .

[0108] Minimize the function in formula (8) using the reverse greedy sampling algorithm Specifically, the sampled user set Initialize to the complete set of all users That is, sample all users, and then iteratively remove them one by one using the objective function. Each user is counted, ultimately resulting in a sampled user set.

[0109] Iteratively remove one by one using the objective function Specifically, let's assume the current sampled user set is [number of users]. ( Based on the reverse greedy approach, the objective function needs to be applied to the current sample set. Every user in , calculation will From the sample set The mean squared error of the reconstruction after removal is calculated. The objective function is as follows:

[0110] (9)

[0111] in, , For hyperparameters, , The autocorrelation matrix of the reconstructed historical opinion values. For noise variance, For the current sample set The function, For only the first A vector with one element being 1 and all other elements being 0.

[0112] According to the greedy principle, the user whose mean square error is minimized after removal and reconstruction should be selected. and will From the sampled user set Removed from the middle.

[0113] As mentioned above, for each user in the current sampled user set... If equation (9) is used to calculate the objective function value, it is necessary to solve it repeatedly. .

[0114] Therefore, in order to improve the performance of the algorithm, the following method is adopted in this embodiment.

[0115] First, maintain the matrix It is the set of sampled users. hour The corresponding inverse matrix. When selecting to remove a user. At that time, according to the Sherman-Morrison inverse formula, we can obtain:

[0116] (10)

[0117] (11)

[0118] in, It is a matrix The OK.

[0119] Based on the above relations (10) and (11), and let Then the objective function (9) can be replaced with the following expression:

[0120] (12)

[0121] As can be observed, the first term in the above equation (12) is related to all the enumerated users. It is irrelevant, therefore a proxy objective function can be considered. And select the objective function with the minimum proxy. users Remove it.

[0122] Based on the properties of trace operation, the surrogate objective function can be simplified as follows:

[0123] (13)

[0124] ;

[0125] in, It is a matrix The List, , , , It is a matrix The List, , It is a matrix The OK, This represents the noise variance.

[0126] It should be noted that the nodes to be removed are greedily selected based on the proxy objective function (13). Only the inner product of vectors needs to be calculated, which can effectively reduce computational complexity.

[0127] In this embodiment, a second historical opinion reconstruction value, considering observation noise, is determined based on the historical opinion reconstruction value. The mean square error of the reconstruction is then calculated based on this second historical opinion reconstruction value. Furthermore, based on the mean square error, an opinion sampling problem with minimum mean square error is constructed. This problem is then solved using a reverse greedy sampling algorithm to obtain the sampled user set. This method, designed for directed social network structures, determines the sampled user set using a reverse greedy sampling algorithm. The determination process only requires calculating the vector inner product, effectively reducing computational complexity, improving algorithm performance, and simultaneously enhancing the accuracy of social network opinion reconstruction.

[0128] Based on the above embodiments, further, after determining the minimum proxy objective function in each round, the parameter matrix is ​​adjusted using a preset formula. , as well as Update.

[0129] Specifically, in order to evaluate the proxy objective function in formula (13), it is necessary to maintain three parameter matrices. as well as In order to obtain vectors , as well as .

[0130] For matrix In the initial stage, there is .because It is sparse, therefore the conjugate gradient method can be used to initialize the matrix. .

[0131] For matrix and , can be based on as well as The two matrices are initialized according to the definition.

[0132] When each round selects the agent with the smallest objective function Then, the parameter matrix can be adjusted according to the preset formula. , as well as Update as follows. The preset formula is as follows:

[0133] (14)

[0134] (15)

[0135] (16)

[0136] in, The user corresponding to the minimum agent objective function.

[0137] In this embodiment, after determining the minimum agent objective function in each round, the parameter matrix is ​​adjusted using a preset formula. , as well as The system has been updated so that the default formulas use matrix-vector multiplication and column-row vector multiplication, avoiding matrix multiplication and thus effectively improving computational performance.

[0138] In another specific embodiment, the complete algorithm of the reverse greedy sampling algorithm provided by the present invention is shown as follows:

[0139] ;

[0140] In another specific embodiment, a comparison table is shown below of the opinion reconstruction errors of the prior art sampling method based on the low-pass graph signal assumption and the inverse greedy sampling algorithm (InvGreedy) provided by the present invention on four opinion datasets, respectively:

[0141] ;

[0142] As shown in the table above, the opinion reconstruction error of the reverse greedy sampling algorithm provided in this invention is reduced by 33.2%, 16%, 9.9%, and 8.8% respectively on the four datasets compared to other sampling methods. This demonstrates the effectiveness of the reverse greedy sampling algorithm and opinion reconstruction algorithm proposed in this invention.

[0143] Figure 3 shows a schematic diagram of the social network opinion reconstruction device based on graph movement smoothness provided by the present invention. As shown in Figure 3, the device includes: a sampling user set determination module 310, used to determine the sampling user set based on a pre-constructed directed social network model and according to a reverse greedy sampling algorithm; a graph movement smoothness determination module 320, used to determine the graph movement smoothness of the true opinion values ​​corresponding to all users in the directed social network model; an opinion observation value acquisition module 330, used to acquire the opinion observation values ​​of each user in the sampling user set; and a social network opinion reconstruction module 340, used to reconstruct the social network opinions of users in the directed social network model based on the graph movement smoothness and the opinion observation values.

[0144] In this embodiment, the sampling user set determination module 310 determines the sampling user set based on a pre-built directed social network model using a reverse greedy sampling algorithm. The graph movement smoothness determination module 320 determines the graph movement smoothness corresponding to all users in the directed social network model, and the opinion observation acquisition module 330 acquires the opinion observations corresponding to each user in the sampling user set. Therefore, the social network opinion reconstruction module 340 reconstructs the social network opinions of users in the directed social network model based on the graph movement smoothness and the opinion observations. This device, targeting the directed social network structure, determines the sampling user set using a reverse greedy sampling algorithm, which not only effectively improves the accuracy of social network opinion reconstruction but also significantly improves the operational efficiency of the reconstruction process.

[0145] It should be noted that the social network opinion reconstruction apparatus based on graph movement smoothness provided in this embodiment can be referred to in correspondence with the social network opinion reconstruction method based on graph movement smoothness described above, and will not be repeated here.

[0146] Figure 4 illustrates a schematic diagram of the physical structure of an electronic device. As shown in Figure 4, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440. The processor 410, communication interface 420, and memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a social network opinion reconstruction method based on graph movement smoothness. This method includes: determining a sampled user set based on a pre-built directed social network model using a reverse greedy sampling algorithm; determining the graph movement smoothness corresponding to all users in the directed social network model; obtaining opinion observations corresponding to each user in the sampled user set; and reconstructing the social network opinions of users in the directed social network model based on the graph movement smoothness and the opinion observations.

[0147] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0148] On the other hand, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a method for reconstructing social network opinions based on graph movement smoothness provided by the above methods. This method includes: determining a sampled user set based on a pre-constructed directed social network model using a reverse greedy sampling algorithm; determining the graph movement smoothness corresponding to all users in the directed social network model; obtaining opinion observations corresponding to each user in the sampled user set; and reconstructing the social network opinions of the users in the directed social network model based on the graph movement smoothness and the opinion observations.

[0149] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0150] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for reconstructing opinions in a social network based on graph movement smoothness, characterized in that, include: Based on a pre-built directed social network model, the set of sampled users is determined according to the reverse greedy sampling algorithm; The graph movement smoothness is determined for all users in the directed social network model; the opinion observations for each user in the sampled user set are obtained; based on the graph movement smoothness and the opinion observations, the social network opinions of the users in the directed social network model are reconstructed; the graph movement smoothness is defined as follows: In the formula, This represents the user's true opinion value. Represents the normalized adjacency matrix. ， Indicates excluding users All other users besides the user The sum of the effects Represents the adjacency matrix. Represents the random walk-graph Laplace matrix. The identity matrix is ​​represented; the opinion reconstruction value of the social network opinions is obtained by the following formula: ;in, For the reconstruction value of opinions, For the sampling matrix, For hyperparameters, These are observations of opinions.

2. The social network opinion reconstruction method based on graph movement smoothness according to claim 1, characterized in that, The method for determining the sampled user set based on a pre-built directed social network model and a reverse greedy sampling algorithm includes: determining a second historical opinion reconstruction value that takes into account observation noise based on the historical opinion reconstruction value; calculating the mean square error of the reconstruction based on the second historical opinion reconstruction value; constructing an opinion sampling problem with minimum mean square error based on the mean square error; and solving the opinion sampling problem using the reverse greedy sampling algorithm to obtain the sampled user set.

3. The social network opinion reconstruction method based on graph movement smoothness according to claim 2, characterized in that, The step of solving the opinion sampling problem using the reverse greedy sampling algorithm to obtain the sampled user set includes: S111, initializing the current sampled user set as the complete set of all users; S112, calculating the mean squared error after removing each user from the current sampled user set using an objective function; S113, determining the user to be removed corresponding to the minimum mean squared error; S114, removing the user to be removed from the current sampled user set, jumping to step S112, and repeating steps S112-S114 until the number of users to be removed reaches a preset threshold, thereby obtaining the sampled user set.

4. The social network opinion reconstruction method based on graph movement smoothness according to claim 3, characterized in that, The objective function is as follows: ;in, This represents the trace operation on a matrix. ， For hyperparameters, ， The autocorrelation matrix of the reconstructed historical opinion values. For noise variance, For the current sample set The function, For only the first A vector with one element being 1 and all other elements being 0.

5. The social network opinion reconstruction method based on graph movement smoothness according to claim 4, characterized in that, The user to be removed is determined by a proxy objective function, which is as follows: ； ;in, It is a matrix The List, ， ， ， It is a matrix The List, ， It is a matrix The OK, This represents the noise variance.

6. The social network opinion reconstruction method based on graph movement smoothness according to claim 5, characterized in that, After determining the minimum agent objective function in each round, the parameter matrix is ​​adjusted using a pre-defined formula. 、 as well as The preset formula is updated as follows: ； ； ;in, The user corresponding to the minimum agent objective function.

7. A social network opinion reconstruction device based on graph movement smoothness, characterized in that, include: The sampled user set determination module is used to determine the sampled user set based on a pre-built directed social network model and a reverse greedy sampling algorithm. The graph movement smoothness determination module is used to determine the graph movement smoothness of the true opinion values ​​corresponding to all users in the directed social network model; the opinion observation value acquisition module is used to acquire the opinion observation values ​​of each user in the sampled user set; the social network opinion reconstruction module is used to reconstruct the social network opinions of users in the directed social network model based on the graph movement smoothness and the opinion observation values; the graph movement smoothness is defined as follows: In the formula, This represents the user's true opinion value. Represents the normalized adjacency matrix. ， Indicates excluding users All other users besides the user The sum of the effects Represents the adjacency matrix. Represents the random walk-graph Laplace matrix. The identity matrix is ​​represented; the opinion reconstruction value of the social network opinions is obtained by the following formula: ;in, For the reconstruction value of opinions, For the sampling matrix, For hyperparameters, These are observations of opinions.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the social network opinion reconstruction method based on graph movement smoothness as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the social network opinion reconstruction method based on graph movement smoothness as described in any one of claims 1 to 6.

Images