Website traffic prediction method and system based on attribute and topology independence constraints

Abstract

The application is a website traffic prediction method and system based on topology and attribute independence constraint. The method obtains historical traffic data of a website and converts it into website traffic graph data. The nodes of the website traffic graph data represent a webpage in the website, and the edges represent the link relationship between the webpages. The node attribute information and graph topology information are obtained from the website traffic graph data. Then, a target function of the website traffic prediction model is constructed based on a graph convolutional neural network. The Hilbert-Schmidt independence criterion or vector 1-norm is used as the independence constraint condition between the attributes and the topology. The optimal solution of the target function is obtained. The iterative formula of the attribute representation and the topology representation is obtained. The converged attribute representation and the topology representation are spliced together by rows to obtain the output of the model. The independence constraint is performed on the attribute and topology information. The mutual interference between the attribute and the topology information is avoided when they are fused. At the same time, the low-frequency and high-frequency information in the website traffic graph data is fully extracted.

Description

Technical Field

[0001] This invention belongs to the field of website traffic prediction technology, specifically a website traffic prediction method and system based on attribute and topology independence constraints. Background Technology

[0002] With the booming development of the internet industry, more and more users are learning and collecting information through the internet. Website traffic reflects the number of views a website receives over a period of time and is one of the important indicators in online marketing research, playing a crucial role for both users and website administrators. Website traffic data, as a new type of data in the internet age, can be abstracted as graph data composed of nodes and edges. Web pages on a website correspond to nodes in the graph data, and node attributes consist of the keywords and traffic associated with the web page. The links between web pages correspond to edges in the graph data; that is, if there is a link between web pages, there is an edge connecting the two nodes. By inputting the abstracted graph data of website traffic into a graph neural network, website traffic can be predicted. Website traffic prediction helps companies obtain important information such as website development scale, user retention, and marketing effectiveness in a timely manner, allowing for faster adjustments to strategic direction. It also helps companies control risks, promptly identify and investigate anomalies, and avoid unnecessary losses. Therefore, website traffic prediction can help companies cope with increasingly fierce industry competition, and its accuracy is directly related to the efficiency and performance of enterprise management.

[0003] Graph neural networks (GNNs) have demonstrated superior performance in processing non-Euclidean data and have been widely applied in fields ranging from computer vision to natural language processing. Essentially, classic GNNs learn node representations by aggregating information from the neighboring nodes around a central node. This results in low-pass filtering and smoothing properties, effectively extracting low-frequency information—the similarities between a node's attributes and those of its neighbors—but struggling to extract high-frequency information—the differences between a node's attributes and those of its neighbors. However, existing research has demonstrated the crucial role of high-frequency information in graphs. For website traffic data, both low-frequency and high-frequency information play a significant role, thus classic GNNs based on smoothing operations fail to achieve the desired results in website traffic prediction. Furthermore, the fusion of attribute and topological information during message propagation in existing classic GNN models leads to mutual interference between these two information types, such as oversmoothing. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the technical problem this invention aims to solve is to propose a website traffic prediction method and system based on topology and attribute independence constraints.

[0005] The technical solution adopted by the present invention to solve the aforementioned technical problem is as follows: On the one hand, this invention provides a website traffic prediction method based on topology and attribute independence constraints, comprising the following steps: Step 1: Obtain the website's historical traffic data and convert it into website traffic graph data. Nodes in the website traffic graph data represent a webpage on the website, and edges represent the links between webpages. Extract node attribute information and graph topology information from the website traffic graph data. Node attribute information includes the keywords and traffic of the webpage. Step 2: Construct a website traffic prediction model based on a graph convolutional neural network. Use node attribute information and graph topology information as input to the website traffic prediction model. The output of the website traffic prediction model is compressed into a one-dimensional column vector formed by vertically stacking the prediction labels of each node through a multilayer perceptron, which is the traffic prediction result of each webpage. Equation (1) is the objective function corresponding to the website traffic prediction model, and the optimal solution of the objective function is the website traffic prediction model. O = min (1) In the formula, This represents the attribute representation extracted from node attribute information by the website traffic prediction model. Represents node attribute information, This represents the topological representation extracted from the graph topology information by the website traffic prediction model. Describing the Frobenius norm, This represents the Laplace matrix corresponding to the website traffic graph data. Represents graph topology information. express The degree matrix, Represents the feature fitting term, This represents the graph regularization term. Represents the trace of a matrix. Indicates matrix transpose. This represents the balance factor between the feature fitting term and the graph regularization term; This is a constraint term for the independence between attribute representation and topological representation. The function representing the independence constraint between attribute representation and topological representation. Indicates learnable parameters; Using the Hilbert-Schmidt independence criterion or the vector 1-norm as independence constraints, the independence constraint term between the attribute representation and the topological representation in the objective function is specifically represented; when the Hilbert-Schmidt independence criterion is used as the independence constraint, the objective function is expressed as: O = min (2) (3) In the formula, The function representing the independence constraint between attribute representation and topological representation. , Represents a matrix of all 1s. This indicates the number of nodes in the website traffic graph data. and Both represent kernel functions; make ,and Then the independence constraint function between attribute representation and topological representation in the objective function is: (4) Ignoring the constants in the independence constraint function, the objective function is expressed as: O = min (5) For the two variables in equation (5) and Taking the derivatives separately and setting them to zero, the iterative formulas for attribute representation and topological representation are expressed as follows: (6) (7) In the formula, Indicates the number of iterations; A learnable weight matrix is ​​introduced during the iteration process. Characterized by balance properties and topological representation To address the issue of inconsistent data scales, the iterative formulas for attribute representation and topological representation are as follows: (8) (9) When the 1-norm of a vector is used as the independence constraint, the objective function is expressed as: O = min (10) In the formula, Describes the 1-norm of a vector. This represents the corresponding operation in the soft threshold function; Solving the objective function of equation (10) using the gradient descent method, the iterative formulas for attribute representation and topological representation are as follows: (11) (12) In the formula, , All represent the learning rate. , Both represent soft threshold functions. , , The function is defined as follows: (13) In the formula, The independent variable of the function is... This indicates the set threshold.

[0006] The attribute representation and topological representation are iteratively updated according to equations (8) and (9) or equations (11) and (12) until convergence; then the converged attribute representation and topological representation are concatenated row by row to obtain the output of the website traffic prediction model. ; (14) In the formula, Indicates a splicing operation; The output of the website traffic prediction model is compressed into a one-dimensional column vector formed by vertically stacking the prediction labels of each node through a multilayer perceptron, which is the traffic prediction result of each webpage. Step 3: Randomly select some nodes or edges from the website traffic graph data to train the website traffic prediction model, and obtain the trained website traffic prediction model. When used for traffic prediction, convert the historical traffic data of the website to be predicted into website traffic graph data, and obtain the node attribute information and graph topology information of the website traffic graph data. Then, input the node attribute information and graph topology information into the trained website traffic prediction model to predict the traffic of each webpage of the website to be predicted.

[0007] Furthermore, this method is applied to the link prediction task, that is, to predict the link relationship between web pages. In step 3, some edges in the website traffic graph data are randomly deleted, and then the node attribute information and graph topology information are input into the website traffic prediction model. According to equation (16), the output of the website traffic prediction model is multiplied by its transpose matrix to obtain the output result of the link prediction task. (16) On the other hand, the present invention also provides a website traffic prediction system, including a processor, a memory, and a computer program; the processor is connected to the memory, and the computer program is stored in the memory. When the system is running, the processor executes the computer program in the memory, causing the system to perform the above-described method.

[0008] A computer-readable storage medium for storing computer instructions; when executed by a processor, the computer instructions perform the above-described method.

[0009] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention targets website traffic data, a novel type of data, by converting it into graph data and using a graph neural network for website traffic prediction. This helps businesses obtain important information such as user retention and marketing effectiveness. Addressing the low-pass filtering characteristics of classic graph neural networks, this invention improves upon them, establishing a website traffic prediction model. Independence constraints are introduced into the model to independently constrain attributes and topological information during message propagation, effectively avoiding mutual interference between these two types of information when fusing them. This allows for the full extraction of both low-frequency and high-frequency information from the website traffic graph data, making it more suitable for website traffic prediction and improving model performance.

[0010] 2. This invention addresses website traffic prediction model from a numerical optimization perspective, explaining and optimizing the model to make it more interpretable, universal, and scalable.

[0011] 3. The website traffic prediction model of this invention is also applicable to link prediction tasks, and its test results are generally better than traditional network embedding methods. This proves that for such tasks, attribute information is not unimportant, but rather it is necessary to avoid mutual interference between attribute and topology information while integrating them, so as to prevent the impact on model performance. Attached Figure Description

[0012] Picture 1 The accompanying drawings are for reference only. Detailed Implementation

[0013] The specific embodiments of the present invention will be explained in detail below with reference to the accompanying drawings. These specific embodiments are only used to further illustrate the present invention and do not limit the scope of protection claimed in this application.

[0014] This invention relates to a website traffic prediction method and system based on attribute and topology independence constraints (hereinafter referred to as the method, see below). Picture 1 The specific steps are as follows: Step 1: Obtain historical traffic data from the website and convert it into website traffic graph data. Nodes in the website traffic graph data represent a webpage on the website, and edges represent links between webpages. Extract node attribute information from the website traffic graph data. Graph topology information ; Historical website traffic data refers to traffic within a specific time period; assuming the website traffic graph data is... , Represents a set of nodes. Represents the set of edges; node attribute information , Representing the matrix space, Indicates the number of nodes. The attribute dimensions of nodes are represented; edge information reflects the graph's topological information, using an adjacency matrix. This indicates that if the node and If there is an edge connecting the nodes, then the nodes and The elements in the corresponding adjacency matrix ,on the contrary ;remember Adjacency matrix The corresponding degree matrix is ​​a diagonal matrix, and the diagonal elements correspond to the row sums of the adjacency matrix; the Laplacian matrix of the website traffic graph data is defined as follows. Its calculation formula is ; Step 2: Construct a website traffic prediction model based on graph convolutional neural networks, incorporating node attribute information. Graph topology information As input to the website traffic prediction model, the output of the website traffic prediction model is the representation of each node; then, the representation of each node is compressed into a single representation using a multilayer perceptron. The column vector is used to obtain the traffic prediction results for each webpage, i.e., the website traffic prediction results; Equation (1) is the objective function of the website traffic prediction model, and the optimal solution of the objective function is the website traffic prediction model. O = min (1) In the formula, This indicates that the website traffic prediction model is based on node attribute information. The attribute representation extracted from it. This indicates that the website traffic prediction model is based on graph topology information. The topological representation extracted from it. Denotes the Frobenius norm; Represents the feature fitting term, used for constrained attribute representation. Compared with the original node attribute information They should be as similar as possible; The graph regularization term is used to measure the overall smoothness of website traffic graph data, making the node representations extracted between two nodes that are connected by an edge as similar as possible. Represents the trace of a matrix. Indicates matrix transpose. This represents the balance factor between the feature fitting term and the graph regularization term; Attribute representation With topological representation Independence constraints between them Representation of attribute characterization With topological representation Independence constraint functions between them are used to characterize constraint attributes. and topological representation As inconsistent as possible Indicates learnable parameters; Using the Hilbert-Schmidt Independence Criterion (HSIC) or the vector 1-norm as independence constraints, the independence constraint terms between attribute representations and topological representations in the objective function are specified, and the above objective function is solved. The Hilbert-Schmidt independence criterion (HSIC) relies on the covariance of random variables for construction. Its main purpose is to measure the independence between two variables, and it contains an important property: if the random variables... and If they are mutually independent, then The Hilbert-Schmidt independence criterion is used to express the independence constraint function between attribute representations and topological representations. Then the objective function is expressed as: O = min (2) (3) In the formula, , Represents a matrix of all 1s. This indicates the number of nodes in the website traffic graph data. and Both represent kernel functions; To simplify the calculation, let ,and Then the independence constraint function between attribute representation and topological representation in the objective function is: (4) To further simplify the calculation, ignoring the constants in the independence constraint function, the objective function is expressed as: O = min (5) For the two variables in equation (5) and Taking the derivatives separately and setting them to zero, the iterative formulas for attribute representation and topological representation are expressed as follows: (6) (7) In the formula, Indicates the number of iterations, attribute representation The initial value is the node attribute information. ; The meanings of the two iterative formulas are as follows: removing topological information during the iterative process of attribute representation, and removing attribute information during the iterative process of topological representation, that is, removing mutual interference factors in the propagation process. A learnable weight matrix is ​​introduced during the iteration process. Characterized by balance properties and topological representation To address the issue of inconsistent data scales, the iterative formulas for attribute representation and topological representation are as follows: (8) (9) The 1-norm of a vector is the sum of the absolute values ​​of all its elements. Also known as the sparsity operator, when used as a regularization term, it sparsifies the model parameters, maximizing the number of zero elements. The 1-norm can be used as an attribute representation. and topological representation When the independence constraint is applied between them, the objective function can be expressed as: O = min (10) In the formula, Describes the 1-norm of a vector. This represents the corresponding operation in the soft threshold function; Solving the objective function of equation (10) using the gradient descent method, the iterative formulas for attribute representation and topological representation are as follows: (11) (12) In the formula, , All represent the learning rate. , Both represent soft thresholding functions, where , , The function is defined as follows: (13) In the formula, The independent variable of the function is... This indicates the set threshold. Characterize the attributes according to equations (8) and (9) or equations (11) and (12). and topological representation Perform iterative updates until the attribute representation is complete. and topological representation Convergence; then characterize the converged attributes. With topological representation By concatenating the lines together, we obtain the output of the website traffic prediction model. That is, a matrix composed of the representations of each node, its expression is: (14) In the formula, Indicates a splicing operation; The output of the website traffic prediction model Compressed using a multilayer perceptron , The predicted labels of each node are stacked vertically to form a one-dimensional column vector, which represents the traffic prediction results of each webpage. Step 3: Use the website traffic graph data from Step 1 to train the website traffic prediction model and obtain the trained website traffic prediction model. When used for traffic prediction, convert the historical traffic data of the website to be predicted into website traffic graph data, and obtain the node attribute information and graph topology information of the website traffic graph data. Then input the node attribute information and graph topology information into the trained website traffic prediction model to predict the traffic of each webpage of the website to be predicted. The website traffic prediction model was trained using a semi-supervised learning approach, and website traffic graph data was randomly selected. Training is performed on a subset of nodes or edges in the dataset. 60% of the nodes corresponding to each label are used as the training set, 20% as the validation set, and 20% as the test set. The training loss is calculated according to the loss function in equation (15). (15) in, This indicates a range of webpage traffic, such as 10-20 times / month. Indicates the largest flow range. This indicates that the webpage will be displayed. Traffic forecast is The probability of; Represents a webpage The true label, that is, its true traffic data.

[0015] The website traffic prediction model of this invention can also be used to handle link prediction tasks, that is, to predict the link relationships between web pages, which is to predict whether there is an edge between two nodes in the website traffic graph data. A semi-supervised training method is used, randomly deleting some edges from the website traffic graph data, and then inputting node attribute information and graph topology information into the website traffic prediction model to obtain the output of the website traffic prediction model. The output of the website traffic prediction model is calculated according to equation (16). Multiplying it by its transpose matrix yields the output of the link prediction task. ; (16) During the training process, the training loss is calculated according to the loss function in equation (17); (17) In the formula, Expressing expectations, This means that the original adjacency matrix of the website traffic graph data is derived by back-calculating the known model predictions. The probability, This indicates the model output predicted when the attributes and topological information are known. The probability, The output obtained by the model prediction is The probability, Represents probability and The relative entropy between them.

[0016] This invention also provides a website traffic prediction system, including a processor, a memory, and a computer program. The processor is connected to the memory, and the computer program is stored in the memory. When the system runs, the processor executes the computer program in the memory, causing the system to perform the aforementioned method. The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), programmable gate arrays (FPGAs), etc. The memory can be a read-only memory (ROM) or a random access memory (RAM), providing instructions and data to the processor. In implementation, each step of the aforementioned method can be completed through integrated logic circuits in the processor or through software instructions.

[0017] The present invention also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the above-described method.

[0018] Simulation experiment: Website traffic prediction is essentially a node classification task, with the prediction result being the interval corresponding to the webpage traffic. To verify the effectiveness of the website traffic prediction model of this invention, node classification tasks were performed on seven real datasets using both the website traffic prediction model of this invention and common graph neural network models in the prior art. The website traffic prediction model of this invention uses the Hilbert-Schmidt independence criterion as the independence constraint term between attributes and topological information.

[0019] Table 1 shows the accuracy statistics of each model on seven real datasets for node classification tasks. The homonym of the datasets is calculated by the proportion of connected nodes with the same label among all nodes. A higher homonym indicates that low-frequency information in the graph data is more important, while a lower homonym (i.e., a higher dissimilarity) indicates that high-frequency information in the graph data is more important. Among them, Cora, Pubmed, and Citeseer are three commonly used citation networks with high homonyms. Nodes represent papers, node attributes represent the keywords corresponding to the papers, and edges between nodes represent the citation relationships between papers. This type of dataset describes that papers of the same or similar categories tend to have citation relationships. The Chameleon and Squirrel datasets describe the relationships between entries in Wikipedia. Nodes represent entries, and node attributes are the keywords corresponding to the entries and the average monthly pageviews. From the distribution of real data, this type of dataset generally has high dissimilarity, so high-frequency information is also important in this type of dataset. The Computer and Photo datasets describe the co-purchase relationships of products on the Amazon website. Nodes represent products, node attributes are product features, and edges between nodes represent that two types of products are frequently purchased together.

[0020] Table 1. Accuracy statistics of each model on real datasets for node classification tasks. ; Under the same dataset partitioning (using 60%, 20%, and 20% of the data as the training, validation, and test sets respectively), the website traffic prediction model of this invention was compared with six existing classic graph neural network models and four graph neural network models designed for heterogeneous graphs. Node classification tasks were performed on each model, and the accuracy rates are shown in Table 1. Table 1 shows that the website traffic prediction model of this invention achieves good results on homogeneous datasets compared to existing graph neural network models. For network traffic datasets with high heterogeneity, such as Chameleon and Squirrel, the accuracy is significantly improved. This is because the website traffic prediction model of this invention introduces an independence constraint term between attribute and topology information. Specifically, classic graph neural networks are based on attribute and topology fusion during message propagation. This constraint precisely satisfies the characteristic that graph data with high homogeneity has consistency between its attributes and topology information. Therefore, classic graph neural networks have high accuracy for graph data with high homogeneity, but they are not suitable for heterogeneous graph data where attributes and topology information are inconsistent. The website traffic prediction model of this invention can effectively solve this problem.

[0021] Under the same experimental conditions, the website traffic prediction model of the present invention and five common network models in the prior art were used to process link prediction tasks, and the statistical results shown in Table 2 were obtained. Table 2. Accuracy statistics of each model for link prediction tasks ; For each dataset, 10% of the edges were randomly removed as test data, and the remaining 90% of the edges, along with the same number of additionally sampled edges that were not present, were used to construct the training data. Experimental results show that the website traffic prediction model of this invention achieves better results on several datasets compared to network embedding methods and graph neural network-based methods. This demonstrates that the website traffic prediction model of this invention can effectively avoid the interference of attribute information on topological information during the combination of topological and attribute information.

[0022] The sources of each model are: [1] Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. In ICLR, 2017. [2] PetarVelickovic,GuillemCucurull,ArantxaCasanova,AdrianaRomero,Pietro Liò, and Yoshua Bengio. Graph attention networks. In ICLR, 2018. [3] William L. Hamilton, Zhitao Ying, and Jure Leskovec. Inductiverepresentation learning on large graphs. In NIPS, pages 1024–1034, 2017. [4] Ming Chen, Zhewei Wei, Zengfeng Huang, Bolin Ding, and YaliangLi. Simple and deep graph convolutional networks. In ICML, pages 1725–1735,2020. [5] Johannes Klicpera, Aleksandar Bojchevski, and Stephan Günnemann.Predict then propagate: Graph neural networks meet personalized pagerank. InICLR, 2019. [6] Keyulu Xu, Chengtao Li, Yonglong Tian, Tomohiro Sonobe, Ken-ichiKawarabayashi, and Stefanie Jegelka. Representation learning on graphs withjumping knowledge networks. In ICML, pages 5449–5458, 2018. [7] Hongbin Pei, Bingzhe Wei, Kevin Chen-Chuan Chang, Yu Lei, and BoYang. Geom-gcn: Geometric graph convolutional networks. In ICLR, 2020. [8] Eli Chien, Jianhao Peng, Pan Li, and Olgica Milenkovic. Adaptiveuniversal generalized pagerank graph neural network. In ICLR, 2021. [9] Deyu Bo, Xiao Wang, Chuan Shi, and Huawei Shen. Beyond low-frequency information in graph convolutional networks. pages 3950–3957,

[10] Jiong Zhu, Yujun Yan, Lingxiao Zhao, Mark Heimann, Leman Akoglu, and Danai Koutra. Beyond homophily in graph neural networks: Current limitations and effective designs. In NeurIPS, 2020.

[11] Aditya Grover and Jure Leskovec. node2vec: Scalable feature learning for networks. In SIGKDD, pages 855–864,

[12] Jian Tang, Meng Qu, Mingzhe Wang, Ming Zhang, Jun Yan, andQiaozhu Mei.LINE: Large-Scale Information Network Embedding. In WWW , pages1067–1077,2015.

[13] Thomas N. Kipf and Max Welling. Variational graph auto-encoders.CoRR,abs / 1611.07308, Thanks for reading the book.

Claims

1. A website traffic prediction method based on topology and attribute independence constraints, characterized in that, Includes the following steps: Step 1: Obtain the website's historical traffic data and convert it into website traffic graph data. Nodes in the website traffic graph data represent a webpage on the website, and edges represent the links between webpages. Extract node attribute information and graph topology information from the website traffic graph data. Node attribute information includes the keywords and traffic of the webpage. Step 2: Construct a website traffic prediction model based on a graph convolutional neural network. Use node attribute information and graph topology information as input to the website traffic prediction model. The output of the website traffic prediction model is compressed into a one-dimensional column vector formed by vertically stacking the prediction labels of each node through a multilayer perceptron, which is the traffic prediction result of each webpage. Equation (1) is the objective function of the website traffic prediction model, and the optimal solution of the objective function is the website traffic prediction model. O = min (1) In the formula, This represents the attribute representation extracted from node attribute information by the website traffic prediction model. Represents node attribute information, This represents the topological representation extracted from the graph topology information by the website traffic prediction model. Denotes the Frobenius norm. The Laplace matrix represents the website traffic graph data. Represents graph topology information. express The degree matrix, Represents the feature fitting term, This represents the graph regularization term. Represents the trace of a matrix. Indicates matrix transpose. This represents the balance factor between the feature fitting term and the graph regularization term; This is a constraint term for the independence between attribute representation and topological representation. The function representing the independence constraint between attribute representation and topological representation. Indicates learnable parameters; Using the 1-norm of a vector as an independence constraint, the independence constraint term between the attribute representation and the topological representation in the objective function is given a more concrete representation; the objective function is expressed as: O = min (10) In the formula, Describes the 1-norm of a vector. This represents the corresponding operation in the soft threshold function; Solving the objective function of equation (10) using the gradient descent method, the iterative formulas for attribute representation and topological representation are as follows: (11) (12) In the formula, , The first , The attribute representation after the next iteration. , The first , Topological representation after the next iteration. , All represent the learning rate. , Both represent soft threshold functions. , , The function is defined as follows: (13) In the formula, Represents the independent variable of the function; This indicates the set threshold. The attribute representation and topology representation are iteratively updated according to equations (11) and (12) until convergence; then the converged attribute representation and topology representation are concatenated row by row to obtain the output of the website traffic prediction model. ; (14) In the formula, Indicates a splicing operation; The output of the website traffic prediction model is compressed using a multilayer perceptron into a vertically stacked representation of the prediction labels of each node. A column vector, representing the traffic prediction results for each webpage; Step 3: Randomly select some nodes or edges from the website traffic graph data to train the website traffic prediction model, and obtain the trained website traffic prediction model. When used for traffic prediction, convert the historical traffic data of the website to be predicted into website traffic graph data, and obtain the node attribute information and graph topology information of the website traffic graph data. Then, input the node attribute information and graph topology information into the trained website traffic prediction model to predict the traffic of each webpage of the website to be predicted.

2. The website traffic prediction method based on topology and attribute independence constraints according to claim 1, characterized in that, When this method is used for link prediction tasks, i.e. predicting the link relationships between web pages, in step 3, some edges in the website traffic graph data are randomly deleted, and then the node attribute information and graph topology information are input into the website traffic prediction model; according to equation (16), the output of the website traffic prediction model is multiplied by its transpose matrix to obtain the output result of the link prediction task. (16) In the formula, This indicates the output of the link prediction task.

3. A website traffic prediction system, characterized in that, It includes a processor, a memory, and a computer program; the processor is connected to the memory, and the computer program is stored in the memory. When the system is running, the processor executes the computer program in the memory, causing the system to perform the method described in any one of claims 1 to 2.

4. A computer-readable storage medium for storing computer instructions; characterized in that, When the computer instructions are executed by the processor, they complete the method described in any one of claims 1 to 2.

Images