Smart city service platform, method and equipment based on cloud computing and big data
Through a smart city service platform based on cloud computing and big data, similarity sorting and dynamic virtual graph update technology are used to solve the problem of cross-provincial or cross-operator call number identification, and the accuracy and effectiveness of telecom fraud identification are improved.
Patent Information
- Application Number
- CN202411242657.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-05
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-09-05
AI Technical Summary
The prior art cannot effectively identify fraud-related numbers for cross-provincial or cross-operator calls, resulting in a reduced accuracy of model identification. The nodes in the figure may appear as isolated nodes or missing connections.
A smart city service platform based on cloud computing and big data is adopted to determine adjacent node sets through similarity sorting, update the dynamic virtual graph, and input it into the telecom fraud identification model, and use graph convolution neural network to identify telecom fraud behavior.
It enhances the connectivity of the graph, improves the accuracy and effectiveness of telecom fraud identification, solves the problem of cross-provincial or cross-operator call number recognition, and improves the recognition accuracy of the model.
Smart Images

Figure CN118761888B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology, and in particular to a smart city service platform, method and device based on cloud computing and big data. Background Art
[0002] With the rapid development of communication technology, telecommunications fraud, as a new type of crime, is highly concealed and cross-regional, posing a serious threat to the property safety of the masses. Telecommunications fraud usually uses fabricated false information to deceive victims through channels such as telephones and communication software, and conducts remote, contactless fraud. In order to effectively identify and prevent telecommunications fraud, traditional detection methods mainly rely on expert experience and simple machine learning algorithms. Although these methods are easy to operate, they have limited accuracy in the model.
[0003] To overcome this problem, graph convolutional neural network (GCN) is an emerging technology that uses graph models to describe the relationship between users, which can better capture user characteristics and thus improve the recognition accuracy of the model. However, although graph convolutional neural networks have shown potential in telecommunications fraud detection, existing technologies still face some challenges in practical applications. Telecommunications user data is usually owned by different operators and companies in different provinces, resulting in data islands, so that each company can only model based on the data of the province and the operator. For those fraudulent numbers that only make calls across provinces or operators, they may appear as isolated nodes or missing connections in the graph, resulting in the model being unable to identify cross-provincial or cross-operator user behavior, which in turn affects the recognition accuracy of the model. Summary of the invention
[0004] The present invention provides a smart city service platform, method and device based on cloud computing and big data, so as to solve the defect that the prior art cannot identify fraudulent numbers that only make cross-provincial or cross-operator calls, thus affecting the accuracy of model recognition.
[0005] In a first aspect, the present invention provides a smart city service platform based on cloud computing and big data, the smart city service platform based on cloud computing and big data includes a control center and at least one smart city service system, the smart city service system includes at least a telecommunications fraud identification subsystem; the telecommunications fraud identification subsystem includes an adjacent node processing module, a virtual map update module and a telecommunications fraud identification module;
[0006] The adjacent node processing module is used to determine similarity results between the target node and multiple user nodes, perform similarity sorting on each of the similarity results, and determine an adjacent node set based on the sorting results, wherein the adjacent node set includes a plurality of user nodes adjacent to the target node;
[0007] The virtual graph updating module is used to update the dynamic virtual graph based on the target node and the adjacent node set;
[0008] The telecommunications fraud identification module is used to input the dynamic virtual graph into the telecommunications fraud identification model to obtain the telecommunications fraud identification result output by the telecommunications fraud identification model; wherein the telecommunications fraud identification model is trained based on the sample dynamic virtual graph and the telecommunications fraud identification result label corresponding to the sample dynamic virtual graph.
[0009] In a second aspect, the present invention further provides a smart city service method based on cloud computing and big data, which is applied to the smart city service platform based on cloud computing and big data described in the first aspect, wherein the smart city service platform includes a control center and at least one smart city service system, wherein the smart city service system includes at least a telecommunications fraud identification subsystem, and the smart city service method includes:
[0010] Determine similarity results between a target node and a plurality of user nodes, perform similarity sorting on each of the similarity results, and determine an adjacent node set based on the sorting results, wherein the adjacent node set includes a plurality of user nodes adjacent to the target node;
[0011] Based on the target node and the set of adjacent nodes, updating a dynamic virtual graph;
[0012] Inputting the dynamic virtual graph into a telecommunications fraud identification model to obtain a telecommunications fraud identification result output by the telecommunications fraud identification model;
[0013] Among them, the telecommunications fraud identification model is trained based on the sample dynamic virtual graph and the telecommunications fraud identification result labels corresponding to the sample dynamic virtual graph.
[0014] According to the smart city service method based on cloud computing and big data provided by the present invention, the step of determining the adjacent node set based on the sorting result includes:
[0015] Clustering the feature vector of the target node and the feature vectors of multiple user nodes to determine a target node cluster, where the target node cluster includes the target node and a plurality of user nodes having similar features to the target node;
[0016] Sampling a number of user nodes in the target node cluster that have similar features to the target node to obtain a sampled node set;
[0017] Based on the sorting result and the sampled node set, an adjacent node set is determined.
[0018] According to the smart city service method based on cloud computing and big data provided by the present invention, the similarity results between the target node and the multiple user nodes are determined, including:
[0019] Based on a first node feature of a target node and first node features of a plurality of user nodes, a difference between the target node and each user node on the same feature dimension is calculated, wherein the first node feature includes a plurality of feature dimensions;
[0020] Based on the first node feature of the target node and the first node features of multiple user nodes, the maximum difference values corresponding to the multiple feature dimensions are calculated, and the maximum difference value corresponding to any of the feature dimensions is the difference between the maximum value in the feature dimension of each of the first node features and the minimum value in the feature dimension of each of the first node features;
[0021] Based on the difference between the target node and each user node on the same feature dimension and the difference between the maximum value and the minimum value of the same feature dimension in all nodes, calculate the second node feature of the target node and the second node features of multiple user nodes;
[0022] The similarity between the second node feature of the target node and the second node features of each of the user nodes is calculated, and the similarity results between the target node and the multiple user nodes are determined according to the calculation results.
[0023] According to the smart city service method based on cloud computing and big data provided by the present invention, the calculating the similarity between the second node feature of the target node and the second node feature of each of the user nodes, and determining the similarity results between the target node and the multiple user nodes according to the calculation results, includes:
[0024] Calculating the Euclidean distance between the second node feature of the target node and the second node feature of each of the user nodes to obtain a calculation result set;
[0025] The similarity results between the target node and the multiple user nodes are determined according to the calculation result set.
[0026] According to the smart city service method based on cloud computing and big data provided by the present invention, the dynamic virtual map is input into the telecommunications fraud identification model to obtain the telecommunications fraud identification result output by the telecommunications fraud identification model, including:
[0027] Inputting the dynamic virtual graph into a feature extraction layer in a telecommunications fraud identification model to obtain an initial feature matrix output by the feature extraction layer, wherein the initial feature matrix includes feature representations of all nodes in the dynamic virtual graph;
[0028] Input the initial feature matrix into the attention calculation layer in the telecommunications fraud identification model to obtain the attention coefficient set output by the attention calculation layer, wherein the attention calculation layer is used to calculate the attention coefficient of the feature representation of each adjacent node in the adjacent node set in the initial feature matrix to the feature representation of the target node;
[0029] Input the attention coefficient set and the initial feature matrix into the feature aggregation layer in the telecommunications fraud identification model to obtain the feature matrix output by the feature aggregation layer, wherein the feature aggregation layer is used to aggregate the feature representation of the target node in the initial feature matrix with the feature representation of each adjacent node in the adjacent node set to obtain the feature matrix;
[0030] The feature matrix is input into the feature generalization layer in the telecommunications fraud identification model to obtain a generalized feature matrix output by the feature generalization layer, wherein the feature generalization layer is used to perform feature extraction and dimension change processing on the feature matrix to determine a generalized feature matrix;
[0031] The generalized feature matrix is input into the probability calculation layer in the telecommunications fraud identification model to obtain the telecommunications fraud identification result output by the probability calculation layer. The probability calculation layer is used to calculate the fraud probability and non-fraud probability of the target node in the generalized feature matrix to determine the telecommunications fraud identification result.
[0032] According to the smart city service method based on cloud computing and big data provided by the present invention, the attention coefficient set is obtained based on the following method:
[0033] Performing linear transformation on the feature representation of the target node in the initial feature matrix and the feature representation of each adjacent node in the adjacent node set, and then concatenating them to determine a first concatenated feature vector set, wherein the first concatenated feature vector set includes a first concatenated feature vector obtained by concatenating the feature representation of the target node and the feature representation of each adjacent node in the adjacent node set after linear transformation;
[0034] Performing vector transposition on each first splicing feature vector in the first splicing feature vector set to obtain a second splicing vector set;
[0035] Based on the activation function and the exponential function, each second splicing vector in the second splicing vector set is calculated to obtain an attention score set, wherein the attention score set includes an attention score of each adjacent node in the adjacent node set to the target node, and each of the attention scores is a positive number;
[0036] Accumulating the attention scores in the attention score set to obtain a total attention score;
[0037] Based on the attention score set and the total of the attention scores, an attention coefficient set is determined, where the attention coefficient set includes the attention coefficient of each adjacent node in the adjacent node set to the target node.
[0038] In a third aspect, the present invention also provides an electronic device, comprising: a memory for storing a computer software program; a processor for reading and executing the computer software program, thereby realizing a smart city service platform based on cloud computing and big data as described in any one of the above.
[0039] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium, in which a computer software program is stored. When the computer software program is executed by a processor, it implements a smart city service platform based on cloud computing and big data as described in any of the above.
[0040] In a fifth aspect, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements a smart city service platform based on cloud computing and big data as described in any one of the above.
[0041] The smart city service platform based on cloud computing and big data provided by the embodiment of the present invention includes a control center and a smart city service system, the smart city service system includes a telecommunications fraud identification subsystem; the telecommunications fraud identification subsystem includes an adjacent node processing module, a virtual graph update module and a telecommunications fraud identification module; the adjacent node processing module determines the similarity results between the target node and multiple user nodes, sorts the similarity results by similarity, and determines an adjacent node set based on the sorting results, the adjacent node set includes several user nodes adjacent to the target node; the virtual graph update module updates the dynamic virtual graph based on the target node and the adjacent node set; the telecommunications fraud identification module inputs the dynamic virtual graph into the telecommunications fraud identification model to obtain the telecommunications fraud identification result output by the telecommunications fraud identification model, therefore, by calculating the similarity between the target node and multiple user nodes, evaluating the correlation between them, sorting the calculated similarity results, selecting several user nodes with the highest similarity, and forming a node set adjacent to the target node. These adjacent nodes are user nodes with high correlation with the target node. The dynamic virtual graph is updated based on the target node and the set of adjacent nodes. The way of updating the dynamic virtual graph can more flexibly reflect the complexity and changes of telecommunications fraud activities and adapt to the needs of telecommunications fraud identification in different scenarios. The updated dynamic virtual graph is input into the pre-trained telecommunications fraud identification model to identify whether the target node is a telecommunications fraud node, which effectively solves the problem in the prior art that fraud-related numbers for cross-provincial or cross-operator calls may appear as isolated nodes or missing connections in the graph, thereby affecting the accuracy of model recognition. In this way, the connectivity of the graph is enhanced, and the accuracy and effectiveness of telecommunications fraud identification are improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 It is a schematic diagram of a smart city service platform based on cloud computing and big data provided by the present invention;
[0043] Figure 2 is a schematic diagram of a smart city service method based on cloud computing and big data provided by the present invention;
[0044] Figure 3 It is a structural schematic diagram of the telecommunications fraud identification model provided by the present invention;
[0045] Figure 4 A schematic diagram of an electronic device according to an embodiment of the present invention;
[0046] Figure 5 A schematic diagram of an embodiment of a computer-readable storage medium provided for an embodiment of the present invention. DETAILED DESCRIPTION
[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.
[0048] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0049] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or explanation". Any embodiment described as "for example" in the present invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any technician in the field to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present invention.
[0050] Reference Figure 1 , Figure 1 It is a schematic diagram of the smart city service platform based on cloud computing and big data provided by the present invention. The smart city service platform based on cloud computing and big data includes a control center and at least one smart city service system. The smart city service system includes at least a telecommunications fraud identification subsystem; the telecommunications fraud identification subsystem includes an adjacent node processing module, a virtual map update module and a telecommunications fraud identification module.
[0051] Optionally, the adjacent node processing module determines similarity results between the target node and multiple user nodes, sorts the similarity results, and determines an adjacent node set based on the sorting results, where the adjacent node set includes several user nodes adjacent to the target node.
[0052] Optionally, the virtual graph update module updates the dynamic virtual graph based on the target node and the set of adjacent nodes.
[0053] Optionally, the telecommunications fraud identification module inputs the dynamic virtual graph into the telecommunications fraud identification model to obtain the telecommunications fraud identification result output by the telecommunications fraud identification model, wherein the telecommunications fraud identification model is trained based on the sample dynamic virtual graph and the telecommunications fraud identification result label corresponding to the sample dynamic virtual graph.
[0054] The smart city service platform based on cloud computing and big data of the embodiment of the present invention includes a control center and a smart city service system, the smart city service system includes a telecommunications fraud identification subsystem; the telecommunications fraud identification subsystem includes an adjacent node processing module, a virtual graph update module and a telecommunications fraud identification module; the adjacent node processing module determines the similarity results between the target node and multiple user nodes, sorts each similarity result by similarity, and determines an adjacent node set based on the sorting result, the adjacent node set includes several user nodes adjacent to the target node; the virtual graph update module updates the dynamic virtual graph based on the target node and the adjacent node set; the telecommunications fraud identification module inputs the dynamic virtual graph into the telecommunications fraud identification model to obtain the telecommunications fraud identification result output by the telecommunications fraud identification model, therefore, by calculating the similarity between the target node and multiple user nodes, evaluating the correlation between them, sorting the calculated similarity results, selecting several user nodes with the highest similarity, and forming a node set adjacent to the target node. These adjacent nodes are user nodes with high correlation with the target node. The dynamic virtual graph is updated based on the target node and the set of adjacent nodes. The way of updating the dynamic virtual graph can more flexibly reflect the complexity and changes of telecommunications fraud activities and adapt to the needs of telecommunications fraud identification in different scenarios. The updated dynamic virtual graph is input into the pre-trained telecommunications fraud identification model to identify whether the target node is a telecommunications fraud node, which effectively solves the problem in the prior art that fraud-related numbers for cross-provincial or cross-operator calls may appear as isolated nodes or missing connections in the graph, thereby affecting the accuracy of model recognition. In this way, the connectivity of the graph is enhanced, and the accuracy and effectiveness of telecommunications fraud identification are improved.
[0055] See also Figure 2 , Figure 2 A schematic diagram of a smart city service method based on cloud computing and big data provided by an embodiment of the present invention, the smart city service method based on cloud computing and big data includes:
[0056] Step 110, determining similarity results between the target node and multiple user nodes, performing similarity sorting on each similarity result, and determining an adjacent node set based on the sorting result, wherein the adjacent node set includes a plurality of user nodes adjacent to the target node.
[0057] Here, the target node refers to the user node or phone number that is being evaluated for fraudulent behavior. User nodes refer to other user nodes or phone numbers in the telecommunications network, which may have call or text message interaction records with the target node. The similarity result is a measure of the correlation between the target node and the user node. The calculation method can be based on a variety of features, such as call records, text message content, etc. The specific calculation method can refer to the following embodiment and will not be repeated here.
[0058] Here, the adjacent node set is a set of several user nodes with the highest correlation with the target node selected based on the similarity sorting result, which is used to update the dynamic virtual graph and further model input.
[0059] In one embodiment, the similarity results between the target node and multiple user nodes are determined, the similarity results are sorted, several similarity results with greater similarity are selected, and an adjacent node set is determined, and the adjacent node set includes several user nodes adjacent to the target node. By sorting all the calculated similarity results from high to low according to the similarity value, after sorting, the user nodes with high similarity values indicate a stronger correlation with the target node, which can ensure that only those nodes that are closely related to the target node are considered, thereby improving the accuracy of the telecommunications fraud identification model. Since telecommunications user data is usually distributed in different operators and provinces, through similarity calculation and selection of high similarity nodes, the scattered data can be effectively integrated to make up for the problem of data islands.
[0060] Step 120: Update the dynamic virtual graph based on the target node and the set of adjacent nodes.
[0061] Here, the dynamic virtual graph refers to a user relationship graph that is dynamically adjusted according to real-time data. It reflects the real-time status and changing trends of the relationships between users. It is not a static graph structure, but an abstract concept that is dynamically updated as the data changes.
[0062] In one embodiment, a dynamic virtual graph is constructed based on the target node and the set of adjacent nodes. When the call list data of each node changes, the set of adjacent nodes also changes accordingly, and the dynamic virtual graph is updated in real time. By updating the dynamic virtual graph, changes in the relationship between users in the telecommunications network can be reflected in a timely manner. The behavior and interaction of users in telecommunications data are dynamically changing. Updating the dynamic virtual graph can ensure that the model makes predictions and decisions based on the latest data, and can more accurately capture the patterns and trends of telecommunications fraud, thereby improving the accuracy and efficiency of identification. The specific update method of the dynamic virtual graph can refer to the following embodiment, which will not be repeated here.
[0063] Step 130, input the dynamic virtual graph into the telecommunications fraud identification model to obtain the telecommunications fraud identification result output by the telecommunications fraud identification model.
[0064] Here, the telecommunications fraud identification model is a model specifically used to identify telecommunications fraud behaviors. By analyzing the patterns and features in user behavior data, the model can automatically determine which behaviors may involve fraud, thereby improving the security and stability of the telecommunications network.
[0065] In one embodiment, the telecommunications fraud identification model can determine the probability that the target node is involved in telecommunications fraud and the probability that it is not involved in telecommunications fraud, thereby determining the telecommunications fraud identification result. The probability of being involved in telecommunications fraud is the result of the telecommunications fraud identification model predicting the target node based on its behavior pattern, interaction with adjacent nodes, and possible fraud patterns. A high probability of being involved in fraud indicates that the behavior characteristics of the target node are more consistent with known fraud patterns. The probability of not being involved in telecommunications fraud behavior is the probability that the telecommunications fraud identification model infers that the target node is not involved in fraud behavior based on the target node's behavior data, its relationship with adjacent nodes, and the comparison of normal user behavior patterns. Through the calculation and comparison of these two probability values, the telecommunications fraud identification model can determine whether the target node is likely to be involved in telecommunications fraud behavior, so that the model can make decisions in uncertain situations, and can provide confidence information about the identification results, helping telecommunications operators to effectively manage and respond to possible fraud risks. The specific structure of the telecommunications fraud identification model can refer to the following embodiments, which will not be repeated here.
[0066] Among them, the telecommunications fraud identification model is trained based on the sample dynamic virtual graph and the telecommunications fraud identification result labels corresponding to the sample dynamic virtual graph.
[0067] The telecommunications fraud identification model is trained using sample dynamic virtual graphs and corresponding telecommunications fraud identification result labels. During the training process, the model learns how to extract features from dynamic virtual graphs, analyze relationships between users, and identify possible fraud patterns. The model adjusts parameters through optimization algorithms (such as gradient descent) so that the model can accurately predict the fraud risk of new unseen sample nodes to obtain a complete telecommunications fraud identification model. Through labeled training data, the telecommunications fraud identification model can learn the characteristics and patterns of real fraud cases, thereby improving the accuracy and reliability of identification.
[0068] The embodiment of the present invention can find the user node most relevant to the target node and form an adjacent node set by determining the similarity between the target node and multiple user nodes and sorting them. This helps to focus on potential fraudulent behavior, because fraud usually involves complex relationships between multiple users. Furthermore, updating the dynamic virtual graph based on the target node and the adjacent node set can reflect real-time communication and interaction patterns, including cross-provincial and cross-operator information. This dynamic update ensures that the telecommunications fraud model can capture the latest user behavior patterns in a timely manner, thereby more accurately identifying potential fraudulent behavior. The updated dynamic virtual graph is input into the telecommunications fraud identification model, which is trained based on the sample dynamic virtual graph and the corresponding fraud identification result label, thereby ensuring the high generalization ability and accuracy of the model. It effectively improves the recognition ability and accuracy of telecommunications fraud behavior, and solves the problem of difficulty in identifying cross-provincial or cross-operator call numbers in the prior art.
[0069] Based on the above embodiment, determining the adjacent node set based on the sorting result includes:
[0070] Clustering the feature vector of the target node and the feature vectors of multiple user nodes to determine a target node cluster, where the target node cluster includes the target node and several user nodes with similar features to the target node;
[0071] Sampling several user nodes in the target node cluster that have similar characteristics to the target node to obtain a sampled node set;
[0072] Based on the sorting result and the sampled node set, the adjacent node set is determined.
[0073] Here, clustering refers to the process of grouping nodes so that the similarity of nodes in the same cluster is maximized and the similarity of nodes in different clusters is minimized. The target node cluster refers to a set of user nodes that contain the target node and its similar features. The sampled node set refers to a set of user nodes extracted from the target node cluster that are similar to the target node's features.
[0074] In one embodiment, the feature vector of the target node and the feature vectors of multiple user nodes are clustered to determine the target node cluster; further, several user nodes with similar features to the target node in the target node cluster are sampled to obtain a sampled node set; further, based on the sorting results and the sampled node set, the adjacent node set is determined. The feature vectors of the target node and other user nodes are clustered using a clustering algorithm. The feature vectors represent various characteristics of the node, such as behavioral characteristics, social relationship characteristics, etc. The clustering algorithm can effectively group nodes with similar characteristics, which helps to identify user nodes with similar behavioral patterns or characteristics to the target node, and can improve the accuracy and efficiency of subsequent analysis. Sampling reduces the amount of data to be processed and the computational overhead, while ensuring that user nodes with significant effects on the target node are retained, which helps to improve the efficiency of data processing and the reliability of the analysis results. Through the sorting results in the above embodiment and the sampled node set determined in this embodiment, the adjacent node set is determined, so that a more accurate dynamic virtual graph can be constructed, thereby providing high-quality input data for subsequent telecommunications fraud identification and improving the accuracy of identification.
[0075] In another embodiment, the feature vector of the target node and the feature vectors of multiple user nodes are clustered to determine the target node cluster; further, based on the sorting result and the target node cluster, the adjacent node set is determined. The sampling step is omitted, and all nodes in the target node cluster are directly sorted by similarity, which simplifies the processing flow. A more comprehensive adjacent node set is determined directly based on all nodes in the target node cluster and the sorting results in the above embodiment.
[0076] Exemplarily, the characteristics of each user (such as call frequency, etc.) are obtained from the user call record data, and several nearest neighbor nodes and several cluster sampling nodes are taken as adjacent nodes of the target node through the calculated similarity measurement, and a dynamic virtual graph is constructed in sequence; specifically, users are taken as nodes and edges between nodes are constructed based on the similarity between users.
[0077] Exemplarily, the clustering algorithm may adopt a k-means algorithm (k-means clustering algorithm).
[0078] The embodiment of the present invention determines the target node cluster by clustering the feature vector of the target node with the feature vectors of multiple user nodes. The clustering step helps to group nodes with similar features, thereby excluding irrelevant nodes in the preliminary screening. Furthermore, by sampling the nodes in the target node cluster, the range of candidate nodes can be further narrowed, and the most representative nodes can be retained. Finally, the sorting results and the sampled node set are combined to ensure that the selected adjacent node set not only has a high similarity with the target node, but also can represent different feature distributions. Such a processing method effectively improves the input data quality of the model and reduces the impact of noise data, thereby enhancing the overall performance and recognition accuracy of the telecommunications fraud identification model.
[0079] Based on any of the above embodiments, in the method, determining the similarity results between the target node and multiple user nodes includes:
[0080] Based on the first node feature of the target node and the first node features of multiple user nodes, a difference between the target node and each user node on the same feature dimension is calculated, where the first node feature includes multiple feature dimensions;
[0081] Based on the first node feature of the target node and the first node features of multiple user nodes, the maximum difference values corresponding to multiple feature dimensions are calculated, and the maximum difference value corresponding to any feature dimension is the difference between the maximum value in the feature dimension of each first node feature and the minimum value in the feature dimension of each first node feature;
[0082] Based on the difference between the target node and each user node on the same feature dimension and the difference between the maximum value and the minimum value of the same feature dimension among all nodes, the second node feature of the target node and the second node features of multiple user nodes are calculated;
[0083] The similarity between the second node feature of the target node and the second node features of each user node is calculated, and the similarity results between the target node and the multiple user nodes are determined according to the calculation results.
[0084] Here, the first node feature refers to the initial feature vector of the target node and the user node, which contains feature values of multiple dimensions. The second node feature refers to the feature vector after normalization, which is used to more accurately calculate the similarity between nodes. Similarity is an indicator to measure the similarity between the target node and the user node.
[0085] Exemplarily, the first node features of all nodes are represented as ; Where n represents the number of nodes. represents the i-th node, represents the jth node; let the values of all feature dimensions of the first node feature of the i-th node be ; where d means the number of dimensions is d. represents the eigenvalue of the first dimension of the i-th node, Represents the feature value of the jth dimension of the i-th node. Assuming that the i-th node is the target node, the formula for calculating the second node feature of the i-th node and the second node feature of the j-th node is as follows:
[0086] Formula 1: ;
[0087] Formula 2: ;
[0088] Since the dimensions of each feature are different, normalization is used to remove the impact of the dimension on the numerical calculation. In this way, the impact of the different value ranges of each feature dimension can be eliminated through standardization, making the feature values of different dimensions comparable.
[0089] The embodiment of the present invention quantifies their differences in various dimensions by calculating the difference between the target node and each user node in each feature dimension, and compares the feature vector of the target node and the feature vector of each user node. For each feature dimension, the maximum value of the difference between the target node and all user nodes is determined. This maximum difference reflects the maximum difference in the features between nodes in this dimension. The feature difference calculated in the first step is standardized, that is, the difference in each feature dimension is divided by the maximum difference in the feature dimension. The purpose of this is to eliminate the dimensional difference between different feature dimensions, so that the contribution of each feature dimension to the similarity calculation is more balanced. Based on the standardized difference, the similarity between the second node feature of the target node and the second node feature of each user node is calculated. Different metrics can be used for the calculation of similarity, such as cosine similarity or Euclidean distance. According to the result of the similarity calculation, the similarity ranking between the target node and multiple user nodes is determined. According to the level of similarity, the user node set most relevant to the target node, that is, the adjacent node set, is selected.
[0090] Based on any of the above embodiments, in the method, calculating the similarity between the second node feature of the target node and the second node features of each user node, and determining the similarity results between the target node and the multiple user nodes according to the calculation results, includes:
[0091] Calculate the Euclidean distance between the second node feature of the target node and the second node feature of each user node to obtain a calculation result set;
[0092] The similarity results between the target node and multiple user nodes are determined based on the calculation result set.
[0093] Here, Euclidean distance is a measurement method for calculating the straight-line distance between two points (or vectors). The calculation result set is a set of distance values obtained by Euclidean distance calculation, which is used to evaluate the similarity between nodes.
[0094] In one embodiment, the Euclidean distance between the second node feature of the target node and the second node feature of each user node is calculated to obtain a calculation result set, wherein any calculation result in the calculation result set corresponds to the Euclidean distance between the second node feature of the target node and the second node feature of a user node; by calculating the Euclidean distance, the similarity between the target node and each user node can be quantified, providing an intuitive numerical comparison basis. The calculation results in the calculation result set are sorted to determine the similarity results between the target node and each user node. By sorting, it can be determined which user nodes are most similar to the target node. The sorting result provides a clear comparison basis for the similarity between the target node and each user node, so that the model can effectively identify and process the most relevant user nodes, thereby improving the recognition accuracy and processing efficiency of the model.
[0095] In another embodiment, other calculation methods may be used to quantify the similarity between the target node and each user node, including but not limited to cosine similarity and Manhattan distance, and similar calculation methods are equivalent replacements, do not deviate from the technical solution of the present invention, and should fall within the protection scope of the present invention, and will not be elaborated herein.
[0096] Exemplarily, the second node feature of the target node i is calculated as , the second node feature of the jth user node is ,It is known that the second node feature of target node i and the second node feature of the jth user node are both standardized values, then the Euclidean distance formula for calculating the second node feature of target node i and the second node feature of the jth user node is:
[0097] Formula 3: ;
[0098] Among them, d represents the number of dimensions of the second node features of the two nodes, and L represents the straight-line distance (i.e., Euclidean distance) between the feature vectors of the two nodes, which is used to measure the similarity between the two nodes.
[0099] The embodiment of the present invention can effectively identify the user node that is most similar to the target node through accurate similarity calculation and sorting, thereby ensuring high correlation between nodes when updating the dynamic virtual graph. This method improves the accuracy of the telecommunications fraud identification model and helps to more effectively identify and prevent telecommunications fraud.
[0100] Based on any of the above embodiments, in the method, the dynamic virtual graph is input into the telecommunications fraud identification model to obtain the telecommunications fraud identification result output by the telecommunications fraud identification model, including:
[0101] Input the dynamic virtual graph into the feature extraction layer in the telecommunications fraud identification model to obtain an initial feature matrix output by the feature extraction layer, wherein the initial feature matrix includes feature representations of all nodes in the dynamic virtual graph;
[0102] The initial feature matrix is input into the attention calculation layer in the telecommunications fraud recognition model to obtain the attention coefficient set output by the attention calculation layer. The attention calculation layer is used to calculate the attention coefficient of the feature representation of each adjacent node in the adjacent node set in the initial feature matrix to the feature representation of the target node.
[0103] The attention coefficient set and the initial feature matrix are input into the feature aggregation layer in the telecommunications fraud recognition model to obtain the feature matrix output by the feature aggregation layer. The feature aggregation layer is used to aggregate the feature representation of the target node in the initial feature matrix with the feature representation of each adjacent node in the adjacent node set to obtain the feature matrix;
[0104] The feature matrix is input into the feature generalization layer in the telecommunications fraud identification model to obtain a generalized feature matrix output by the feature generalization layer. The feature generalization layer is used to perform feature extraction and dimension change processing on the feature matrix to determine the generalized feature matrix.
[0105] The generalized feature matrix is input into the probability calculation layer in the telecommunications fraud identification model to obtain the telecommunications fraud identification result output by the probability calculation layer. The probability calculation layer is used to calculate the fraud probability and non-fraud probability of the target node in the generalized feature matrix to determine the telecommunications fraud identification result.
[0106] To better understand the telecommunications fraud identification model of the embodiment of the present invention, Figure 3 As shown, the telecommunications fraud identification model includes a feature extraction layer, an attention calculation layer, a feature aggregation layer, a feature generalization layer and a probability calculation layer.
[0107] In one embodiment, the dynamic virtual graph is input into the feature extraction layer in the telecommunications fraud identification model to obtain an initial feature matrix; the main function of the feature extraction layer is to extract the initial feature representation of the node from the input dynamic virtual graph, and each node in the dynamic virtual graph is represented by a feature vector, which contains various attributes and relationship information of the node, and integrates the feature vectors of all nodes into an initial feature matrix to provide support for subsequent calculations. Further, the initial feature matrix is input into the attention calculation layer in the telecommunications fraud identification model to calculate the attention coefficient set; the attention calculation layer determines the influence of adjacent nodes on the target node by calculating the attention coefficient. Specifically, it calculates the attention coefficient of the feature representation of each adjacent node in the initial feature matrix to the feature representation of the target node. These coefficients represent the importance of each adjacent node, so that the model can automatically focus on more important adjacent node features and ignore irrelevant features. By focusing on important features, telecommunications fraud nodes can be more accurately identified. Further, the attention coefficient set and the initial feature matrix are input into the feature aggregation layer in the telecommunications fraud identification model to obtain the feature matrix; the feature aggregation layer is used to aggregate the feature representation of the target node in the initial feature matrix with the feature representation of each adjacent node in the adjacent node set to obtain a comprehensive feature matrix. Through the aggregation operation, the feature information of the target node and its adjacent nodes is combined to form a richer feature representation, which helps to improve the generalization ability of the model. Further, the feature matrix is input into the feature generalization layer to obtain the generalized feature matrix in the telecommunications fraud identification model; the feature generalization layer further extracts features and performs dimension change processing on the aggregated feature matrix to generate a generalized feature matrix. This process includes operations such as dimensionality reduction and nonlinear transformation, so that the feature representation is more suitable for subsequent classification tasks. Through feature extraction and dimensionality reduction, redundant information is removed to improve the generalization ability of the model. Further, the generalized feature matrix is input into the probability calculation layer in the telecommunications fraud identification model to obtain the telecommunications fraud identification result; the probability calculation layer receives the generalized feature matrix and uses the classifier to calculate the probability of fraud and non-fraud of the target node. Through these probability values, it is ultimately determined whether the target node is a telecommunications fraud node, providing a basis for identifying telecommunications fraud nodes.
[0108] For example, the feature extraction layer in the telecommunications fraud identification model can be a graph attention network in a graph convolutional neural network; the attention calculation layer in the telecommunications fraud identification model can be obtained by introducing an attention mechanism in a graph convolutional neural network. In a traditional graph convolutional neural network (GCN), the influence of a node's neighbor nodes on its features is uniform, and each neighbor node contributes the same to the target node. After the introduction of the attention mechanism, the weights can be dynamically adjusted according to the similarity of node features, increasing the attention to important neighbor nodes and reducing the attention to unimportant neighbor nodes, thereby improving the expressiveness and discrimination of features.
[0109] Exemplarily, the feature generalization layer in the telecommunications fraud identification model can be a multi-layer perceptron MLP. As a deep neural network structure, the multi-layer perceptron (MLP) has powerful nonlinear mapping capabilities. It can learn nonlinear relationships between complex features and help extract and learn higher-order feature representations. The feature generalization layer can extract more representative and abstract features from the input generalized feature matrix by stacking multiple fully connected layers (the basic component of MLP). These features can better capture the complex patterns and laws of fraudulent behavior. Therefore, designing the feature generalization layer as a multi-layer perceptron (MLP) helps to achieve more efficient and accurate feature extraction and representation in the telecommunications fraud identification model, thereby improving the overall performance and recognition accuracy of the model.
[0110] Exemplarily, the probability calculation layer in the telecommunications fraud identification model may include a Softmax function. The task of the telecommunications fraud identification model is multi-category classification (such as identifying different categories involving fraud and non-fraud). The Softmax function can effectively output the probability of each category, that is, it can output the probability of each node belonging to fraud or non-fraud. If the probability of fraud is greater than the probability of non-fraud, then the user corresponding to the node is output as being involved in fraud, and then the telecommunications fraud identification result is output. If the probability of fraud is less than the probability of non-fraud, then the user corresponding to the node is output as not being involved in fraud, and then the telecommunications fraud identification result is output. Therefore, designing the probability calculation layer to include a Softmax function can enable the telecommunications fraud identification model to produce a standardized probability output, providing a more intuitive and explainable classification result.
[0111] The embodiment of the present invention obtains the initial feature representation of all nodes in the dynamic virtual graph through the feature extraction layer, so that the node features are more comprehensive; then, the attention calculation layer calculates the attention coefficient of the adjacent nodes to the target node, so that the model can focus on more important node features; then, the feature aggregation layer aggregates the feature representation of the target node and its adjacent nodes to improve the overall expressiveness of the features; then, the feature generalization layer further extracts and processes the dimensionality changes of the features, making the features more generalized and abstract; finally, the probability calculation layer calculates the probability of fraud and the probability of non-fraud, and outputs the final telecommunications fraud identification result. Through this series of steps, the model can more accurately identify telecommunications fraud and improve the accuracy and reliability of identification. Through the deep mining and processing of node features, the model can more comprehensively and accurately identify telecommunications fraud behaviors, improve the recognition ability and reliability of the model, and effectively prevent telecommunications fraud risks.
[0112] Based on any of the above embodiments, in this method, the attention coefficient set is obtained based on the following method:
[0113] Performing linear transformation on the feature representation of the target node in the initial feature matrix and the feature representation of each adjacent node in the adjacent node set, and then concatenating them to determine a first concatenated feature vector set, wherein the first concatenated feature vector set includes a first concatenated feature vector obtained by concatenating the feature representation of the target node and the feature representation of each adjacent node in the adjacent node set after linear transformation;
[0114] Performing vector transposition on each first splicing feature vector in the first splicing feature vector set to obtain a second splicing vector set;
[0115] Based on the activation function and the exponential function, each second splicing vector in the second splicing vector set is calculated to obtain an attention score set, where the attention score set includes the attention score of each adjacent node in the adjacent node set to the target node, and each attention score is a positive number;
[0116] Accumulate the attention scores of the attention score set to obtain the total attention score;
[0117] Based on the attention score set and the total attention score, an attention coefficient set is determined, where the attention coefficient set includes the attention coefficient of each adjacent node in the adjacent node set to the target node.
[0118] Here, linear transformation refers to mapping the input feature vector to a new feature space through matrix multiplication to enhance the feature representation capability. Concatenation refers to connecting two feature vectors into a longer vector to combine feature information from different nodes. An activation function is a nonlinear function, such as ReLU (activation function), which is used to introduce nonlinear characteristics so that the model can represent more complex feature relationships. The exponential function usually uses the exp function to convert the attention score into a positive number to ensure that all attention scores are positive. The attention score is used to measure the degree of attention of a neighboring node to the target node. The higher the value, the more important the neighboring node is to the target node.
[0119] In one embodiment, the features of the target node and the adjacent nodes are linearly transformed and spliced to determine a first spliced feature vector set; the original features of the target node and the adjacent nodes are projected into a new feature space through linear transformation (such as matrix multiplication), and the splicing operation combines the transformed features of the target node and the adjacent nodes to form a first spliced feature vector. The spliced feature vector can integrate the information of the target node and its adjacent nodes, and provide rich context information for subsequent attention calculations. Further, the first spliced feature vector is vector transposed to obtain a second spliced vector set. The transposition operation facilitates matrix operations and dot product operations between vectors in subsequent calculations. Further, the attention score set is calculated based on the activation function and the exponential function. The activation function introduces nonlinearity to enhance the expression ability of the model. The exponential function ensures that the attention score is positive, which is conducive to normalization processing. Further, based on the attention score and the sum of the attention scores, the attention coefficient set is determined. In this way, the model can dynamically adjust the weights of neighboring nodes, increase attention to important node features, and thus improve the effect of feature aggregation.
[0120] For example, the computation graph convolutional neural network In the layer, the formula for the attention coefficient of node j to node i is as follows:
[0121]
[0122] in, is the transpose of the attention learning parameters of the graph convolutional neural network; refers to and splicing; It refers to the The feature representation of node i in the layer, It represents the result of the feature of node i after being transformed by the weight matrix W. Represents the result of the feature of node j after being transformed by the weight matrix W; Refers to the set of neighbor nodes of node i; is an exponential function used to convert the attention score into a non-negative value; It is an activation function and an improved .
[0123] Through these calculation steps, the embodiment of the present invention effectively calculates the attention coefficients between nodes in the dynamic virtual graph structure. The attention mechanism enables the telecommunications fraud identification model to dynamically adjust the weights of neighboring nodes and capture more complex and detailed node relationships, thereby improving the representation ability and prediction accuracy of the telecommunications fraud identification model on graph data.
[0124] See also Figure 4 , Figure 4Schematic diagram of an electronic device provided by an embodiment of the present invention. Figure 4 As shown, an embodiment of the present invention provides an electronic device 400, including a memory 410, a processor 420, and a computer program 411 stored in the memory 410 and executable on the processor 420. When the processor 420 executes the computer program 411, the following steps are implemented:
[0125] Determine similarity results between a target node and a plurality of user nodes, perform similarity sorting on each of the similarity results, and determine an adjacent node set based on the sorting results, wherein the adjacent node set includes a plurality of user nodes adjacent to the target node;
[0126] Based on the target node and the set of adjacent nodes, updating a dynamic virtual graph;
[0127] Inputting the dynamic virtual graph into a telecommunications fraud identification model to obtain a telecommunications fraud identification result output by the telecommunications fraud identification model;
[0128] Among them, the telecommunications fraud identification model is trained based on the sample dynamic virtual graph and the telecommunications fraud identification result labels corresponding to the sample dynamic virtual graph.
[0129] See also Figure 5 , Figure 5 Schematic diagram of an embodiment of a computer-readable storage medium provided in an embodiment of the present invention. Figure 5 As shown, this embodiment provides a computer-readable storage medium 500, on which a computer program 511 is stored. When the computer program 511 is executed by a processor, the following steps are implemented:
[0130] Determine similarity results between a target node and a plurality of user nodes, perform similarity sorting on each of the similarity results, and determine an adjacent node set based on the sorting results, wherein the adjacent node set includes a plurality of user nodes adjacent to the target node;
[0131] Based on the target node and the set of adjacent nodes, updating a dynamic virtual graph;
[0132] Inputting the dynamic virtual graph into a telecommunications fraud identification model to obtain a telecommunications fraud identification result output by the telecommunications fraud identification model;
[0133] Among them, the telecommunications fraud identification model is trained based on the sample dynamic virtual graph and the telecommunications fraud identification result labels corresponding to the sample dynamic virtual graph.
[0134] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.
[0135] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0136] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0137] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0139] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.
[0140] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.
Claims
1. A smart city service platform based on cloud computing and big data, characterized by: The smart city service platform based on cloud computing and big data includes a control center and at least one smart city service system, and the smart city service system includes at least a telecommunications fraud identification subsystem; the telecommunications fraud identification subsystem includes an adjacent node processing module, a virtual map update module and a telecommunications fraud identification module; The adjacent node processing module is used to determine similarity results between the target node and multiple user nodes, perform similarity sorting on each of the similarity results, and determine an adjacent node set based on the sorting results, wherein the adjacent node set includes a plurality of user nodes adjacent to the target node; The virtual graph updating module is used to update the dynamic virtual graph based on the target node and the adjacent node set; The telecommunications fraud identification module is used to input the dynamic virtual graph into the telecommunications fraud identification model to obtain the telecommunications fraud identification result output by the telecommunications fraud identification model; wherein the telecommunications fraud identification model is trained based on the sample dynamic virtual graph and the telecommunications fraud identification result label corresponding to the sample dynamic virtual graph; The step of determining the adjacent node set based on the sorting result includes: Clustering the feature vector of the target node and the feature vectors of multiple user nodes to determine a target node cluster, where the target node cluster includes the target node and a plurality of user nodes having similar features to the target node; Sampling a number of user nodes in the target node cluster that have similar features to the target node to obtain a sampled node set; Based on the sorting result and the sampled node set, an adjacent node set is determined.
2. A smart city service method based on cloud computing and big data, characterized in that: The smart city service platform based on cloud computing and big data as described in claim 1, wherein the smart city service platform includes a control center and at least one smart city service system, wherein the smart city service system includes at least a telecommunications fraud identification subsystem, and the smart city service method includes: Determine similarity results between a target node and a plurality of user nodes, perform similarity sorting on each of the similarity results, and determine an adjacent node set based on the sorting results, wherein the adjacent node set includes a plurality of user nodes adjacent to the target node; Based on the target node and the set of adjacent nodes, updating a dynamic virtual graph; Inputting the dynamic virtual graph into a telecommunications fraud identification model to obtain a telecommunications fraud identification result output by the telecommunications fraud identification model; The telecommunications fraud identification model is obtained by training based on the sample dynamic virtual graph and the telecommunications fraud identification result label corresponding to the sample dynamic virtual graph; The step of determining the adjacent node set based on the sorting result includes: Clustering the feature vector of the target node and the feature vectors of multiple user nodes to determine a target node cluster, where the target node cluster includes the target node and a plurality of user nodes having similar features to the target node; Sampling a number of user nodes in the target node cluster that have similar features to the target node to obtain a sampled node set; Based on the sorting result and the sampled node set, an adjacent node set is determined.
3. The smart city service method based on cloud computing and big data according to claim 2 is characterized in that: The determining of similarity results between the target node and the plurality of user nodes comprises: Based on a first node feature of a target node and first node features of a plurality of user nodes, a difference between the target node and each user node on the same feature dimension is calculated, wherein the first node feature includes a plurality of feature dimensions; Based on the first node feature of the target node and the first node features of multiple user nodes, the maximum difference values corresponding to the multiple feature dimensions are calculated, and the maximum difference value corresponding to any of the feature dimensions is the difference between the maximum value in the feature dimension of each of the first node features and the minimum value in the feature dimension of each of the first node features; Based on the difference between the target node and each user node on the same feature dimension and the difference between the maximum value and the minimum value of the same feature dimension in all nodes, calculate the second node feature of the target node and the second node features of multiple user nodes; The similarity between the second node feature of the target node and the second node features of each of the user nodes is calculated, and the similarity results between the target node and the multiple user nodes are determined according to the calculation results.
4. The smart city service method based on cloud computing and big data according to claim 3 is characterized in that: The calculating the similarity between the second node feature of the target node and the second node features of each of the user nodes, and determining the similarity results between the target node and the plurality of user nodes according to the calculation result, comprises: Calculating the Euclidean distance between the second node feature of the target node and the second node feature of each of the user nodes to obtain a calculation result set; The similarity results between the target node and the multiple user nodes are determined according to the calculation result set.
5. The smart city service method based on cloud computing and big data according to claim 2 is characterized in that: The step of inputting the dynamic virtual graph into a telecommunications fraud identification model to obtain a telecommunications fraud identification result output by the telecommunications fraud identification model includes: Inputting the dynamic virtual graph into a feature extraction layer in a telecommunications fraud identification model to obtain an initial feature matrix output by the feature extraction layer, wherein the initial feature matrix includes feature representations of all nodes in the dynamic virtual graph; Input the initial feature matrix into the attention calculation layer in the telecommunications fraud identification model to obtain the attention coefficient set output by the attention calculation layer, wherein the attention calculation layer is used to calculate the attention coefficient of the feature representation of each adjacent node in the adjacent node set in the initial feature matrix to the feature representation of the target node; Input the attention coefficient set and the initial feature matrix into the feature aggregation layer in the telecommunications fraud identification model to obtain the feature matrix output by the feature aggregation layer, wherein the feature aggregation layer is used to aggregate the feature representation of the target node in the initial feature matrix with the feature representation of each adjacent node in the adjacent node set to obtain the feature matrix; The feature matrix is input into the feature generalization layer in the telecommunications fraud identification model to obtain a generalized feature matrix output by the feature generalization layer, wherein the feature generalization layer is used to perform feature extraction and dimension change processing on the feature matrix to determine a generalized feature matrix; The generalized feature matrix is input into the probability calculation layer in the telecommunications fraud identification model to obtain the telecommunications fraud identification result output by the probability calculation layer. The probability calculation layer is used to calculate the fraud probability and non-fraud probability of the target node in the generalized feature matrix to determine the telecommunications fraud identification result.
6. The smart city service method based on cloud computing and big data according to claim 5 is characterized in that: The attention coefficient set is obtained based on the following method: Performing linear transformation on the feature representation of the target node in the initial feature matrix and the feature representation of each adjacent node in the adjacent node set, and then concatenating them to determine a first concatenated feature vector set, wherein the first concatenated feature vector set includes a first concatenated feature vector obtained by concatenating the feature representation of the target node and the feature representation of each adjacent node in the adjacent node set after linear transformation; Performing vector transposition on each first splicing feature vector in the first splicing feature vector set to obtain a second splicing vector set; Based on the activation function and the exponential function, each second splicing vector in the second splicing vector set is calculated to obtain an attention score set, wherein the attention score set includes an attention score of each adjacent node in the adjacent node set to the target node, and each of the attention scores is a positive number; Accumulating the attention scores in the attention score set to obtain a total attention score; Based on the attention score set and the total of the attention scores, an attention coefficient set is determined, where the attention coefficient set includes the attention coefficient of each adjacent node in the adjacent node set to the target node.
7. 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 computer program, the smart city service method based on cloud computing and big data is implemented as described in any one of claims 2 to 6.
