E-SIM user behavior analysis method and system
By constructing user connection diagrams and traffic feature vectors and using the network to judge user connection rules, the problem of inaccurate package design in the existing technology is solved, and a more accurate recognition of SIM traffic package types is achieved.
Patent Information
- Application Number
- CN202411146093.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-20
- Publication Date
- 2024-11-19
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art is difficult to provide personalized traffic packages accurately based on user behavior, and the user behavior is not fixed, resulting in the package design not being accurate enough.
By obtaining the user's connection time period, application category and application usage traffic, building a two-dimensional connection diagram and traffic feature vector, using the connection judgment network and package detection network, judging the time period pattern and traffic characteristics of the user's connection network, and then determining the SIM traffic Package type.
It realizes accurate analysis of user behavior, provides more accurate SIM traffic package types, and meets the traffic needs of different user scenarios.
Smart Images

Figure CN118984449A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of communication technology, and in particular to an E-SIM user behavior analysis method and system. Background Art
[0002] At present, the traffic usage of users is determined by their behaviors. For example, users tend to use WiFi networks at home and SIM cards for network connection on the road. According to the habits of users, for example, users who take a break from 22:00 to 6:00 and users who take a break from 02:00 to 10:00 use traffic at different times, so the packages required by users are also different. However, since each user's behavior is specific, and the user's daily behavior is not fixed but has a certain regularity, it is necessary to explore the regularity in user behavior, so it is necessary to provide more accurate packages based on the detection of user behavior. Summary of the invention
[0003] The purpose of the present invention is to provide a method and system for analyzing user behavior of an E-SIM, so as to solve the above problems existing in the prior art.
[0004] In a first aspect, an embodiment of the present invention provides a method for analyzing user behavior of an E-SIM, including:
[0005] Obtain the user connection time period, application category, application usage time period and corresponding application usage traffic; the user connection time point shown indicates the time period in which the user uses the SIM card to connect to the network within 30 days; the application usage time period indicates the time period in which the application corresponding to the application category is used by the user within 30 days; the application usage traffic indicates the traffic used in the application usage time period corresponding to the application category;
[0006] Based on the user connection time period, segment the time period into fixed time lengths, construct an image, and obtain a two-dimensional connection graph; the two-dimensional connection graph represents the connection status of a fixed time length within 30 days;
[0007] By connecting the discrimination network, based on the two-dimensional connection graph, the regularity of the time period in which the user connects to the network is determined to obtain a connection feature graph;
[0008] Extracting features based on the application usage time period and the corresponding application usage traffic to obtain a traffic feature vector;
[0009] The connection feature graph and the traffic feature vector are input into a package detection network to obtain a SIM traffic package type.
[0010] Optionally, segmenting the user connection time period by a fixed time length to construct an image to obtain a two-dimensional connection graph includes:
[0011] Divide the user connection time period into fixed time lengths to obtain n user divided time periods;
[0012] Constructing n connection vectors with an initial value of 0; the length of the connection vector is m, where m represents the number of segments into which the fixed time length is evenly divided;
[0013] Modify the value in the connection vector according to the user segmentation time period; the value of the connection vector is 1, indicating that the SIM card is used to connect to the network in the corresponding time period, and the value of the connection vector is 0, indicating that the SIM card is not used to connect to the network in the corresponding time period;
[0014] According to the time point from far to near, the n connection vectors are superimposed from top to bottom to obtain a two-dimensional connection graph.
[0015] Optionally, the method of determining the regularity of the time period in which the user connects to the network based on the two-dimensional connection graph by connecting to the discrimination network to obtain the connection feature graph includes:
[0016] Obtaining a two-dimensional convolution kernel in a connection discrimination network; the length of the two-dimensional convolution kernel is less than the length of the two-dimensional connection graph; the width of the two-dimensional convolution kernel is 2;
[0017] The two-dimensional convolution kernel is sequentially convolved on the two-dimensional connection image with a step size of 1 to obtain a first convolution feature map; the first convolution feature map represents the characteristics of the state change of the SIM connection network at time points with a fixed time difference;
[0018] Based on the first convolution feature map, multiple vectors are segmented, and similarities are determined respectively to obtain a first similarity feature map;
[0019] The two-dimensional convolution kernel is sequentially convolved on the first convolution feature map with a step size of 1 to obtain a second convolution feature map;
[0020] Based on the second convolution feature map, split the multiple vectors, respectively determine the similarity, and obtain a second similarity feature map;
[0021] Multiple convolutions correspond to obtaining multiple i-th similar feature maps; i is a natural number greater than 0 and less than or equal to m;
[0022] Based on the first convolutional feature map and multiple i-th similarity feature maps, a connection feature map is obtained.
[0023] Optionally, the segmenting of multiple vectors based on the first convolution feature map, respectively determining similarities, and obtaining a first similarity feature map includes:
[0024] Keeping the width of the first convolution feature map unchanged, dividing the length of the first convolution feature map in sequence to obtain a plurality of first segmentation vectors;
[0025] Calculate the standard deviation of multiple values in the first segmentation vector to obtain a first similarity value;
[0026] If the first similarity value is less than a similarity threshold, the corresponding first segmentation vector in the first convolution feature map is set to 0 to obtain a first similarity feature map.
[0027] Optionally, obtaining a connection feature map based on the first convolution feature map and the plurality of i-th similar feature maps includes:
[0028] Finding the midpoint of the i-th similar feature graph according to the length and width of the i-th similar feature graph to obtain the i-th midpoint;
[0029] Overlap the i-th midpoint and superimpose the i-th similar feature graph to obtain a superimposed three-dimensional graph;
[0030] According to the length and width of the first convolution feature map, find the midpoint of the first convolution feature map to obtain the first convolution midpoint;
[0031] The i-th midpoint corresponding to the superimposed three-dimensional graph is overlapped with the midpoint of the first convolution, and the first convolution feature graph is superimposed on the surface corresponding to the first similar feature graph to obtain a connection feature graph; the length of the connection feature graph is equal to the length of the first convolution feature graph; the width of the connection feature graph is equal to the width of the first convolution feature graph;
[0032] The positions of the connection feature map except the i-th similarity feature map and the first convolution feature map are filled with 0.
[0033] Optionally, the training method of the connection discrimination network includes:
[0034] Obtain training data and annotate similar time periods; the training data includes a plurality of training i-th similar feature graphs and training connection feature graphs; the annotated similar time periods represent time periods in which the status of users using SIM cards to connect to the network is similar in the n user segmentation time periods;
[0035] Inputting the trained i-th similar feature graph into a similarity judgment network, judging similarity, and obtaining a first similar time period;
[0036] The loss value is calculated using the first similar time period and the annotated similar time period, and the connection discriminant network is trained by back propagation;
[0037] Based on the training connection feature graph, a connection discrimination network is trained.
[0038] Optionally, the training of a connection discrimination network based on the training connection feature graph includes:
[0039] Inputting the training connection feature graph into a similarity judgment network, judging similarity, and obtaining a second similar time period;
[0040] The loss value is calculated using the second similar time period and the annotated similar time period, and the back propagation training connects the discriminant network.
[0041] Optionally, extracting features based on the application usage time period and the corresponding application usage traffic to obtain a traffic feature vector includes:
[0042] Arrange the application usage traffic in sequence according to the corresponding application usage time periods to obtain a time traffic vector;
[0043] The time flow vector is input into the time flow neural network to obtain flow characteristics.
[0044] Optionally, the method for acquiring the second similar feature map is the same as the method for acquiring the first similar feature map.
[0045] In a second aspect, an embodiment of the present invention provides an E-SIM user behavior analysis system, including:
[0046] Acquisition module: obtains user connection time period, application category, application usage time period and corresponding application usage traffic; the user connection time point shown indicates the time period in which the user uses a SIM card to connect to the network within 30 days; the application usage time period indicates the time period in which the application corresponding to the application category is used by the user within 30 days; the application usage traffic indicates the traffic used in the application usage time period corresponding to the application category;
[0047] Segmentation module: based on the user connection time period, segmentation is performed with a fixed time length, an image is constructed, and a two-dimensional connection graph is obtained; the two-dimensional connection graph represents the connection status of a fixed time length within 30 days;
[0048] Connection identification module: through the connection identification network, based on the two-dimensional connection graph, the regularity of the time period when the user connects to the network is determined to obtain a connection feature graph;
[0049] Traffic module: extracting features based on the application usage time period and the corresponding application usage traffic to obtain a traffic feature vector;
[0050] Prediction module: input the connection feature graph and the traffic feature vector into the package detection network to obtain the SIM traffic package type.
[0051] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:
[0052] The embodiment of the present invention also provides a user behavior analysis method and system for E-SIM, the method comprising: obtaining a user connection time period, an application category, an application usage time period and a corresponding application usage flow; the user connection time point shown indicates a time period in which the user uses a SIM card to connect to the network within 30 days; the application usage time period indicates a time period in which the application corresponding to the application category is used by the user within 30 days; the application usage flow indicates the flow used in the application usage time period corresponding to the application category; based on the user connection time period, segmenting with a fixed time length, constructing an image, and obtaining a two-dimensional connection graph; the two-dimensional connection graph indicates a connection state of a fixed time length within 30 days; through a connection discrimination network, based on the two-dimensional connection graph, judging the regularity of the time period in which the user connects to the network, and obtaining a connection feature graph; based on the application usage time period and the corresponding application usage flow, extracting features to obtain a flow feature vector; inputting the connection feature graph and the flow feature vector into a package detection network to obtain a SIM flow package type.
[0053] In this embodiment, the user's network data and connection status are collected to analyze the user's behavior in various scenarios. According to the status of the network connection every day, a two-dimensional connection image that reflects the time relationship is constructed. By designing a special two-dimensional convolution kernel and a convolution method of the two-dimensional convolution kernel, the correlation relationship at the same time in two adjacent days is found. Due to the particularity of convolution, similar connection methods in the convolution process also have similar feature maps after feature extraction, so the features corresponding to the data at the same time of multiple days are split to obtain the first segmentation vector. According to the first segmentation vector, similar features are added to achieve a more accurate technical effect when detecting the type of SIM traffic package. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] Figure 1 The present invention provides a flowchart of a method for analyzing user behavior of an E-SIM. DETAILED DESCRIPTION
[0055] The present invention will be described in detail below in conjunction with the accompanying drawings.
[0056] Example 1
[0057] like Figure 1 As shown, an embodiment of the present invention provides a method for analyzing user behavior of an E-SIM, the method comprising:
[0058] S101: Obtain the user connection time period, application category, application usage time period and corresponding application usage traffic; the user connection time point shown represents the time period within 30 days when the user uses a SIM card to connect to the network; the application usage time period represents the time period within 30 days when the application corresponding to the application category is used by the user; the application usage traffic represents the traffic used in the application usage time period corresponding to the application category.
[0059] In this embodiment, the application category represents the application software used, such as a browser, video software, and reading software.
[0060] S102: Based on the user connection time period, segment the time period into fixed time lengths, construct an image, and obtain a two-dimensional connection graph; the two-dimensional connection graph represents the connection status of a fixed time length within 30 days.
[0061] In this embodiment, the fixed time length is 1 day.
[0062] S103: using a connection discrimination network, based on the two-dimensional connection graph, determining the regularity of the time period in which the user connects to the network, and obtaining a connection feature graph;
[0063] S104: Extract features based on the application usage time period and the corresponding application usage traffic to obtain a traffic feature vector.
[0064] Among them, the time flow neural network is used to obtain the flow feature vector.
[0065] S105: Input the connection feature graph and the traffic feature vector into a package detection network to obtain a SIM traffic package type.
[0066] Among them, in this embodiment, the SIM traffic package types include night packages, 3-day packages, daytime fixed-time packages and application-specific packages.
[0067] Among them, the package detection network is a fully-connected neural network (FCN), and the connection feature graph is stretched from top to bottom and from left to right from small to large in terms of the number of channels into a one-dimensional vector, which is input into the package detection network together with the traffic feature vector.
[0068] Among them, the marked SIM traffic package types are used to train the package detection network, the time traffic neural network and the trained connection discrimination network.
[0069] Optionally, segmenting the user connection time period by a fixed time length to construct an image to obtain a two-dimensional connection graph includes:
[0070] The user connection time period is divided into fixed time lengths to obtain n user divided time periods.
[0071] In this embodiment, the fixed time length is 1 day, and the time period for connecting to the network using the SIM card within the 30 days is divided into 30 user-divided time periods.
[0072] Construct n connection vectors with an initial value of 0; the length of the connection vector is m, and m represents the number of segments into which the fixed time length is evenly divided.
[0073] In this embodiment, if recording is performed every minute, m is 24*60=1440, which means that whether the user uses the corresponding SIM card to connect to the network is recorded every minute in one day; each time segment after cutting according to the number of segments represents the time within 1 minute.
[0074] The value in the connection vector is modified according to the user segmentation time period; a value of 1 in the connection vector indicates that the SIM is used to connect to the network in the corresponding time period, and a value of 0 in the connection vector indicates that the SIM is not used to connect to the network in the corresponding time period.
[0075] Among them, m and n are natural numbers greater than 0; for example, if the user does not use the SIM to connect to the network from 00:00 to 06:59, 7*60=420, then the value corresponding to the subscript 0-419 in the connection vector is set to 0; if the user uses the SIM to connect to the network from 07:00 to 07:59, 1*60, then the value corresponding to the subscript 420-479 in the connection vector is set to 1.
[0076] According to the time point from far to near, the n connection vectors are superimposed from top to bottom to obtain a two-dimensional connection graph.
[0077] Optionally, the method of determining the regularity of the time period in which the user connects to the network based on the two-dimensional connection graph by connecting to the discrimination network to obtain the connection feature graph includes:
[0078] A two-dimensional convolution kernel in a connection discrimination network is obtained; the length of the two-dimensional convolution kernel is less than the length of the two-dimensional connection graph; and the width of the two-dimensional convolution kernel is 2.
[0079] The two-dimensional convolution kernel is sequentially convolved on the two-dimensional connection image with a step size of 1 to obtain a first convolution feature map; the first convolution feature map represents the characteristics of the state change of the SIM connection network extracted at time points with a fixed time difference.
[0080] The two-dimensional convolution kernel is convolved from top to bottom and from left to right.
[0081] In this embodiment, the two-dimensional convolution kernel is convolved from top to bottom to obtain the correlation relationship at the same time in two adjacent days, such as the change between 12:00 on the first day and 12:00 on the second day.
[0082] Based on the first convolution feature map, multiple vectors are segmented, and similarities are determined respectively to obtain a first similarity feature map.
[0083] The two-dimensional convolution kernel is sequentially convolved on the first convolution feature map with a step size of 1 to obtain a second convolution feature map.
[0084] Based on the second convolution feature map, multiple vectors are segmented, and similarities are determined respectively to obtain a second similarity feature map.
[0085] Multiple convolutions correspond to obtaining multiple i-th similar feature maps; the i is a natural number greater than 0 and less than or equal to m.
[0086] The convolution stops when there are no similar vectors or when i equals m.
[0087] Based on the first convolutional feature map and multiple i-th similarity feature maps, a connection feature map is obtained.
[0088] Optionally, the segmenting of multiple vectors based on the first convolution feature map, respectively determining similarities, and obtaining a first similarity feature map includes:
[0089] The width of the first convolution feature map is kept unchanged, and the length of the first convolution feature map is divided in sequence to obtain multiple first segmentation vectors.
[0090] The length of the first segmentation vector is m-1, the number of the first segmentation vectors is np, and p is the length of the convolution kernel.
[0091] The first segmentation vector represents the features between two adjacent days.
[0092] The standard deviation of multiple values in the first segmentation vector is calculated to obtain a first similarity value.
[0093] If the first similarity value is less than a similarity threshold, the corresponding first segmentation vector in the first convolution feature map is set to 0 to obtain a first similarity feature map.
[0094] Among them, in this embodiment, the similarity sentence is 0.9.
[0095] Optionally, obtaining a connection feature map based on the first convolution feature map and the plurality of i-th similar feature maps includes:
[0096] Finding the midpoint of the i-th similar feature graph according to the length and width of the i-th similar feature graph to obtain the i-th midpoint;
[0097] Overlap the i-th midpoint and superimpose the i-th similar feature graph to obtain a superimposed three-dimensional graph;
[0098] According to the length and width of the first convolution feature map, find the midpoint of the first convolution feature map to obtain the first convolution midpoint;
[0099] The i-th midpoint corresponding to the superimposed three-dimensional graph is overlapped with the midpoint of the first convolution, and the first convolution feature graph is superimposed on the surface corresponding to the first similar feature graph to obtain a connection feature graph; the length of the connection feature graph is equal to the length of the first convolution feature graph; the width of the connection feature graph is equal to the width of the first convolution feature graph;
[0100] The positions of the connection feature map except the i-th similarity feature map and the first convolution feature map are filled with 0.
[0101] Optionally, the training method of the connection discrimination network includes:
[0102] Obtain training data and annotate similar time periods; the training data includes a plurality of training i-th similar feature graphs and training connection feature graphs; the annotated similar time periods represent time periods in which the status of users using SIM cards to connect to the network is similar in n user segmentation time periods.
[0103] In this embodiment, the training ith similarity feature graph represents the ith similarity feature graph obtained during a historical user connection time period, and the training connection feature graph represents the connection feature graph obtained during a historical user connection time period.
[0104] The trained i-th similar feature graph is input into a similarity judgment network to judge the similarity and obtain a first similar time period.
[0105] The similarity judgment network is a fully-connected neural network (FCN) used to judge similarity.
[0106] The first similar time period represents a time period in which the state of users using SIM cards to connect to the network in n user segmentation time periods output from training the i-th similar feature graph is similar. In this embodiment, the user segmentation time period is 1 day, and the first similar time period represents the time in each day when the state of users using SIM cards to connect to the network is similar.
[0107] The loss value is calculated using the first similar time period and the labeled similar time period, and the back propagation training connects the discriminant network.
[0108] In this embodiment, the loss value is calculated using a cross entropy loss function.
[0109] Based on the training connection feature graph, a connection discrimination network is trained.
[0110] Optionally, the training of a connection discrimination network based on the training connection feature graph includes:
[0111] The training connection feature graph is input into a similarity judgment network to judge the similarity and obtain a second similar time period.
[0112] The second similar time period represents a time period in which the status of users using SIM cards to connect to the network in n user segmentation time periods output by the training connection feature graph is similar. In this embodiment, the user segmentation time period is 1 day, and the second similar time period represents the moments in each day in which the status of users using SIM cards to connect to the network is similar.
[0113] The loss value is calculated using the second similar time period and the annotated similar time period, and the back propagation training connects the discriminant network.
[0114] In this embodiment, the loss value is calculated using a cross entropy loss function.
[0115] Optionally, extracting features based on the application usage time period and the corresponding application usage traffic to obtain a traffic feature vector includes:
[0116] The application usage traffic is arranged in sequence according to the corresponding application usage time periods to obtain a time traffic vector.
[0117] The time flow vector is input into a fully connected neural network to obtain a flow feature vector.
[0118] Wherein, the fully connected neural network is obtained by training.
[0119] Optionally, the method for acquiring the second similar feature map is the same as the method for acquiring the first similar feature map.
[0120] Example 2
[0121] Based on the above-mentioned E-SIM user behavior analysis method, an embodiment of the present invention further provides an E-SIM user behavior analysis system, which includes an acquisition module, a segmentation module, a connection determination module, a traffic module and a prediction module.
[0122] The acquisition module is used to obtain the user connection time period, application category, application usage time period and the corresponding application usage traffic; the user connection time point shown represents the time period within 30 days when the user uses a SIM card to connect to the network; the application usage time period represents the time period within 30 days when the application corresponding to the application category is used by the user; the application usage traffic represents the traffic used in the application usage time period corresponding to the application category.
[0123] The segmentation module is used to segment the user connection time period with a fixed time length, construct an image, and obtain a two-dimensional connection graph; the two-dimensional connection graph represents the connection status of a fixed time length within 30 days.
[0124] The connection determination module is used to determine the regularity of the time period in which the user connects to the network based on the two-dimensional connection graph through the connection determination network, and obtain a connection feature graph.
[0125] The traffic module is used to extract features based on the application usage time period and the corresponding application usage traffic to obtain a traffic feature vector.
[0126] The prediction module is used to input the connection feature graph and the traffic feature vector into the package detection network to obtain the SIM traffic package type.
Claims
1. A method for analyzing user behavior of E-SIM, It is characterized in that include: Obtain user connection time period, application category, application usage time period and corresponding application usage traffic; The user connection time points shown represent the time periods during which the user uses a SIM card to connect to the network within 30 days; the application usage time periods represent the time periods during which the application corresponding to the application category is used by the user within 30 days; the application usage traffic represents the traffic used during the application usage time periods corresponding to the application category; Based on the user connection time period, segment the time period into fixed time lengths, construct an image, and obtain a two-dimensional connection graph; the two-dimensional connection graph represents the connection status of a fixed time length within 30 days; By connecting the discrimination network, based on the two-dimensional connection graph, the regularity of the time period in which the user connects to the network is determined to obtain a connection feature graph; Extracting features based on the application usage time period and the corresponding application usage traffic to obtain a traffic feature vector; The connection feature graph and the traffic feature vector are input into a package detection network to obtain a SIM traffic package type.
2. The method for analyzing user behavior of an E-SIM according to claim 1, It is characterized in that The method of segmenting the user connection time period by a fixed time length, constructing an image, and obtaining a two-dimensional connection graph includes: Divide the user connection time period into fixed time lengths to obtain n user divided time periods; Constructing n connection vectors with an initial value of 0; the length of the connection vector is m, where m represents the number of segments into which the fixed time length is evenly divided; Modify the value in the connection vector according to the user segmentation time period; the value of the connection vector is 1, indicating that the SIM card is used to connect to the network in the corresponding time period, and the value of the connection vector is 0, indicating that the SIM card is not used to connect to the network in the corresponding time period; According to the time point from far to near, the n connection vectors are superimposed from top to bottom to obtain a two-dimensional connection graph.
3. The method for analyzing user behavior of an E-SIM according to claim 1, It is characterized in that The method of determining the regularity of the time period in which the user connects to the network based on the two-dimensional connection graph by connecting the network to the discrimination network to obtain the connection feature graph includes: Obtaining a two-dimensional convolution kernel in a connection discrimination network; the length of the two-dimensional convolution kernel is less than the length of the two-dimensional connection graph; the width of the two-dimensional convolution kernel is 2; The two-dimensional convolution kernel is sequentially convolved on the two-dimensional connection image with a step size of 1 to obtain a first convolution feature map; the first convolution feature map represents the characteristics of the state change of the SIM connection network at time points with a fixed time difference; Based on the first convolution feature map, multiple vectors are segmented, and similarities are determined respectively to obtain a first similarity feature map; The two-dimensional convolution kernel is sequentially convolved on the first convolution feature map with a step size of 1 to obtain a second convolution feature map; Based on the second convolution feature map, split the multiple vectors, respectively determine the similarity, and obtain a second similarity feature map; Multiple convolutions correspond to obtaining multiple i-th similar feature maps; i is a natural number greater than 0 and less than or equal to m; Based on the first convolutional feature map and multiple i-th similarity feature maps, a connection feature map is obtained.
4. The method for analyzing user behavior of an E-SIM according to claim 3, It is characterized in that The method of segmenting a plurality of vectors based on the first convolution feature map and respectively determining similarities to obtain a first similarity feature map includes: Keeping the width of the first convolution feature map unchanged, dividing the length of the first convolution feature map in sequence to obtain a plurality of first segmentation vectors; Calculate the standard deviation of multiple values in the first segmentation vector to obtain a first similarity value; If the first similarity value is less than a similarity threshold, the corresponding first segmentation vector in the first convolution feature map is set to 0 to obtain a first similarity feature map.
5. The method for analyzing user behavior of an E-SIM according to claim 3, It is characterized in that The method of obtaining a connection feature map based on the first convolution feature map and the plurality of i-th similar feature maps comprises: Finding the midpoint of the i-th similar feature graph according to the length and width of the i-th similar feature graph to obtain the i-th midpoint; Overlap the i-th midpoints and superimpose the i-th similar feature graphs to obtain a superimposed three-dimensional graph; According to the length and width of the first convolution feature map, find the midpoint of the first convolution feature map to obtain the first convolution midpoint; The i-th midpoint corresponding to the superimposed three-dimensional graph is overlapped with the midpoint of the first convolution, and the first convolution feature graph is superimposed on the surface corresponding to the first similar feature graph to obtain a connection feature graph; the length of the connection feature graph is equal to the length of the first convolution feature graph; the width of the connection feature graph is equal to the width of the first convolution feature graph; The positions of the connection feature map except the i-th similarity feature map and the first convolution feature map are filled with 0.
6. The method for analyzing user behavior of an E-SIM according to claim 1, It is characterized in that The training method of the connection discrimination network comprises: Obtain training data and annotate similar time periods; the training data includes a plurality of training i-th similar feature graphs and training connection feature graphs; the annotated similar time periods represent time periods in which the status of users using SIM cards to connect to the network is similar in the n user segmentation time periods; Inputting the trained i-th similar feature graph into a similarity judgment network, judging similarity, and obtaining a first similar time period; The loss value is calculated using the first similar time period and the annotated similar time period, and the connection discriminant network is trained by back propagation; Based on the training connection feature graph, a connection discrimination network is trained.
7. The method for analyzing user behavior of an E-SIM according to claim 6, It is characterized in that The step of training a connection discrimination network based on the training connection feature graph includes: Inputting the training connection feature graph into a similarity judgment network, judging similarity, and obtaining a second similar time period; The loss value is calculated using the second similar time period and the annotated similar time period, and the back propagation training connects the discriminant network.
8. The method for analyzing user behavior of an E-SIM according to claim 1, It is characterized in that The extracting features based on the application usage time period and the corresponding application usage traffic to obtain a traffic feature vector includes: Arrange the application usage traffic in sequence according to the corresponding application usage time periods to obtain a time traffic vector; The time flow vector is input into the time flow neural network to obtain flow characteristics.
9. The method for analyzing user behavior of an E-SIM according to claim 1, It is characterized in that The method for obtaining the second similar feature map is the same as the method for obtaining the first similar feature map.
10. An E-SIM user behavior analysis system, It is characterized in that include: Acquisition module: obtains user connection time period, application category, application usage time period and corresponding application usage traffic; The user connection time points shown represent the time periods during which the user uses a SIM card to connect to the network within 30 days; the application usage time periods represent the time periods during which the application corresponding to the application category is used by the user within 30 days; the application usage traffic represents the traffic used during the application usage time periods corresponding to the application category; Segmentation module: based on the user connection time period, segmentation is performed with a fixed time length, an image is constructed, and a two-dimensional connection graph is obtained; the two-dimensional connection graph represents the connection status of a fixed time length within 30 days; Connection identification module: through the connection identification network, based on the two-dimensional connection graph, the regularity of the time period when the user connects to the network is determined to obtain a connection feature graph; Traffic module: extracting features based on the application usage time period and the corresponding application usage traffic to obtain a traffic feature vector; Prediction module: input the connection feature graph and the traffic feature vector into the package detection network to obtain the SIM traffic package type.
Citation Information
Patent Citations
Method and device for recommending telephone traffic set
CN103905971A
Multi-factor embedded personalized package recommendation method based on deep neural network
CN110955828A
Internet card user loss prediction method and system based on user portraits
CN113962160A
Flow ticket segmentation method
CN116522177A
Model training method, information recommendation method and device, medium and electronic equipment
CN117196721A
Cited By
Subscriber identification system based on SIM card
CN119967415A
A user identification system based on SIM card
CN119967415B
E-SIM-based user behavior analysis method
CN121037829A
An e-sim-based user behavior analysis method
CN121037829B
