A SIM package recommendation method and system based on user behavior analysis

Abstract

The present invention discloses a SIM package recommendation method and system based on user behavior analysis. Based on a user portrait tree, the probability of traffic usage is judged to obtain a user behavior probability. Based on the user portrait tree, the user behavior probability and the application category probability, a network is constructed to obtain a user recommendation network. Through the user recommendation network, a recommended package is obtained based on the usage traffic and time traffic key-value pairs at multiple time points. A user portrait tree is constructed based on the application category, application program and time point of traffic usage by general users. A user recommendation network that satisfies user recommendations is constructed according to the specific structure of the user portrait tree. Thereby achieving the technical effect of recommending to users traffic packages that are not stuck and are in line with the software status of the users themselves.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and system for recommending SIM packages based on user behavior analysis. Background Art

[0002] At present, the application software in the mobile phone can be operated by obtaining traffic through SIM card, and the Internet can be used for video playback, text reading, etc. by taking advantage of its intuitive, fast, good presentation, rich content, strong interactivity, no geographical restrictions, and audience segmentation. However, each user has personalized content for obtaining data on the network. Only the total amount of fixed traffic and only 4G or 5G traffic packages require users to judge by themselves, resulting in traffic overflow, insufficient traffic, and freezes, which cannot meet user needs and cannot recommend satisfactory traffic packages to users. Summary of the invention

[0003] The purpose of the present invention is to provide a SIM package recommendation method and system based on user behavior analysis to solve the above problems existing in the prior art.

[0004] In a first aspect, an embodiment of the present invention provides a SIM package recommendation method based on user behavior analysis, including:

[0005] Obtain application usage time, application program and current traffic information; the current traffic information includes the usage traffic and transmission time length of multiple time points of the current traffic package; the usage traffic is the traffic used at each time point; the transmission time length is the time length during which the transmission user can receive complete data for viewing; the application usage time indicates the time during which the user uses the application in a day; the application program indicates the software of the application used by the user in a day;

[0006] Based on the application and the usage traffic at multiple time points, obtaining application category probability through an application classification network;

[0007] Based on the current traffic information, a relationship between the transmission time length and the used traffic is constructed to obtain a time-traffic key-value pair;

[0008] Based on application usage time, application program, application category and usage traffic, user behavior is judged to obtain a user portrait tree; the user portrait tree represents the hierarchical state of usage traffic;

[0009] Based on the user portrait tree, the probability of traffic usage is determined to obtain the user behavior probability;

[0010] Based on the user portrait tree, user behavior probability and application category probability, a network is constructed to obtain a user recommendation network;

[0011] The recommended packages are obtained through the user recommendation network based on the usage traffic and time-traffic key-value pairs at multiple time points.

[0012] Optionally, constructing a network based on the user portrait tree, user behavior probability and application category probability to obtain a user recommendation network includes:

[0013] According to the first layer of the user portrait tree, a feature extraction network is obtained;

[0014] According to the application category corresponding to the second layer of the user portrait tree, an application category probability structure, a first application category neural network and a second application category neural network are obtained; the application category probability structure stores application category probabilities;

[0015] According to the application corresponding to the third layer of the user portrait tree, a first program network and a self-detection structure are obtained;

[0016] Obtaining a first convolutional network according to the application time flow corresponding to the fourth layer of the user portrait tree;

[0017] Acquire a feature classification network; the feature classification network is used to classify users into suitable traffic packages according to feature information;

[0018] The feature extraction network, the application category probability structure, the first application category neural network, the second application category neural network, the first program network, the self-detection structure, the first convolutional network and the feature classification network are used to construct a user recommendation network.

[0019] Optionally, the step of constructing a user recommendation network by combining the feature extraction network, the application category probability structure, the first application category neural network, the second application category neural network, the first program network, the self-detection structure, the first convolutional network, and the feature classification network comprises:

[0020] The application category probability structure stores the application category probability; the self-detection structure stores the user behavior probability;

[0021] The input of the application category probability structure is the output of the feature extraction network;

[0022] The output of the application category probability structure is the input of the first application category neural network and the input of the second application category neural network;

[0023] The input of the first program network is the output of the first application category neural network and the output of the second application category neural network

[0024] The output of the first program network is the input of the self-detection structure;

[0025] The output of the self-detection structure and the output of the first convolutional network are input to the feature classification network.

[0026] Optionally, obtaining a recommended package based on usage traffic and time-traffic key-value pairs at multiple time points through the user recommendation network includes:

[0027] Inputting the usage flow at the multiple time points into a feature extraction network, extracting features, and obtaining a first flow feature;

[0028] Multiplying the first traffic feature by the application category probability stored in the application category probability structure respectively to obtain a first application usage traffic feature and a second application usage traffic feature;

[0029] Inputting the first application usage traffic feature into a first application category neural network to obtain a first application feature;

[0030] Inputting the second application usage traffic feature into a second application category neural network to obtain a second application feature;

[0031] Arranging the first application feature and the second application feature and inputting them into a first program network to obtain a first program feature;

[0032] Multiplying the first program feature by the user behavior probability stored in the self-detection structure respectively to obtain a plurality of user behavior flow features;

[0033] Inputting the first part of the curve image in the time-flow key-value pair into the first convolutional network to obtain a first time-flow feature;

[0034] The multiple user behavior traffic features and the first time traffic features are input into a feature classification network to obtain a recommended package.

[0035] Optionally, the user behavior is judged based on application usage time, application program, application category and usage traffic to obtain a user portrait tree, including:

[0036] Construct a tree structure with a depth of 4;

[0037] Adding a root node to the first level of the tree structure;

[0038] Adding nodes corresponding to the application categories to the second layer of the tree structure;

[0039] categorizing the application programs according to the application categories;

[0040] Adding the node corresponding to the application to the third layer of the tree structure;

[0041] According to the application usage time, the corresponding usage flow is divided to obtain a plurality of application time flows; the application time flow represents the usage flow of the application program at each time point during the application usage time;

[0042] The nodes corresponding to the application time flow are added to the fourth layer of the tree structure to obtain a user portrait tree.

[0043] Optionally, the relationship between the transmission time length and the used traffic is constructed based on the current traffic information to obtain a time-traffic key-value pair, including:

[0044] Normalize the usage flow and multiply it by a fixed value to obtain a normalized flow value; obtain multiple normalized flow values ​​corresponding to the usage flow at multiple time points;

[0045] Fitting multiple normalized flow values ​​to obtain a first flow curve;

[0046] Converting the first flow curve into an image to obtain a first flow curve image;

[0047] Based on the first traffic curve image and the transmission time length, a time-traffic key-value pair is obtained.

[0048] Optionally, obtaining a time-flow key-value pair based on the first flow curve image and the transmission time length includes:

[0049] Segmenting the first flow curve image according to the transmission time length to obtain a plurality of first partial curve images;

[0050] A plurality of transmission time lengths and the corresponding first part of the curve image are used to construct a key-value pair to obtain a time-flow key-value pair.

[0051] Optionally, obtaining the application category probability through an application classification network based on the application program and the usage traffic at multiple time points includes:

[0052] Classifying the usage traffic at the multiple time points according to application categories to obtain multiple application category traffic;

[0053] Adding the traffic of the multiple application categories to obtain the total traffic of the application category;

[0054] The multiple application category flows are divided respectively by the total application category flow to obtain the application category probability.

[0055] Optionally, judging the probability of traffic usage based on the user portrait tree to obtain the user behavior probability includes:

[0056] According to the user portrait tree, application traffic is obtained; the application traffic represents the usage traffic of the application in one day; and multiple application traffic is obtained corresponding to multiple applications;

[0057] Adding the traffic of the multiple applications to obtain a total traffic;

[0058] The traffic of the multiple application programs is divided by the total traffic respectively to obtain the user behavior probability.

[0059] In a second aspect, an embodiment of the present invention provides a SIM package recommendation system based on user behavior analysis, including:

[0060] Acquisition module: Acquisition of application usage time, application program and current traffic information; the current traffic information includes the usage traffic and transmission time length of multiple time points of the current traffic package; the usage traffic is the traffic used at each time point; the transmission time length is the time length during which the transmission user can receive complete data for viewing; the application usage time indicates the time during which the user uses the application in a day; the application program indicates the software of the application used by the user in a day;

[0061] Application category probability module: based on the application program and the usage traffic at multiple time points, obtains the application category probability through the application classification network;

[0062] Key-value pair module: based on the current traffic information, construct the relationship between the transmission time length and the used traffic to obtain the time-traffic key-value pair;

[0063] Tree structure module: based on application usage time, application program, application category and usage traffic, the user behavior is judged to obtain a user portrait tree; the user portrait tree represents the hierarchical status of usage traffic;

[0064] User behavior probability module: Based on the user portrait tree, the probability of traffic usage is determined to obtain the user behavior probability;

[0065] Network construction module: constructs a network based on the user portrait tree, user behavior probability and application category probability to obtain a user recommendation network;

[0066] Recommendation module: obtains recommended packages based on the usage traffic and time-traffic key-value pairs at multiple time points through the user recommendation network.

[0067] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0068] The embodiment of the present invention also provides a SIM package recommendation method and system based on user behavior analysis, the method comprising: obtaining application usage time, application program and current traffic information; the current traffic information comprises the usage traffic and transmission time length of multiple time points of the current traffic package; the usage traffic is the traffic used at each time point; the transmission time length is the time length for the transmission user to receive complete data for viewing; the application usage time represents the time when the user uses the application in a day; the application program represents the software of the application used in a day; based on the application program and the usage traffic at multiple time points, the application category probability is obtained through the application classification network; based on the current traffic information, the relationship between the transmission time length and the usage traffic is constructed to obtain the time traffic key-value pair; based on the application usage time, application program, application category and usage traffic, the user behavior is judged to obtain the user portrait tree; the user portrait tree represents the hierarchical state of the usage traffic; based on the user portrait tree, the probability of traffic usage is judged to obtain the user behavior probability; based on the user portrait tree, the user behavior probability and the application category probability, a network is constructed to obtain the user recommendation network; through the user recommendation network, the recommended package is obtained based on the usage traffic and the time traffic key-value pair at multiple time points.

[0069] In the present invention, a user portrait tree is constructed by the application categories, applications and time points of the traffic used by general users. A user recommendation network that meets the user's recommendation is constructed according to the specific structure of the user portrait tree. In addition, the application category probability structure and the self-detection structure are added to construct a user recommendation network with characteristics according to the characteristics of each user. In the user recommendation network, not only multiple fully connected networks are used to detect the features that meet the characteristics of each user in the traffic used at multiple time points, but also the first convolutional network is used to judge the state of the curve composed of the traffic used at different time points under the transmission time length, and the jamming of the video corresponding to the high rate at a certain traffic flow is used to predict the traffic situation without jamming. Thereby achieving the technical effect of recommending to the user a traffic package that is not jammed and meets the software usage status of the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 This is a flow chart of a SIM package recommendation method based on user behavior analysis provided by an embodiment of the present invention

[0071] Figure 2 It is a structural schematic diagram of a user recommendation network in a SIM package recommendation method based on user behavior analysis provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0072] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0073] Example 1

[0074] like Figure 1 As shown, an embodiment of the present invention provides a SIM package recommendation method based on user behavior analysis, the method comprising:

[0075] S101: Obtain application usage time, application program and current traffic information; the current traffic information includes the usage traffic and transmission time length of multiple time points of the current traffic package; the usage traffic is the traffic used at each time point; the transmission time length is the time length for the transmission user to receive complete data for viewing; the application usage time indicates the time the user uses the application in a day; the application program indicates the software of the application used by the user in a day.

[0076] In this embodiment, 1 second is regarded as 1 time point.

[0077] The transmission time length includes a specific start point and an end point, and the corresponding time point can be found for cutting.

[0078] S102: Based on the application program and the usage traffic at multiple time points, an application category probability is obtained through an application classification network.

[0079] The application category probability represents the probability of a user using traffic under one application category.

[0080] S103: Based on the current traffic information, a relationship between the transmission time length and the used traffic is constructed to obtain a time-traffic key-value pair.

[0081] S104: Based on application usage time, application program, application category and usage traffic, user behavior is identified to obtain a user portrait tree; the user portrait tree represents a hierarchical state of usage traffic.

[0082] S105: Based on the user portrait tree, determine the probability of traffic usage and obtain the user behavior probability.

[0083] The user behavior probability represents the probability of a user using traffic under one application.

[0084] S106: Based on the user portrait tree, user behavior probability and application category probability, a network is constructed to obtain a user recommendation network.

[0085] The structural diagram of the user recommendation network is as follows: Figure 2 shown.

[0086] S107: Obtain a recommended package through the user recommendation network based on the usage traffic and time traffic key-value pairs at multiple time points.

[0087] Among them, the traffic packages include 4G or 5G packages, directional traffic packages of the video platform corresponding to the application, etc.

[0088] Optionally, constructing a network based on the user portrait tree, user behavior probability and application category probability to obtain a user recommendation network includes:

[0089] According to the first layer of the user portrait tree, a feature extraction network is obtained.

[0090] Wherein, the feature extraction network is a fully-connected neural network (FCNN).

[0091] According to the application category corresponding to the second layer of the user portrait tree, an application category probability structure, a first application category neural network and a second application category neural network are obtained; the application category probability structure stores application category probabilities;

[0092] The first application category neural network is a fully connected neural network (Fully-connected neural network, FCNN), and the second application category neural network is a fully connected neural network (Fully-connected neural network, FCNN). The number of neurons in the input layer of the first application category neural network is equal to the number of neurons in the input layer of the second application category neural network, and the number of neurons in the output layer of the feature extraction network is equal to the number of neurons in the output layer of the feature extraction network.

[0093] The application categories are divided into two categories: video and text.

[0094] According to the application corresponding to the third layer of the user portrait tree, a first program network and a self-detection structure are obtained; the user behavior probability is stored in the self-detection structure.

[0095] The first program network is a fully-connected neural network (FCNN). The number of neurons in the input layer of the first program network is equal to the sum of the number of neurons in the input layer of the first application category neural network and the number of neurons in the input layer of the second application category neural network.

[0096] A first convolutional network is obtained according to the application time flow corresponding to the fourth layer of the user portrait tree.

[0097] Among them, in this embodiment, the first convolutional network is a convolutional neural network (Convolutional Neural Networks, CNN).

[0098] Obtain a feature classification network; the feature classification network is used to classify users into suitable traffic packages according to feature information.

[0099] The feature classification network is a fully-connected neural network (FCNN). The number of neurons in the input layer of the feature classification network is equal to the product of the number of neurons in the input layer of the first application category neural network multiplied by the number of probabilities in the self-detection structure.

[0100] The feature extraction network, the application category probability structure, the first application category neural network, the second application category neural network, the first program network, the self-detection structure, the first convolutional network and the feature classification network are used to construct a user recommendation network.

[0101] Optionally, the step of constructing a user recommendation network by combining the feature extraction network, the application category probability structure, the first application category neural network, the second application category neural network, the first program network, the self-detection structure, the first convolutional network, and the feature classification network comprises:

[0102] The input of the application category probability structure is the output of the feature extraction network.

[0103] The output of the application category probability structure is the input of the first application category neural network and the input of the second application category neural network.

[0104] The input of the first program network is the output of the first application category neural network and the output of the second application category neural network.

[0105] The output of the first program network is the input of the self-detection structure.

[0106] The output of the self-detection structure and the output of the first convolutional network are input to the feature classification network.

[0107] Optionally, obtaining a recommended package based on usage traffic and time-traffic key-value pairs at multiple time points through the user recommendation network includes:

[0108] The usage flow at the multiple time points is input into a feature extraction network to extract features to obtain a first flow feature.

[0109] The first traffic feature is a feature of traffic distribution at multiple time points.

[0110] Multiplying the first traffic feature by the application category probability stored in the application category probability structure respectively to obtain a first application usage traffic feature and a second application usage traffic feature;

[0111] Inputting the first application usage traffic feature into a first application category neural network to obtain a first application feature;

[0112] Inputting the second application usage traffic feature into a second application category neural network to obtain a second application feature;

[0113] Arranging the first application feature and the second application feature and inputting them into a first program network to obtain a first program feature;

[0114] Multiplying the first program feature by the user behavior probability stored in the self-detection structure respectively to obtain a plurality of user behavior flow features;

[0115] The first part of the curve image in the time-flow key-value pair is input into the first convolutional network to obtain a first time-flow feature.

[0116] The first time-traffic feature indicates the relationship between time and traffic for receiving a complete and viewable data without causing data freeze.

[0117] The multiple user behavior traffic features and the first time traffic features are input into a feature classification network to obtain a recommended package.

[0118] Optionally, the user behavior is judged based on application usage time, application program, application category and usage traffic to obtain a user portrait tree, including:

[0119] Construct a tree structure with a depth of 4;

[0120] A root node is added to the first level of the tree structure.

[0121] The nodes corresponding to the application categories are added to the second layer of the tree structure.

[0122] In this embodiment, the application categories are divided into two categories: text and video. The second layer has two nodes.

[0123] The application programs are classified according to the application categories.

[0124] Among them, there are 3 applications belonging to the video application category, and there are 3 applications belonging to the text application category.

[0125] The nodes corresponding to the application are added to the third layer of the tree structure.

[0126] Among them, three nodes are used as child nodes of the node corresponding to the second-layer video category, and two nodes are used as child nodes of the node corresponding to the second-layer text category.

[0127] According to the application usage time, the corresponding usage traffic is divided to obtain a plurality of application time traffics; the application time traffic represents the usage traffic of the application program at each time point during the application usage time.

[0128] If there are 60 time points in the application, the usage flow corresponding to the 60 time points is obtained as the application time flow.

[0129] The nodes corresponding to the application time flow are added to the fourth layer of the tree structure to obtain a user portrait tree.

[0130] Among them, 60 nodes are used as child nodes of the node corresponding to the corresponding application.

[0131] In this embodiment, the value of the node in the user portrait tree is 1, and then the structure of the user portrait tree is used without using its value.

[0132] Optionally, the relationship between the transmission time length and the used traffic is constructed based on the current traffic information to obtain a time-traffic key-value pair, including:

[0133] The used traffic is normalized and then multiplied by a fixed value to obtain a normalized traffic value; the used traffic at multiple time points corresponds to multiple normalized traffic values.

[0134] In this embodiment, the fixed value is 100.

[0135] A plurality of normalized flow values ​​are fitted to obtain a first flow curve.

[0136] In this embodiment, polynomial fitting is used for fitting.

[0137] The first flow curve is converted into an image to obtain a first flow curve image.

[0138] The length of the first flow curve image is the horizontal coordinate, the width of the first flow curve image is the vertical coordinate, the lower left corner of the first flow curve image is the origin, and the first flow curve is projected onto the first flow curve image.

[0139] The length of the first flow curve image represents a time point, and the width of the first flow curve image represents a usage flow.

[0140] Based on the first traffic curve image and the transmission time length, a time-traffic key-value pair is obtained.

[0141] Optionally, obtaining a time-flow key-value pair based on the first flow curve image and the transmission time length includes:

[0142] The first curve is divided according to the transmission time length to obtain a plurality of first partial curve images.

[0143] A key-value pair is constructed by combining the multiple transmission time lengths with the corresponding first part of the curve image to obtain a time-flow key-value pair.

[0144] Among them, the transmission time length is the key, and the first part of the curve image is the corresponding value.

[0145] Optionally, obtaining the application category probability through an application classification network based on the application program and the usage traffic at multiple time points includes:

[0146] The usage traffic at the multiple time points is classified according to application categories to obtain multiple application category traffic.

[0147] In this embodiment, the application types are classified according to the fact that the interfaces of the data displayed in the application structure are mostly video or text. The usage flow at multiple time points corresponding to multiple applications corresponding to one application category is added together to obtain the application category flow.

[0148] Adding the traffic of the multiple application categories to obtain the total traffic of the application category;

[0149] The multiple application category flows are divided respectively by the total application category flow to obtain the application category probability.

[0150] Optionally, judging the probability of traffic usage based on the user portrait tree to obtain the user behavior probability includes:

[0151] According to the user portrait tree, application traffic is obtained; the application traffic represents the usage traffic of the application in one day; and multiple application traffic is obtained corresponding to multiple applications.

[0152] The application traffic is the sum of the application time traffic corresponding to multiple subnodes corresponding to one application node.

[0153] The traffic of the multiple applications is added together to obtain the total traffic.

[0154] The traffic of the multiple application programs is divided by the total traffic respectively to obtain the user behavior probability.

[0155] Example 2

[0156] Based on the above-mentioned SIM package recommendation method based on user behavior analysis, an embodiment of the present invention further provides a SIM package recommendation system based on user behavior analysis, the system comprising:

[0157] The acquisition module is used to obtain application usage time, application program and current traffic information; the current traffic information includes the usage traffic and transmission time length of multiple time points of the current traffic package; the usage traffic is the traffic used at each time point; the transmission time length is the time length during which the transmission user can receive complete data for viewing; the application usage time indicates the time the user uses the application in a day; the application program indicates the software of the application used by the user in a day.

[0158] The application category probability module is used to obtain the application category probability through the application classification network based on the application program and the usage traffic at multiple time points.

[0159] The key-value pair module is used to construct a relationship between the transmission time length and the used traffic based on the current traffic information to obtain a time-traffic key-value pair.

[0160] The tree structure module is used to identify user behavior based on application usage time, application program, application category and usage traffic to obtain a user portrait tree; the user portrait tree represents the hierarchical status of usage traffic.

[0161] The user behavior probability module is used to determine the probability of traffic usage based on the user portrait tree and obtain the user behavior probability.

[0162] The network construction module is used to construct a network based on the user portrait tree, user behavior probability and application category probability to obtain a user recommendation network.

[0163] The recommendation module is used to obtain recommended packages based on the usage traffic and time-traffic key-value pairs at multiple time points through the user recommendation network.

Claims

1. A SIM package recommendation method based on user behavior analysis, characterized in that: include: Get application usage time, application programs and current traffic information; The current traffic information includes the usage traffic and transmission time length of the current traffic package at multiple time points; The traffic usage refers to the traffic usage at each time point; the transmission time refers to the time during which the transmission user can receive complete data for viewing; the application usage time refers to the time during which the user uses the application in a day; the application refers to the software of the application used by the user in a day; Based on the application and the usage traffic at multiple time points, obtaining the application category probability through the application classification network; the obtaining the application category probability based on the application and the usage traffic at multiple time points through the application classification network includes: Classifying the usage traffic at the multiple time points according to application categories to obtain multiple application category traffic; Adding the traffic of the multiple application categories to obtain the total traffic of the application category; Dividing the traffic of the multiple application categories by the total traffic of the application category respectively, to obtain the probability of the application category; wherein the application category is divided into two categories: video and text; Based on the current traffic information, a relationship between the transmission time length and the used traffic is constructed to obtain a time-traffic key-value pair; Based on application usage time, application program, application category and usage traffic, user behavior is judged to obtain a user portrait tree; the user portrait tree represents the hierarchical state of usage traffic; The user behavior is judged based on the application usage time, application program, application category and usage traffic to obtain a user portrait tree, including: Construct a tree structure with a depth of 4; Adding a root node to the first level of the tree structure; Adding nodes corresponding to the application categories to the second layer of the tree structure; categorizing the application programs according to the application categories; Adding the node corresponding to the application to the third layer of the tree structure; According to the application usage time, the corresponding usage flow is divided to obtain a plurality of application time flows; the application time flow represents the usage flow of the application program at each time point during the application usage time; Adding the nodes corresponding to the application time flow into the fourth layer of the tree structure to obtain a user portrait tree; Based on the user portrait tree, the probability of traffic usage is determined to obtain the user behavior probability; The method of determining the probability of traffic usage based on the user portrait tree and obtaining the user behavior probability includes: According to the user portrait tree, application traffic is obtained; the application traffic represents the usage traffic of the application in one day; and multiple application traffic is obtained corresponding to multiple applications; Adding the traffic of the multiple applications to obtain a total traffic; Dividing the traffic of the multiple application programs by the total traffic to obtain a user behavior probability; Based on the user portrait tree, user behavior probability and application category probability, a network is constructed to obtain a user recommendation network; The step of constructing a network based on the user portrait tree, user behavior probability and application category probability to obtain a user recommendation network includes: According to the first layer of the user portrait tree, a feature extraction network is obtained; According to the application category corresponding to the second layer of the user portrait tree, an application category probability structure, a first application category neural network and a second application category neural network are obtained; the application category probability structure stores application category probabilities; According to the application corresponding to the third layer of the user portrait tree, a first program network and a self-detection structure are obtained; the user behavior probability is stored in the self-detection structure; Obtaining a first convolutional network according to the application time flow corresponding to the fourth layer of the user portrait tree; Acquire a feature classification network; the feature classification network is used to classify users into suitable traffic packages according to feature information; The feature extraction network, the application category probability structure, the first application category neural network, the second application category neural network, the first program network, the self-detection structure, the first convolutional network and the feature classification network are used to construct a user recommendation network; Obtaining recommended packages based on the usage traffic and time traffic key-value pairs at multiple time points through the user recommendation network; The method of obtaining a recommended package based on the usage traffic and time traffic key-value pairs at multiple time points through the user recommendation network includes: Inputting the usage flow at the multiple time points into a feature extraction network, extracting features, and obtaining a first flow feature; Multiplying the first traffic feature by the application category probability stored in the application category probability structure respectively to obtain a first application usage traffic feature and a second application usage traffic feature; Inputting the first application usage traffic feature into a first application category neural network to obtain a first application feature; Inputting the second application usage traffic feature into a second application category neural network to obtain a second application feature; Arranging the first application feature and the second application feature and inputting the resultant into a first program network to obtain a first program feature; Multiplying the first program feature by the user behavior probability stored in the self-detection structure respectively to obtain a plurality of user behavior flow features; Inputting the first part of the curve image in the time-flow key-value pair into the first convolutional network to obtain a first time-flow feature; The multiple user behavior traffic features and the first time traffic features are input into a feature classification network to obtain a recommended package.

2. A SIM package recommendation method based on user behavior analysis according to claim 1, characterized in that: The method of constructing a user recommendation network by combining the feature extraction network, the application category probability structure, the first application category neural network, the second application category neural network, the first program network, the self-detection structure, the first convolutional network and the feature classification network comprises: The input of the application category probability structure is the output of the feature extraction network; The output of the application category probability structure is the input of the first application category neural network and the input of the second application category neural network; The input of the first program network is the output of the first application category neural network and the output of the second application category neural network; The output of the first program network is the input of the self-detection structure; The output of the self-detection structure and the output of the first convolutional network are input to the feature classification network.

3. The method for recommending SIM packages based on user behavior analysis according to claim 1, characterized in that: The relationship between the transmission time length and the used traffic is constructed based on the current traffic information to obtain a time-traffic key-value pair, including: Normalize the usage flow and multiply it by a fixed value to obtain a normalized flow value; obtain multiple normalized flow values ​​corresponding to the usage flow at multiple time points; Fitting multiple normalized flow values ​​to obtain a first flow curve; Converting the first flow curve into an image to obtain a first flow curve image; Based on the first traffic curve image and the transmission time length, a time-traffic key-value pair is obtained.

4. A SIM package recommendation method based on user behavior analysis according to claim 3, characterized in that: The obtaining of a time-flow key-value pair based on the first flow curve image and the transmission time length includes: Segmenting the first flow curve image according to the transmission time length to obtain a plurality of first partial curve images; A plurality of transmission time lengths and the corresponding first part of the curve image are used to construct a key-value pair to obtain a time-flow key-value pair.

5. A SIM package recommendation system based on user behavior analysis, characterized in that: include: Acquisition module: obtain application usage time, application program and current traffic information; The current traffic information includes the usage traffic and transmission time length of the current traffic package at multiple time points; The traffic usage refers to the traffic usage at each time point; the transmission time refers to the time during which the transmission user can receive complete data for viewing; the application usage time refers to the time during which the user uses the application in a day; the application refers to the software of the application used by the user in a day; Application category probability module: based on the application program and the usage traffic at multiple time points, obtains the application category probability through the application classification network; The obtaining of application category probability based on the application program and the usage traffic at multiple time points through an application classification network includes: Classifying the usage traffic at the multiple time points according to application categories to obtain multiple application category traffic; Adding the traffic of the multiple application categories to obtain the total traffic of the application category; Dividing the traffic of the multiple application categories by the total traffic of the application category respectively, to obtain the probability of the application category; wherein the application category is divided into two categories: video and text; Key-value pair module: based on the current traffic information, construct the relationship between the transmission time length and the used traffic to obtain the time-traffic key-value pair; Tree structure module: based on application usage time, application program, application category and usage traffic, the user behavior is judged to obtain a user portrait tree; the user portrait tree represents the hierarchical status of usage traffic; The user behavior is judged based on the application usage time, application program, application category and usage traffic to obtain a user portrait tree, including: Construct a tree structure with a depth of 4; Adding a root node to the first level of the tree structure; Adding nodes corresponding to the application categories to the second layer of the tree structure; categorizing the application programs according to the application categories; Adding the node corresponding to the application to the third layer of the tree structure; According to the application usage time, the corresponding usage flow is divided to obtain a plurality of application time flows; the application time flow represents the usage flow of the application program at each time point during the application usage time; Adding the nodes corresponding to the application time flow into the fourth layer of the tree structure to obtain a user portrait tree; User behavior probability module: based on the user portrait tree, the probability of traffic usage is determined to obtain the user behavior probability; the method of determining the probability of traffic usage based on the user portrait tree to obtain the user behavior probability includes: According to the user portrait tree, application traffic is obtained; the application traffic represents the usage traffic of the application in one day; and multiple application traffic is obtained corresponding to multiple applications; Adding the traffic of the multiple applications to obtain a total traffic; Dividing the traffic of the multiple application programs by the total traffic to obtain a user behavior probability; Network construction module: constructs a network based on the user portrait tree, user behavior probability and application category probability to obtain a user recommendation network; The step of constructing a network based on the user portrait tree, user behavior probability and application category probability to obtain a user recommendation network includes: According to the first layer of the user portrait tree, a feature extraction network is obtained; According to the application category corresponding to the second layer of the user portrait tree, an application category probability structure, a first application category neural network and a second application category neural network are obtained; the application category probability structure stores application category probabilities; According to the application corresponding to the third layer of the user portrait tree, a first program network and a self-detection structure are obtained; the user behavior probability is stored in the self-detection structure; Obtaining a first convolutional network according to the application time flow corresponding to the fourth layer of the user portrait tree; Acquire a feature classification network; the feature classification network is used to classify users into suitable traffic packages according to feature information; The feature extraction network, the application category probability structure, the first application category neural network, the second application category neural network, the first program network, the self-detection structure, the first convolutional network and the feature classification network are used to construct a user recommendation network; Recommendation module: obtains recommended packages based on the usage traffic and time traffic key-value pairs at multiple time points through the user recommendation network; The method of obtaining a recommended package based on the usage traffic and time traffic key-value pairs at multiple time points through the user recommendation network includes: Inputting the usage flow at the multiple time points into a feature extraction network, extracting features, and obtaining a first flow feature; Multiplying the first traffic feature by the application category probability stored in the application category probability structure respectively to obtain a first application usage traffic feature and a second application usage traffic feature; Inputting the first application usage traffic feature into a first application category neural network to obtain a first application feature; Inputting the second application usage traffic feature into a second application category neural network to obtain a second application feature; Arranging the first application feature and the second application feature and inputting the resultant into a first program network to obtain a first program feature; Multiplying the first program feature by the user behavior probability stored in the self-detection structure respectively to obtain a plurality of user behavior flow features; Inputting the first part of the curve image in the time-flow key-value pair into the first convolutional network to obtain a first time-flow feature; The multiple user behavior traffic features and the first time traffic features are input into a feature classification network to obtain a recommended package.

Images