Internet of things card package configuration method and system based on data analysis

Abstract

The invention discloses an Internet of Things card package configuration method and system based on data analysis, and relates to the field of Internet of Things card package management, and the technical scheme is characterized in that the method comprises the steps: collecting comprehensive user information in a period time, carrying out the format conversion of the comprehensive user information, and obtaining standard comprehensive data; using an improved multi-feature K-means technology to analyze the standard comprehensive data of the users, grouping the Internet of Things card users, and obtaining multiple groups of data sets; constructing and using the sample set to train a package personality design model, importing a plurality of groups of real-time data sets into the trained package personality design model, and obtaining a real-time Internet of Things card package scheme; visual interaction is carried out on the internet of things card package scheme, and personalized configuration of the internet of things card package is realized through data analysis.

Description

Technical Field

[0001] The present invention relates to the field of Internet of Things card package management, and more specifically, to an Internet of Things card package configuration method and system based on data analysis. Background Art

[0002] Faced with massive amounts of user data, user information is often scattered across multiple systems, with inconsistent data formats (such as text, numbers, etc.), making it difficult to integrate and utilize. A large amount of data contains a wide variety of data categories, resulting in more and more factors affecting package configuration. Based on a single feature or simple rules, it is easy to cause unreasonable configuration, which makes it difficult to accurately characterize user differences when configuring packages due to traditional user segmentation. Moreover, while the rich data increases the reference value of package design, it also brings inevitable obstacles to data analysis, processing, and in-depth mining.

[0003] In addition, in various IoT card packages, traditional package design usually adopts a "one-size-fits-all" strategy, which is often based on fixed templates, lacks personalized considerations, and cannot meet the specific needs of different users. User feedback is often ignored or not handled in a timely manner, affecting service quality.

[0004] Therefore, based on the above problems, it is necessary to design the IoT card package configuration based on data analysis. Summary of the invention

[0005] In view of the problems existing in the prior art, the purpose of the present invention is to provide an Internet of Things card package configuration method and system based on data analysis, so as to realize personalized configuration of the Internet of Things card package through data analysis.

[0006] To achieve the above object, the present invention provides the following technical solution: The Internet of Things card package configuration method based on data analysis includes: Step S1: Collect comprehensive user information within a period of time, and perform format conversion processing on the comprehensive user information to obtain standard comprehensive data; Step S2: using the improved multi-feature K-means technology to analyze the standard comprehensive data of users, grouping IoT card users, and obtaining multiple data sets; Step S3: construct and use the sample set to train the package personalized design model, import the real-time multiple data sets into the trained package personalized design model, and obtain the real-time IoT card package plan; Step S4: Visualize and interact with the IoT card package plan.

[0007] Preferably, the comprehensive user information includes the user's basic information and behavior preference data, the usage behavior data and package billing data of the Internet of Things card, and the user's evaluation feedback data; The user's basic information includes identity information, device information, and geographic location information; Behavioral preference data includes package query frequency, app resource usage ratio, package change frequency and additional service data; The usage behavior data of IoT cards includes traffic usage data, voice usage data, and SMS usage data; Package billing data includes the total consumption amount of the package and the consumption amount corresponding to each usage behavior data; User evaluation feedback data refers to the user's satisfaction with the package.

[0008] Preferably, the method for performing format conversion processing on the comprehensive user information includes: Step A1: construct a blank dictionary, set the initial value, use Python to generate different arrays from the initial value to form a data set, identify the arrays in the comprehensive user information, delete the arrays in the data set that are identical to it, fill the deleted data set into the blank dictionary, and arrange the arrays in order of size to form a conversion dictionary; Step A2: extracting text data from the comprehensive user information, removing irrelevant symbols, using a sentence segmentation tool to segment sentences in the text, using a word segmentation tool to segment sentences into words, calculating the weight index of each word, arranging each word according to the weight index, and the arrangement order is the same as the array order in the conversion dictionary, and converting the text data into numerical data; No. The weight index of the word ; in, Indicates Words In the Text The proportion of Indicates the total number of texts. Indicates that it contains Words The number of texts, Indicates Words The importance level, and represents the adjustment factor; Step A3: Collect all numerical data, fill missing values, replace outliers and standardize data to form standard comprehensive data.

[0009] Preferably, the method of constructing a blank dictionary includes: Set up a Each column is mapped to a unique key to form a key-cell pair, which is used to index the array in each column; After the array is filled, the blank columns that are not filled in the array are deleted; in, Indicates the length of the blank column. Indicates the width of the blank column. Indicates the level of the blank grid column.

[0010] Preferably, the method of analyzing standard comprehensive data using the improved multi-feature K-means technology comprises: Step D1: preset the K value and initialize K cluster centers using the probability feature method; Step D2: Calculate the distance between each user's standard comprehensive data and the cluster center, and assign the standard comprehensive data to the cluster center with the nearest distance, dividing the users into K groups; Step D3: Identify the number of users in each group. If the number of users is less than the limit threshold, release the users in the group and delete the cluster center corresponding to the group. Recalculate the distance between the released users and the remaining cluster centers and reallocate them to obtain the updated group K_1 into which the users are divided. Step D4: extract the standard comprehensive data of each user in the K_1 group and update the weight of each data category in each group; Step D5: Calculate the standard comprehensive data of all users in each group using weighted average, i.e. , collect the means of all data categories and update the cluster centers; in, Indicates the first The mean of the data categories, Indicates the number of users in the group. Indicates the first User about The weight of each data category, In the group User about data values ​​for each data category, The index of the user; Step D6: Repeat steps D2 to D6 until the change of all cluster centers is less than a preset threshold, and then obtain users divided into K_1 groups to form multiple data sets.

[0011] Preferably, the method of initializing K cluster centers using a probabilistic feature method includes: Step W1: Calculate the cumulative value of the standard comprehensive data of each user, arrange them in order, and extract the standard comprehensive data corresponding to the median of the cumulative value as the first cluster center; Step W2: Calculate the distance from each user's standard comprehensive data to the first cluster center ,in, Indicates The numerical value corresponding to each data category, Indicates the first cluster center The numerical value corresponding to each data category, is the index of the data category; Step W3: Calculate the probability of each user's standard comprehensive data being selected as the next cluster center ,in, Indicates The distance from the standard comprehensive data of each user to the first cluster center, Indicates The distance from the standard comprehensive data of each user to the first cluster center, The index of the user; Step W4: Use the roulette algorithm to select the next cluster center according to the probability of each standard comprehensive data being selected as the next cluster center; Step W5: Repeat steps W2 to W4 until K-1 cluster centers are selected, and collect the first cluster center and K-1 cluster centers as the initialized K cluster centers.

[0012] Preferably, updating the weight of each data category in each group includes: Count the number of values ​​of each data category in each group and calculate the probability of each data category in the group ,in, Indicates the first The number of values ​​for each data category, The total number of values ​​in the group. Indicates the first The probability of each data category; Calculate the weight of each data category in the group ,in, is a positive number, which updates the weight of each data category in each group.

[0013] Preferably, the method of constructing and using the sample set to train the package personalized design model includes: The framework of the package personalized design model is based on neural network learning technology, including input layer, hidden layer and output layer. The input of the input layer is set as multiple sets of data sets, and the output layer sets P_C output folders in parallel. Each output folder is set with 4 task heads, namely traffic prediction head, phone prediction head, information prediction head and additional service prediction head. P_C output folders represent P_C configuration schemes designed for users. P_C configuration schemes are collected together to form IoT card package schemes. The four task heads represent the specific designs of traffic, phone calls, information and additional services in each configuration scheme. The traffic prediction header, phone prediction header, and information prediction header use a linear activation function and set the loss function to mean square error, which is used to measure the difference between the predicted value and the actual corresponding value. The additional service prediction header uses a ReLU activation function and sets the loss function to multi-class cross entropy loss, which is used to calculate the error between the predicted additional service label and the actual corresponding label. The total loss function of the package personalized design model is set to the sum of the loss functions of each task head; The optimization goal of the package personalization design model is to minimize user costs and maximize operator revenue; The sample sets are input into the personalized package design model in sequence. An adaptive learning rate strategy is used. The Adam or SGD optimizer is adopted to output the loss function of each task head in sequence. The training is stopped when the total loss function value no longer changes or the maximum number of iterations is reached to obtain a trained personalized package design model.

[0014] Preferably, the setting of the sample set includes: The sample set includes T_U groups of historical samples. Each group of samples includes multiple sets of user data sets and corresponding IoT card package plans. The timestamp of each sample is extracted and the samples are split according to quarters or months. The samples before the split are full samples, and the samples after the split are periodic samples. When using the sample set to train the package personalized design model, it is preferred to use the periodic samples to pre-train the model, and then use the full samples to continue training, thereby obtaining the trained package personalized design model.

[0015] An Internet of Things card package configuration system based on data analysis, comprising: Data collection and processing module: used to collect comprehensive user information within a period of time, and perform format conversion processing on the comprehensive user information to obtain standard comprehensive data; Feature clustering module: Based on standard comprehensive data, the improved multi-feature K-means technology is used to group IoT card users and obtain multiple data sets; Personalized design module: Build and train a personalized package design model, import multiple sets of real-time data sets into the trained personalized package design model, and obtain real-time IoT card package solutions; Interface interaction module: interact with operators through a visual interface.

[0016] Compared with the prior art, the present invention has the following beneficial effects: Through automated data collection and preprocessing, data in different formats (such as text, numbers, etc.) are uniformly converted and standardized, solving the problems of data heterogeneity and large data volume, laying the foundation for subsequent analysis, improving data processing efficiency, reducing labor costs, and providing the possibility for large-scale user analysis.

[0017] By adopting the improved K-means clustering algorithm, taking into account multiple features, optimizing the cluster center through probabilistic methods, and dynamically adjusting according to the number of users in the group, a more fine-grained and accurate user grouping is achieved. The diverse characteristics of users are mined, a more detailed user group portrait is achieved, and user behavior and needs are better understood, providing data support for future business decisions, enhancing data-driven decision-making capabilities, and providing support for personalized services.

[0018] By building a multi-task learning neural network model, the parallel output clips correspond to different configuration schemes, and each scheme contains multiple prediction tasks. It can provide personalized package configuration combinations according to the characteristics of different users. It provides flexible and diverse package configurations to meet the personalized needs of different users and improve user experience. Through user grouping and personalized package design, operators can implement more accurate marketing strategies and improve market competitiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention proposes a structural schematic diagram of an Internet of Things card package configuration system based on data analysis; Figure 2 It is a schematic diagram of the method applied to the Internet of Things card package configuration system based on data analysis in the present invention. DETAILED DESCRIPTION

[0020] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here. Example

[0021] Reference Figure 1 and Figure 2 Embodiment 1 further illustrates an Internet of Things card package configuration system based on data analysis proposed by the present invention.

[0022] Step S1: Collect comprehensive user information within a period of time, and perform format conversion processing on the comprehensive user information to obtain standard comprehensive data; Comprehensive user information includes basic user information and behavior preference data, IoT card usage behavior data and package billing data, as well as user evaluation feedback data; The period of time can be one day, one month or one year, etc. The following data can be extracted through web collection equipment or crawler tools.

[0023] The user's basic information includes identity information, device information, and geographic location information; User identity information: including the user's identity type (individual user or corporate user), age, gender, etc. (if it involves an individual user). For corporate users, it may also include information such as industry type and corporate size.

[0024] Device information: the type of IoT device used by the user (such as sensors, smart home devices, vehicle-mounted devices, etc.), device brand, device model, number of device connections, etc.

[0025] Geographic location information: The user's geographical area, such as country, city, and even precise location information (such as the connection location of a base station), which helps analyze the user's geographical distribution and usage habits.

[0026] Behavioral preference data includes package query frequency, app resource usage ratio, package change frequency and additional service data; Package query frequency refers to the user's checking of package usage within a period of time, such as checking traffic volume, call duration, etc.

[0027] App usage ratio refers to the percentage of resources used by each app in the package resources.

[0028] The package change frequency refers to the number of times a user's package is changed or modified within a certain period of time (or cycle time).

[0029] Additional service data refers to the user's usage of additional services, such as data overlay packages, voice packages, international roaming packages, SMS reminders, and other services outside the package.

[0030] The usage behavior data of IoT cards includes traffic usage data, voice usage data, and SMS usage data; Traffic usage data can be average monthly traffic (total monthly traffic consumption), daily traffic consumption (traffic usage on each day, which can help analyze user usage habits), peak hour traffic (traffic usage during specific periods, such as daytime or nighttime traffic consumption) and traffic usage scenarios (such as video traffic, sensor data transmission traffic, system update data traffic, etc.).

[0031] Voice usage data includes voice call duration, call frequency (number of calls and their distribution), and call time distribution within a specific period such as monthly, daily or weekly average.

[0032] The SMS data may be SMS usage, that is, the number of SMS messages sent and received per month.

[0033] Package billing data includes the total consumption amount of the package and the consumption amount corresponding to each usage behavior data; The consumption amount corresponding to each usage behavior data includes the specific consumption amount of traffic, voice, text messages, etc. In addition, whether the user has arrears or unpaid bill transfer.

[0034] User evaluation feedback data refers to the user's satisfaction with the package.

[0035] Feedback on user satisfaction with the current package (such as collected through customer service channels, user surveys, etc.). Trends and inclinations in changing packages and operators, feedback on network signal quality, such as the network signal quality in the user's area (such as 4G / 5G signal strength, network stability), and whether users have complaints or feedback data about network connection problems (such as drop rate, delay, etc.).

[0036] The method for format conversion processing of comprehensive user information includes: Step A1: construct a blank dictionary, set the initial value, use Python to generate different arrays from the initial value to form a data set, identify the arrays in the comprehensive user information, delete the arrays in the data set that are identical to it, fill the deleted data set into the blank dictionary, and arrange the arrays in order of size to form a conversion dictionary; The initial value can be 1, 2, 3, 4, 5, 6, 7, 8, 9, 0, etc. Python can flexibly generate various forms of arrays or number combinations. Arrays can also be said to be number combinations, for example, 1111, 11, 13211, etc.

[0037] Step A2: extracting text data from the comprehensive user information, removing irrelevant symbols, using a sentence segmentation tool to segment sentences in the text, using a word segmentation tool to segment sentences into words, calculating the weight index of each word, arranging each word according to the weight index, and the arrangement order is the same as the array order in the conversion dictionary, and converting the text data into numerical data; The sentence segmentation tool can be a regular expression, which is used to segment sentences in the text. The word segmentation tool can be the jieba library, which is used to segment the individual words in the sentence. You can also use the natural language processing library (NLTK or spaCy) to segment sentences and words. The weight index of the word ; in, Indicates Words In the Text The proportion of Indicates the total number of texts. Indicates that it contains Words The number of texts, Indicates Words The importance level, and represents the adjustment factor; Among them, the importance levels can be pre-set based on experience or needs. For example, the importance levels can be set to 1, 2, 3, 4, 5, 6 and 7, and the importance increases in order. The adjustment coefficient can be pre-set based on experimental data analysis or historical data analysis.

[0038] Step A3: Collect all numerical data, fill missing values, replace outliers and standardize data to form standard comprehensive data.

[0039] If the values ​​of some features are missing, you can choose to delete the corresponding rows, or use the mean, median, etc. to fill the missing values. For outliers, you can use methods such as box plots and Z-score to detect and remove or replace outliers to avoid extreme values ​​from having a negative impact on clustering results. Since different features may have different dimensions and ranges (for example, traffic consumption may be hundreds of MB, and call duration may be tens of minutes), use Min-Max normalization to standardize the data so that the values ​​of all features are on the same scale.

[0040] To construct a blank dictionary: Set up a Each column is mapped to a unique key to form a key-cell pair, which is used to index the array in each column; After the array is filled, the blank columns that are not filled in the array are deleted; in, Indicates the length of the blank column. Indicates the width of the blank column. Indicates the level of blank grid columns; Step S2: using the improved multi-feature K-means technology to analyze the standard comprehensive data of users, grouping IoT card users, and obtaining multiple data sets; Methods for analyzing standard synthetic data using the improved multi-feature K-means technique include: Step D1: preset the K value and initialize K cluster centers using the probability feature method; Step D2: Calculate the distance between each user's standard comprehensive data and the cluster center. ,in, Indicates the cluster center The data value corresponding to each data category, Indicates the user's standard comprehensive data The data values ​​corresponding to the data categories are calculated, and the standard comprehensive data are assigned to the nearest cluster center, dividing the users into K groups; Step D3: Identify the number of users in each group. If the number of users is less than the limit threshold, release the users in the group and delete the cluster center corresponding to the group. Recalculate the distance between the released users and the remaining cluster centers and reallocate them to obtain the updated group K_1 into which the users are divided. Step D4: extract the standard comprehensive data of each user in the K_1 group and update the weight of each data category in each group; The boundary threshold and the segmentation threshold are pre-set based on experimental data analysis or historical data analysis, and can also be set based on operator experience.

[0041] Step D5: Calculate the standard comprehensive data of all users in each group using weighted average, i.e. , collect the means of all data categories and update the cluster centers; in, Indicates the first The mean of the data categories, Indicates the number of users in the group. Indicates the first User about The weight of each data category, Indicates the first User about data values ​​for each data category, The index of the user; Step D6: Repeat steps D2 to D6 until the change of all cluster centers is less than a preset threshold, and then obtain users divided into K_1 groups to form multiple data sets.

[0042] The K-means clustering algorithm relies heavily on the selection of the initial cluster centers. The traditional K-means algorithm randomly selects the initial centers, which may lead to problems such as slow algorithm convergence and poor clustering effect.

[0043] Methods for initializing K cluster centers using the probabilistic feature method include: Step W1: Calculate the cumulative value of the standard comprehensive data of each user, arrange them in order, and extract the standard comprehensive data corresponding to the median of the cumulative value as the first cluster center; The cumulative value refers to the cumulative addition of all the values ​​of the user's standard comprehensive data. The purpose is to select the middle user standard comprehensive data as the first cluster center and select the basis for the generation of subsequent K-1 cluster centers.

[0044] Step W2: Calculate the distance from each user's standard comprehensive data to the first cluster center ,in, Indicates The numerical value corresponding to each data category, Indicates the first cluster center The numerical value corresponding to each data category, is the index of the data category; Step W3: Calculate the probability of each user's standard comprehensive data being selected as the next cluster center ,in, Indicates The distance from the standard comprehensive data of each user to the first cluster center, Indicates The distance from the standard comprehensive data of each user to the first cluster center, The index of the user; The above means that the farther the standard comprehensive data is from the existing center point, the greater the probability of being selected as the next cluster center point. This can ensure that the new cluster center point is as far away from the existing cluster center point as possible, thereby avoiding the initial cluster center point from being concentrated in a certain area.

[0045] Step W4: Use the roulette algorithm to select the next cluster center according to the probability of each standard comprehensive data being selected as the next cluster center; Step W5: Repeat steps W2 to W4 until K-1 cluster centers are selected, and collect the first cluster center and K-1 cluster centers as the initialized K cluster centers.

[0046] The initial center is selected by the probability feature method to ensure that the initial center point is as dispersed as possible and improve the clustering effect. The roulette algorithm is a random selection method based on probability. Each standard comprehensive data corresponds to an area on the roulette wheel. The size of the area is proportional to the probability of the standard comprehensive data being selected. The roulette wheel is randomly rotated. The area where the pointer stops is selected as the next clustering center.

[0047] Update the weights of each data category within each group, including: Count the number of values ​​of each data category in each group and calculate the probability of each data category in the group ,in, Indicates the first The number of values ​​for each data category, Indicates the total number of values ​​in the group. Indicates the first The probability of each data category; Calculate the weight of each data category in the group ,in, Is a positive number used to prevent the denominator from being zero. Then the weights of each data category in each group are updated.

[0048] If the data is two-dimensional, you can directly draw a scatter plot to observe the clustering effect. Use different colors to mark different clusters and observe the distribution of data points in the clusters and the separation between clusters.

[0049] Step S3: construct and use the sample set to train the package personalized design model, import the real-time multiple data sets into the trained package personalized design model, and obtain the real-time IoT card package plan; The method of building and using sample sets to train a package personalization design model includes: The framework of the package personalized design model is based on neural network learning technology, including input layer, hidden layer and output layer. The input of the input layer is set as multiple sets of data sets, and the output layer sets P_C output folders in parallel. Each output folder is set with 4 task heads, namely traffic prediction head, phone prediction head, information prediction head and additional service prediction head. P_C output folders represent P_C configuration schemes designed for users. P_C configuration schemes are collected together to form IoT card package schemes. The four task heads represent the specific designs of traffic, phone calls, information and additional services in each configuration scheme. The traffic prediction header, phone prediction header, and information prediction header use a linear activation function and set the loss function to mean square error, which is used to measure the difference between the predicted value and the actual corresponding value. The additional service prediction header uses a ReLU activation function and sets the loss function to multi-class cross entropy loss, which is used to calculate the error between the predicted additional service label and the actual corresponding label. The total loss function of the package personalized design model is set to the sum of the loss functions of each task head; The optimization goal of the package personalization design model is to minimize user costs and maximize operator revenue; The sample sets are input into the personalized package design model in sequence. An adaptive learning rate strategy is used. The Adam or SGD optimizer is adopted to output the loss function of each task head in sequence. The training is stopped when the total loss function value no longer changes or the maximum number of iterations is reached to obtain a trained personalized package design model.

[0050] The sample set settings include: The sample set includes T_U groups of historical samples. Each group of samples includes multiple sets of user data sets and corresponding IoT card package plans. The timestamp of each sample is extracted and the samples are split according to quarters or months. The samples before the split are full samples, and the samples after the split are periodic samples. When using the sample set to train the package personalized design model, it is preferred to use the periodic samples to pre-train the model, and then use the full samples to continue training, thereby obtaining the trained package personalized design model.

[0051] Step S4: Visualize and interact with the IoT card package plan.

[0052] Visual interaction refers to displaying the IoT card package plan on a computer interface or display device to facilitate observation and operation by management personnel.

[0053] Embodiment 2 Reference Figure 1 and Figure 2 , Example 2 further illustrates an Internet of Things card package configuration system based on data analysis proposed by the present invention.

[0054] A method for configuring an Internet of Things card package based on data analysis comprises the following steps: Step S1: Collect comprehensive user information within a period of time, and perform format conversion processing on the comprehensive user information to obtain standard comprehensive data; Step S2: using the improved multi-feature K-means technology to analyze the standard comprehensive data of users, grouping IoT card users, and obtaining multiple data sets; Step S3: construct and use the sample set to train the package personalized design model, import the real-time multiple data sets into the trained package personalized design model, and obtain the real-time IoT card package plan; Step S4: Visualize and interact with the IoT card package plan.

[0055] A data analysis-based Internet of Things card package configuration system, applied to a data analysis-based Internet of Things card package configuration method, includes: Data collection and processing module: used to collect comprehensive user information within a period of time, and perform format conversion processing on the comprehensive user information to obtain standard comprehensive data; Feature clustering module: Based on standard comprehensive data, the improved multi-feature K-means technology is used to group IoT card users and obtain multiple data sets; Personalized design module: Build and train a personalized package design model, import multiple sets of real-time data sets into the trained personalized package design model, and obtain real-time IoT card package solutions; Interface interaction module: interact with operators through a visual interface; The modules are connected to each other via wired and / or wireless means.

[0056] In addition, according to the implementation of the present application, the process described in the figure of the Internet of Things card package configuration system based on data analysis can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, which stores machine-readable instructions, and the machine-readable instructions can be run by a processor to execute instructions corresponding to the method steps provided in the present application. Of course, the architecture shown in the figure of the Internet of Things card package configuration system based on data analysis is only exemplary. When implementing different devices, adaptive selection or adjustment can be made according to actual needs.

[0057] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0058] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for configuring an Internet of Things card package based on data analysis, characterized in that: The Internet of Things card package configuration method based on data analysis includes: Step S1: Collect comprehensive user information within a period of time, and perform format conversion processing on the comprehensive user information to obtain standard comprehensive data; Step S2: using the improved multi-feature K-means technology to analyze the standard comprehensive data of users, grouping IoT card users, and obtaining multiple data sets; Step S3: construct and use the sample set to train the package personalized design model, import the real-time multiple data sets into the trained package personalized design model, and obtain the real-time IoT card package plan; Step S4: Visualize and interact with the IoT card package plan.

2. The method for configuring Internet of Things card packages based on data analysis according to claim 1, characterized in that: The comprehensive user information includes the user's basic information and behavior preference data, IoT card usage behavior data and package billing data, and user evaluation feedback data; The user's basic information includes identity information, device information, and geographic location information; Behavioral preference data includes package query frequency, app resource usage ratio, package change frequency and additional service data; The usage behavior data of IoT cards includes traffic usage data, voice usage data, and SMS usage data; Package billing data includes the total consumption amount of the package and the consumption amount corresponding to each usage behavior data; User evaluation feedback data refers to the user's satisfaction with the package.

3. The method for configuring Internet of Things card packages based on data analysis according to claim 2, characterized in that: The method for performing format conversion processing on the comprehensive user information comprises: Step A1: construct a blank dictionary, set the initial value, use Python to generate different arrays from the initial value to form a data set, identify the arrays in the comprehensive user information, delete the arrays in the data set that are identical to it, fill the deleted data set into the blank dictionary, and arrange the arrays in order of size to form a conversion dictionary; Step A2: extracting text data from the comprehensive user information, removing irrelevant symbols, using a sentence segmentation tool to segment sentences in the text, using a word segmentation tool to segment sentences into words, calculating the weight index of each word, arranging each word according to the weight index, and the arrangement order is the same as the array order in the conversion dictionary, and converting the text data into numerical data; No. The weight index of the word ; in, Indicates Words In the Text The proportion of Indicates the total number of texts. Indicates that it contains Words The number of texts, Indicates Words The importance level, and represents the adjustment factor; Step A3: Collect all numerical data, fill missing values, replace outliers and standardize data to form standard comprehensive data.

4. The method for configuring Internet of Things card packages based on data analysis according to claim 3 is characterized in that: The method of constructing a blank dictionary includes: Set up a Each column is mapped to a unique key to form a key-cell pair, which is used to index the array in each column; After the array is filled, the blank columns that are not filled in the array are deleted; in, Indicates the length of the blank column. Indicates the width of the blank column. Indicates the level of the blank grid column.

5. The method for configuring Internet of Things card packages based on data analysis according to claim 4, characterized in that: The method for analyzing standard comprehensive data using the improved multi-feature K-means technology includes: Step D1: preset the K value and initialize K cluster centers using the probability feature method; Step D2: Calculate the distance between each user's standard comprehensive data and the cluster center, and assign the standard comprehensive data to the cluster center with the nearest distance, dividing the users into K groups; Step D3: Identify the number of users in each group. If the number of users is less than the limit threshold, release the users in the group and delete the cluster center corresponding to the group. Recalculate the distance between the released users and the remaining cluster centers and reallocate them to obtain the updated group K_1 into which the users are divided. Step D4: extract the standard comprehensive data of each user in the K_1 group and update the weight of each data category in each group; Step D5: Calculate the standard comprehensive data of all users in each group using weighted average, i.e. , collect the means of all data categories and update the cluster centers; in, Indicates the first The mean of the data categories, Indicates the number of users in the group. Indicates the first User about The weight of each data category, Indicates the first User about data values ​​for each data category, The index of the user; Step D6: Repeat steps D2 to D6 until the change of all cluster centers is less than a preset threshold, and then obtain users divided into K_1 groups to form multiple data sets.

6. The method for configuring Internet of Things card packages based on data analysis according to claim 5, characterized in that: The method of initializing K cluster centers using the probability feature method includes: Step W1: Calculate the cumulative value of the standard comprehensive data of each user, arrange them in order, and extract the standard comprehensive data corresponding to the median of the cumulative value as the first cluster center; Step W2: Calculate the distance from each user's standard comprehensive data to the first cluster center ,in, Indicates The numerical value corresponding to each data category, Indicates the first cluster center The numerical value corresponding to each data category, is the index of the data category; Step W3: Calculate the probability of each user's standard comprehensive data being selected as the next cluster center ,in, Indicates The distance from the standard comprehensive data of each user to the first cluster center, Indicates The distance from the standard comprehensive data of each user to the first cluster center, The index of the user; Step W4: Use the roulette algorithm to select the next cluster center according to the probability of each standard comprehensive data being selected as the next cluster center; Step W5: Repeat steps W2 to W4 until K-1 cluster centers are selected, and collect the first cluster center and K-1 cluster centers as the initialized K cluster centers.

7. The method for configuring Internet of Things card packages based on data analysis according to claim 6, characterized in that: The updating of the weight of each data category in each group includes: Count the number of values ​​of each data category in each group and calculate the probability of each data category in the group ,in, Indicates the first The number of values ​​for each data category, Indicates the total number of values ​​in the group. Indicates the first The probability of each data category; Calculate the weight of each data category in the group ,in, is a positive number, which updates the weight of each data category in each group.

8. The method for configuring Internet of Things card packages based on data analysis according to claim 7, characterized in that: The method for constructing and using a sample set to train a package personality design model comprises: The framework of the package personalized design model is based on neural network learning technology, including input layer, hidden layer and output layer. The input of the input layer is set as multiple sets of data sets, and the output layer sets P_C output folders in parallel. Each output folder is set with 4 task heads, namely traffic prediction head, phone prediction head, information prediction head and additional service prediction head. P_C output folders represent P_C configuration schemes designed for users. P_C configuration schemes are collected together to form IoT card package schemes. The four task heads represent the specific designs of traffic, phone calls, information and additional services in each configuration scheme. The traffic prediction header, phone prediction header, and information prediction header use a linear activation function and set the loss function to mean square error, which is used to measure the difference between the predicted value and the actual corresponding value. The additional service prediction header uses a ReLU activation function and sets the loss function to multi-class cross entropy loss, which is used to calculate the error between the predicted additional service label and the actual corresponding label. The total loss function of the package personalized design model is set to the sum of the loss functions of each task head; The optimization goal of the package personalization design model is to minimize user costs and maximize operator revenue; The sample sets are input into the personalized package design model in sequence. An adaptive learning rate strategy is used. The Adam or SGD optimizer is adopted to output the loss function of each task head in sequence. The training is stopped when the total loss function value no longer changes or the maximum number of iterations is reached to obtain a trained personalized package design model.

9. The method for configuring Internet of Things card packages based on data analysis according to claim 8, characterized in that: The setting of the sample set includes: The sample set includes T_U groups of historical samples. Each group of samples includes multiple sets of user data sets and corresponding IoT card package plans. The timestamp of each sample is extracted and the samples are split according to quarters or months. The samples before the split are full samples, and the samples after the split are periodic samples. When using the sample set to train the package personalized design model, it is preferred to use the periodic samples to pre-train the model, and then use the full samples to continue training, thereby obtaining the trained package personalized design model.

10. A data analysis-based Internet of Things card package configuration system, applied to a data analysis-based Internet of Things card package configuration method according to any one of claims 1 to 9, characterized in that: The Internet of Things card package configuration system based on data analysis includes: Data collection and processing module: used to collect comprehensive user information within a period of time, and perform format conversion processing on the comprehensive user information to obtain standard comprehensive data; Feature clustering module: Based on standard comprehensive data, the improved multi-feature K-means technology is used to group IoT card users and obtain multiple data sets; Personalized design module: Build and train a personalized package design model, import multiple sets of real-time data sets into the trained personalized package design model, and obtain real-time IoT card package solutions; Interface interaction module: interact with operators through a visual interface.

Images