Dynamic short message pricing method based on customer behavior characteristics

Abstract

The invention provides a dynamic short message pricing method based on customer behavior characteristics, belongs to the technical field of short message pricing, and also relates to the technical field of communication service pricing. According to the method, client short message sending behavior data is collected in real time, behavior characteristics are analyzed by using a machine learning algorithm, and an optimal charging scheme is automatically matched based on a multi-dimensional tiered pricing model, so that the problems that the existing fixed rate or simple grading pricing cannot adapt to the actual sending behavior of a client, the resource distribution is unreasonable and a behavior incentive mechanism is lacked are solved; dynamic pricing, optimized resource configuration and win-win of customers and service providers are realized. Therefore, accurate matching of pricing and actual sending behaviors of clients is realized, network resource allocation is optimized, the clients are stimulated to standardize the sending behaviors, and the client experience and the service provider operation benefits are improved.

Description

Technical Field

[0001] This invention belongs to the field of SMS pricing technology, specifically relating to a dynamic SMS pricing method based on customer behavior characteristics. Background Technology

[0002] The SMS service industry currently uses a fixed rate or a simple tiered pricing model, which has significant drawbacks.

[0003] Under the fixed-rate model, a uniform price standard is applied regardless of the customer's sending frequency, time period selection, or SMS delivery effectiveness. High-frequency sending customers generate a large volume of business but cannot obtain corresponding price discounts, reducing customer loyalty; low-frequency customers do not enjoy flexible pricing that matches their usage, resulting in insufficient satisfaction.

[0004] Simple tiered pricing divides price ranges solely based on the volume of messages sent, failing to consider key behavioral characteristics such as sending time, SMS success rate, and complaint rate. This makes it difficult for service providers to adjust network resource allocation through pricing, resulting in both excessive network load during peak hours and idle resources during off-peak hours, leading to low resource utilization efficiency. Furthermore, the existing pricing model lacks incentive mechanisms to encourage customers to maintain standardized sending behavior, failing to guide them to optimize their sending strategies and easily leading to problems such as low delivery rates and high complaint rates, increasing the operating costs of service providers.

[0005] Therefore, there is an urgent need for a text message pricing method that can accurately match customer behavior characteristics and dynamically adjust pricing standards to make up for the shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a dynamic SMS pricing method based on customer behavior characteristics, which achieves precise matching between pricing and actual customer sending behavior, optimizes network resource allocation, incentivizes customers to standardize sending behavior, and improves customer experience and service provider operational efficiency.

[0007] The present invention employs the following technical solution.

[0008] A dynamic SMS pricing method based on customer behavior characteristics includes:

[0009] Step 1: Collect multi-dimensional behavioral data of customers during the SMS sending process in real time;

[0010] Step 2: Process the collected behavioral data using the random forest algorithm, extract customer behavior feature vectors, and train the random forest algorithm to build a customer behavior profile;

[0011] Step 3: Set up a multi-dimensional tiered pricing model, which includes a basic sending volume tiered model, a time-based discount model, and a behavior reward model;

[0012] Step 4: Based on customer behavior feature vectors, the optimal billing scheme is matched from the pricing model library using a weighted scoring method;

[0013] Step 5: Perform dynamic billing based on the optimal billing plan and generate a detailed bill. The detailed bill includes billing details for each dimension and supports real-time querying.

[0014] Preferably, in step 1, the multi-dimensional behavioral data includes the customer ID of the customer SMS, sending timestamp, sending time period, number of messages sent, number of recipients, SMS content type, delivery success rate, and complaint records. The data collection frequency of the multi-dimensional behavioral data is no less than once per second.

[0015] Preferably, in step 1, the SMS content types include notification SMS, marketing SMS, and verification SMS.

[0016] Preferably, in step 1, the time periods for sending customer SMS messages include set off-peak hours, busy hours, and peak hours.

[0017] Preferably, in step 2, the feature vector includes core indicators such as the frequency of customer SMS sending, the proportion of time period distribution, the proportion of content type, the delivery rate, and the complaint rate.

[0018] Preferably, step 2 specifically includes:

[0019] Input features for a random forest model are generated based on collected customer SMS behavior data;

[0020] Next, the random forest model is trained and its feature importance is evaluated by inputting its input features and multi-dimensional behavioral data. This process includes:

[0021] Using customer value level as the prediction target, a random forest classification model is trained. The model learns the correlation between input features (behavioral features) and customer value. The model parameters of the random forest classification model are set as follows:

[0022] Number of decision trees: 100;

[0023] Maximum depth per tree: 10;

[0024] Minimum number of samples for node splitting: 5;

[0025] Random sampling ratio: Each tree uses 70% of the multi-dimensional behavioral data and 60% of the input features as samples (to ensure tree diversity).

[0026] After the random forest classification model is trained, it outputs the importance scores of 12 input features;

[0027] The six input features with the highest importance scores were then used as customer behavior feature vectors. :

[0028] ;

[0029] in Customer behavior feature vector The first in Each input feature ;

[0030] Subsequently, the customer behavior feature vector was analyzed. The six input features are normalized using the following formula:

[0031] ;

[0032] Among them and These are customer behavior feature vectors. The first in The minimum and maximum values ​​of each input feature across all customers. ;

[0033] Finally, customer behavior profiles are constructed, and the methods include:

[0034] Standardized feature vectors of all customers As input parameters for the K-Means clustering algorithm, customers are divided into 5 behavioral profiles using the K-Means clustering algorithm.

[0035] Preferably, in step 2, the input features of the random forest model specifically include:

[0036] Sending frequency: average daily sending volume, hourly sending fluctuation variance;

[0037] Time-of-day distribution: percentage of data sent during off-peak hours and percentage of data sent during peak hours;

[0038] Content type breakdown: Marketing SMS messages, verification SMS messages;

[0039] Quality: Delivery rate, complaint rate;

[0040] Interaction: Average number of receivers, receiver repetition rate;

[0041] Time patterns: percentage of consecutive sending days, percentage of sending at night.

[0042] Preferably, in step 3, the basic transmission volume ladder model is calculated using the following formula:

[0043] ;

[0044] in Billing amount based on basic transmission volume; For the first The number of text messages sent to customers in each tier. For the number of steps, just as For the range of 0-10,000 SMS messages sent, For SMS messages sent in the range of 10,000 to 50,000, For SMS messages sent in the range of 50,000 or more; For the first Tiered unit rates for sending text messages;

[0045] The time-based discount model is calculated using the following formula:

[0046] ;

[0047] in This is the billing amount after the time-limited discount; This is the time-based discount factor, representing the off-peak hours. =0.8, during peak hours =1.0, during peak hours =1.2;

[0048] The behavioral reward model is calculated using the following formula:

[0049] ;

[0050] in This is the final billing amount; The reward is a percentage of the delivery rate; when the delivery rate is >99%. =0.05, otherwise =0; A higher rebate percentage is awarded for low complaint rates, specifically when the complaint rate is <0.1%. =0.03, otherwise =0.

[0051] Preferably, step 4 specifically includes:

[0052] For each candidate solution in the pricing model library Calculate its matching score with customer characteristics. :

[0053] ;

[0054] in Candidate solutions Match score; Customer behavior feature vectors set according to specific requirements The first in The weight coefficients of each input feature, and satisfying ; Candidate solutions In the Standard values ​​for each input feature dimension;

[0055] The candidate scheme with the lowest score is then selected as the optimal billing scheme, and its calculation formula is as follows:

[0056] ;

[0057] in This is the sequence number of the optimal billing scheme.

[0058] Preferably, in step 5, the detailed bill specifically includes the customer ID, billing cycle, bill generation time, final amount due, and bill status.

[0059] The beneficial effects of the present invention are as follows, compared with the prior art:

[0060] Dynamic pricing adapted to customer behavior: By combining multi-dimensional behavioral feature analysis with a tiered pricing model, high-frequency sending customers can enjoy lower unit rates, and the actual usage cost of customers is accurately matched with their sending behavior, thereby improving customer stickiness.

[0061] Optimize network resource allocation: By leveraging time-based discount coefficients, customers are encouraged to send SMS messages during off-peak hours, reducing network load during peak hours, improving network resource utilization efficiency, and reducing congestion.

[0062] Incentivizing customers to send messages in a standardized manner: The behavioral reward model links delivery rate, complaint rate and cost, prompting customers to optimize SMS content and recipient quality, reduce complaint rate, improve service quality and reduce service provider operating costs.

[0063] Automated and efficient billing: Without human intervention, the system automatically completes data collection, analysis, pricing matching and billing, improving billing efficiency and reducing manual management costs. Attached Figure Description

[0064] Figure 1 This is a flowchart of the dynamic SMS pricing method based on customer behavior characteristics in this invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, other embodiments obtained by those skilled in the art without creative effort are all within the protection scope of this invention.

[0066] like Figure 1 As shown, a dynamic SMS pricing method based on customer behavior characteristics includes:

[0067] Step 1: Collect multi-dimensional behavioral data of customers during the SMS sending process in real time;

[0068] In a preferred but non-limiting embodiment of the present invention, in step 1, the multi-dimensional behavioral data includes the customer ID of the customer's SMS message, sending timestamp, sending time period, number of messages sent, number of recipients, SMS content type (notification, marketing, verification), delivery success rate, complaint records, etc. The data collection frequency of the multi-dimensional behavioral data is no less than once per second to ensure data real-time performance. The customer ID is a pre-defined unique identifier for each customer.

[0069] In a preferred but non-limiting embodiment of the present invention, in step 1, the SMS content types include notification SMS, marketing SMS, and verification SMS.

[0070] In a preferred but non-limiting embodiment of the present invention, in step 1, the time period for sending customer SMS messages includes setting off-peak hours (such as 0:00-8:00), busy hours (such as 8:00-24:00 on non-holidays), and peak hours (such as 8:00-24:00 on holidays).

[0071] Step 2: Process the collected behavioral data using the random forest algorithm, extract customer behavior feature vectors, and train the random forest algorithm to build a customer behavior profile;

[0072] In a preferred but non-limiting embodiment of the present invention, in step 2, the feature vector includes core indicators such as the frequency of customer SMS sending (number of messages sent per unit time), the proportion of time period distribution (proportion of messages sent during off-peak / busy / peak hours), the proportion of content types, the delivery rate, and the complaint rate.

[0073] In a preferred but non-limiting embodiment of the present invention, step 2 specifically includes:

[0074] The random forest algorithm is used to process customer behavior data. The core of this approach is to mine key behavioral patterns from raw data through ensemble learning of multiple decision trees, quantifying and extracting core features such as sending frequency and time period distribution, and constructing customer behavior profiles based on these feature vectors. The process includes:

[0075] Raw behavioral data preprocessing and feature engineering;

[0076] Random forest model training and feature importance evaluation;

[0077] Customer behavior feature vector extraction;

[0078] Customer behavior profile construction based on feature vectors.

[0079] Collect customer SMS behavior data for 30 consecutive days, including:

[0080] Basic attributes: Customer ID, sending timestamp, SMS content, number of recipients;

[0081] Outcome: Delivery status (success / failure), complaint flag (present / absent);

[0082] Environmental attributes: Sending time (mapped to off-peak / busy / peak hours), content type (notification / marketing / verification, categorized by keyword matching or NLP).

[0083] Input features for a random forest model are generated based on collected customer SMS behavior data;

[0084] Next, the random forest model is trained and its feature importance is evaluated by inputting its input features and multi-dimensional behavioral data. This process includes:

[0085] (1) Model training objectives

[0086] Using customer value levels (labeled into high / medium / low tiers, manually labeled based on historical spending and renewal rates) as the prediction target, a random forest classification model is trained. The model learns the correlation between input features (behavioral features) and customer value. The model parameters of the random forest classification model are set as follows:

[0087] Number of decision trees: 100 (balancing model accuracy and computational efficiency);

[0088] Maximum depth per tree: 10 (to avoid overfitting);

[0089] Minimum number of samples for node splitting: 5;

[0090] Random sampling ratio: Each tree uses 70% of the multi-dimensional behavioral data and 60% of the input features as samples (to ensure tree diversity).

[0091] (3) Calculation of feature importance

[0092] After the random forest classification model is trained, it outputs the importance scores of 12 input features (based on the contribution of the feature to the decision tree split; the higher the score, the stronger the feature's ability to distinguish customer value).

[0093] The six input features with the highest importance scores were then used as customer behavior feature vectors. :

[0094] ;

[0095] in Customer behavior feature vector The first in Each input feature ;

[0096] Subsequently, the customer behavior feature vector was analyzed. The six input features are normalized (mapped to the [0,1] interval) to eliminate dimensional differences. The calculation formula for normalization is as follows:

[0097] ;

[0098] Among them and These are customer behavior feature vectors. The first in The minimum and maximum values ​​of each input feature across all customers. ;

[0099] Finally, customer behavior profiles are constructed, and the methods include:

[0100] Standardized feature vectors of all customers As input parameters for the K-Means clustering algorithm, customers are divided into 5 behavioral profiles using the K-Means clustering algorithm.

[0101] The technical effect of step 2 is:

[0102] Feature extraction accuracy: By evaluating the feature importance of random forest, the core indicators most strongly associated with customer behavior (such as sending frequency and delivery rate) are selected, avoiding interference from redundant features and improving the representativeness of the vectors.

[0103] Profile stability: Based on 30 days of continuous data training and through integrated learning of multiple decision trees, the impact of single data fluctuations is reduced, making customer profiles more stable (over 92% consistency across months).

[0104] Interpretability of behavioral patterns: Compared to black-box models (such as neural networks), the feature importance scores of random forests can intuitively explain "which behaviors have the greatest impact on customer profiles" (such as "average daily sending volume" having the highest weight), making it easier for service providers to understand customer behavior logic.

[0105] Dynamic update capability: Supports weekly update of feature vectors (recalculating data from the last 30 days), capturing changes in customer behavior in real time (such as switching from "low-frequency verification" to "high-frequency quality"), providing timely support for dynamic pricing.

[0106] This method enables the entire process of transformation from raw behavioral data to structured feature vectors and then to concrete customer profiles, laying a data foundation for the accurate matching of subsequent dynamic SMS pricing schemes.

[0107] In a preferred but non-limiting embodiment of the present invention, in step 2, the input features of the random forest model specifically include:

[0108] Sending frequency (number of messages sent per unit time): daily average number of messages sent, hourly sending fluctuation variance; the formula for calculating the daily average number of messages sent is the total number of messages sent by the customer over 30 consecutive days / 30 days; the formula for calculating the hourly sending fluctuation variance is the mean variance of the number of messages sent in each hour of each day over 30 consecutive days.

[0109] Time Period Distribution Ratio (Off-peak / Busy / Peak Period Sending Ratio): Off-peak sending ratio, peak period sending ratio; The formula for calculating the off-peak sending ratio is the customer's SMS off-peak sending volume over 30 consecutive days / the customer's total SMS sending volume over 30 consecutive days; The formula for calculating the peak period sending ratio is the customer's peak period sending volume over 30 consecutive days / the customer's total sending volume over 30 consecutive days.

[0110] Content type breakdown: Marketing SMS percentage, Verification SMS percentage; The formula for calculating the marketing SMS percentage is the number of marketing SMS messages sent by the customer in 30 consecutive days / the total number of messages sent by the customer in 30 consecutive days; The formula for calculating the verification SMS percentage is the number of verification SMS messages sent by the customer in 30 consecutive days / the total number of messages sent by the customer in 30 consecutive days.

[0111] Quality: Delivery rate, complaint rate; Delivery rate is the average daily delivery success rate, which is calculated as the average of the number of successfully delivered text messages sent by the customer over 30 consecutive days / the average number of text messages sent by the customer over 30 consecutive days; Complaint rate is the slope of the complaint rate trend, which is calculated by taking the complaint rate of the customer's text messages sent over 30 consecutive days on a weekly basis, and then performing a linear fit based on the complaint rate and its corresponding complaint time. The slope of the straight line obtained by the linear fit is the slope of the complaint rate trend.

[0112] Interaction: Average number of recipients, recipient duplication rate; the average number of recipients is the average number of recipients for each SMS sent by the customer over 30 consecutive days; the recipient duplication rate is calculated as the number of numbers that received duplicate SMS messages from the customer over 30 consecutive days / the total number of numbers that received SMS messages from the customer over 30 consecutive days.

[0113] Time patterns: percentage of consecutive sending days, percentage of nighttime sending; the formula for the percentage of consecutive sending days is the number of consecutive days the customer sent SMS messages over 30 consecutive days / 30 days; the formula for the percentage of nighttime sending is the number of SMS messages sent by the customer between 22:00 and 24:00 over 30 consecutive days / the total number of SMS messages sent by the customer over 30 consecutive days.

[0114] Step 3: Set up a multi-dimensional tiered pricing model, which includes a basic sending volume tiered model, a time-based discount model, and a behavior reward model;

[0115] In a preferred but non-limiting embodiment of the present invention, in step 3, the basic transmission volume tiered model uses the customer's cumulative transmission volume as the pricing basis and sets tiered rates; the higher the transmission volume, the lower the unit rate. The basic transmission volume tiered model is calculated using the following formula:

[0116] ;

[0117] in Billing amount based on basic transmission volume (RMB); For the first Number of text messages sent to customers in each tier. For the number of steps, just as For the range of 0-10,000 SMS messages sent, For SMS messages sent in the range of 10,000 to 50,000, For SMS messages sent in the range of 50,000 or more; For the first The tiered unit rate for sending text messages (yuan / message) is as follows: =0.05, =0.045, =0.04;

[0118] The time-based discount pricing model sets a discount coefficient based on the SMS sending time period, adjusting the sending cost for different time periods and optimizing network resource allocation. The time-based discount model is calculated using the following formula:

[0119] ;

[0120] in The billing amount (in yuan) is the discounted amount for the time period. This is the time-based discount factor, representing the off-peak hours. =0.8, during peak hours =1.0, during peak hours =1.2;

[0121] The behavioral reward pricing model sets reward rebates based on customer SMS delivery rate and complaint rate to incentivize customers to optimize their sending behavior. The behavioral reward model is calculated using the following formula:

[0122] ;

[0123] in This is the final billing amount (in yuan); The reward is a percentage of the delivery rate; when the delivery rate is >99%. =0.05, otherwise =0; A higher rebate percentage is awarded for low complaint rates, specifically when the complaint rate is <0.1%. =0.03, otherwise =0.

[0124] Step 4: Based on customer behavior feature vectors, the optimal billing scheme is matched from the pricing model library using a weighted scoring method; ensuring that the pricing scheme comprehensively adapts to customer behavior.

[0125] In a preferred but non-limiting embodiment of the present invention, step 4 specifically includes:

[0126] For each candidate solution in the pricing model library Calculate its matching score with customer characteristics. :

[0127] ;

[0128] in Candidate solutions (Candidate solutions can be based on a tiered model of basic delivery volume, a time-based discount model, or a behavioral reward model) matching score (the smaller the value, the higher the matching degree between the solution and the customer characteristics). Customer behavior feature vectors set according to specific requirements The first in The weight coefficients of each input feature (behavioral feature) satisfy the following conditions: For example, the weight of the amount of data sent is 0.4, and the weight of the time period distribution is 0.2. Candidate solutions In the Standard values ​​of each input feature dimension ( An ideal matching value pre-set for a pricing scheme based on a specific behavioral characteristic (such as sending volume), used to quantify the degree to which the scheme matches the customer's actual behavior. The absolute difference between the customer's standardized characteristics and the solution's standard values ​​(reflecting a single-dimensional matching deviation);

[0129] The candidate scheme with the lowest score is then selected as the optimal billing scheme, and its calculation formula is as follows:

[0130] ;

[0131] in This is the sequence number of the optimal billing scheme.

[0132] The technical effect of step 4 is as follows:

[0133] Precisely match customer behavior: By using multi-dimensional feature weighted scoring, pricing deviations caused by a single dimension (such as only looking at the number of messages sent) are avoided, so that the billing plan is deeply adapted to the customer's actual behavior (such as the time of sending and the quality of SMS messages), thereby improving the customer's acceptance of the pricing.

[0134] Dynamically adapt to changes in behavior: When customer behavior characteristics (such as sending volume changing from low frequency to high frequency) change, the system can recalculate the score in real time and switch to a better solution, realizing dynamic adjustment of pricing as behavior changes.

[0135] Optimize solution selection efficiency: Replace manual judgment with quantitative scoring to reduce the subjectivity of solution matching, and has low computational complexity (linear time complexity), supporting real-time solution updates for millions of customers.

[0136] Improve the accuracy of resource allocation: For example, matching off-peak discount schemes to customers with a high proportion of off-peak transmissions can guide customers to send during off-peak hours, indirectly optimizing network resource allocation and reducing load during peak hours.

[0137] This method allows service providers to tailor billing plans for each customer, maximizing customer value and satisfaction while protecting their own revenue.

[0138] Step 5: Perform dynamic billing based on the optimal billing plan and generate a detailed bill. The detailed bill includes billing details for each dimension and supports real-time querying.

[0139] In a preferred but non-limiting embodiment of the present invention, in step 5, the detailed bill specifically includes the customer ID, billing period (e.g., month X, 202X), bill generation time, final amount due, and bill status (unpaid / paid).

[0140] An example of the present invention is shown below:

[0141] 1. Implementation steps

[0142] (1) Collect customer SMS sending data in real time, including the sending time, recipient number, content type, delivery status (success / failure), and complaint records for each SMS. Clean the collected data and remove invalid data (such as duplicate sending records and data with incorrect format) to ensure data accuracy.

[0143] (2) Use the random forest algorithm to train and analyze the preprocessed data. Calculate the customer's sending frequency (total monthly sending volume / 30 days), time period distribution ratio (the proportion of sending volume during off-peak / busy / peak hours to the total monthly sending volume), content type ratio (the proportion of notification / marketing / verification SMS messages), delivery rate (number of successfully delivered SMS messages / total number of sent SMS messages), and complaint rate (number of complaint SMS messages / total number of sent SMS messages) on a monthly basis to form a customer-specific behavioral feature vector.

[0144] (3) Optimal pricing scheme matching: The optimal pricing selector performs weighted scoring based on customer behavior feature vectors and selects the optimal scheme from the pricing model library. For example:

[0145] Customer A: Monthly sending volume of 60,000 messages ( =10,000 entries =40,000 entries =10,000 messages), with 60% sent during off-peak hours, 30% during busy hours, and 10% during peak hours, achieving a delivery rate of 99.5% and a complaint rate of 0.05%.

[0146] Basic billing =10000×0.05+40000×0.045+10000×0.04=2700 yuan; Time-based discount coefficient = (60% × 0.8 + 30% × 1.0 + 10% × 1.2) = 0.86, =2700 × 0.86 = 2322 yuan; Behavioral reward rebate =0.05, =0.03, final billing =2322×(1-0.05-0.03)=2136.24 yuan.

[0147] (4) Execute billing according to the matched optimal pricing scheme, generate a detailed bill including basic fees, time period discount amount and behavior reward amount, and push it to customers through SMS, APP and other channels to support customers to query and verify in real time.

[0148] 2. Implementation Safeguards

[0149] The module formed by the method of this invention adopts a distributed architecture deployment, ensuring the stability of data processing and billing, supporting simultaneous online billing for millions of customers, with a response latency of no more than 100ms. The pricing model library is regularly updated, adjusting tiered rates, discount coefficients, and reward ratios based on market demand, network load, and changes in customer behavior to ensure model adaptability. A data security protection mechanism is established to encrypt, store, and transmit collected customer behavior data, complying with relevant data security regulations and protecting customer privacy.

[0150] The beneficial effects of the present invention are as follows, compared with the prior art:

[0151] Dynamic pricing adapted to customer behavior: By combining multi-dimensional behavioral feature analysis with a tiered pricing model, high-frequency sending customers can enjoy lower unit rates, and the actual usage cost of customers is accurately matched with their sending behavior, thereby improving customer stickiness.

[0152] Optimize network resource allocation: By leveraging time-based discount coefficients, customers are encouraged to send SMS messages during off-peak hours, reducing network load during peak hours, improving network resource utilization efficiency, and reducing congestion.

[0153] Incentivizing customers to send messages in a standardized manner: The behavioral reward model links delivery rate, complaint rate and cost, prompting customers to optimize SMS content and recipient quality, reduce complaint rate, improve service quality and reduce service provider operating costs.

[0154] Automated and efficient billing: Without human intervention, the system automatically completes data collection, analysis, pricing matching and billing, improving billing efficiency and reducing manual management costs.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention without departing from the spirit and scope of the present invention. Any modifications or equivalent substitutions should be covered within the scope of protection of the claims of the present invention.

Claims

1. A dynamic short message billing method based on customer behavior characteristics, characterized in that, The method comprises the following steps: Step 1: Real-time collection of multi-dimensional behavior data in the process of sending short messages by customers; Step 2: Processing of the collected behavior data by using a random forest algorithm, extraction of a customer behavior feature vector, and training and establishment of a customer behavior portrait by using the random forest algorithm; Step 3: Setting of a multi-dimensional step pricing model, wherein the multi-dimensional step pricing model comprises a basic sending quantity step model, a time period discount model, and a behavior reward model; Step 4: Matching of an optimal charging scheme from a charging model library by using a weighted scoring method based on the customer behavior feature vector; Step 5: Execution of dynamic charging according to the optimal charging scheme, and generation of a detailed bill, wherein the detailed bill comprises charging details of each dimension and supports real-time query.

2. The dynamic short message pricing method based on customer behavior characteristics according to claim 1, characterized in that, In step 1, the multi-dimensional behavior data comprises a customer ID, a sending timestamp, a sending time period, a sending quantity, a number of receiving parties, a short message content type, a delivery success rate, and a complaint record of the customer short message, and the data collection frequency of the multi-dimensional behavior data is not less than 1 time / second.

3. The dynamic short message pricing method based on customer behavior characteristics according to claim 2, characterized in that, In step 1, the short message content type comprises a notification type short message, a marketing type short message, and a verification type short message.

4. The dynamic short message pricing method based on customer behavior characteristics according to claim 3, characterized in that, In step 1, the sending time period of the customer short message comprises a set idle time period, a busy time period, and a peak time period.

5. The dynamic short message pricing method based on customer behavior characteristics according to claim 4, characterized in that, In step 2, the feature vector comprises a sending frequency, a time period distribution proportion, a content type proportion, a delivery rate, and a complaint rate of the customer short message.

6. The dynamic short message pricing method based on customer behavior characteristics according to claim 5, characterized in that, Step 2 specifically comprises the following steps: Generating input features of a random forest model based on the collected short message behavior data of the customer; Then, inputting the input features and the multi-dimensional behavior data into the random forest model to train the random forest model and evaluate the feature importance, which specifically comprises the following steps: Taking a customer value level as a prediction target, training a random forest classification model, learning the correlation between the input features as behavior features and the customer value by using the model, and setting the model parameters of the random forest classification model as follows: Number of decision trees: 100 Maximum depth of each tree: 10 Minimum number of samples for node splitting: 5 Random sampling ratio: 70% of the multi-dimensional behavior data and 60% of the input features are randomly sampled as samples for each tree (to ensure the diversity of the tree) After the training of the random forest classification model is completed, the importance scores of the 12 input features are outputted Subsequently, the 6 input features with the highest importance scores are selected as the customer behavior feature vector : ； wherein is a customer behavior feature vector is a first input feature in ; Subsequently, the 6 input features in the customer behavior feature vector are normalized, and the normalization formula is as follows: ； wherein with are the customer behavior feature vectors the minimum and maximum values of the th input feature in all customers, ; Finally, a customer behavior portrait is constructed, and the method comprises the following steps: standardized feature vectors of all customers As input parameters to the K-Means clustering algorithm, the customers are divided into 5 behavioral personas by the K-Means clustering algorithm.

7. The dynamic short message pricing method based on customer behavior characteristics according to claim 6, characterized in that, In step 2, the input features of the random forest model specifically comprise the following: Sending frequency: daily average sending quantity and hour-level sending fluctuation variance Time period distribution proportion: idle time sending proportion and peak time period sending proportion Content type proportion: marketing type short message proportion and verification type short message proportion Quality: delivery rate and complaint rate Interaction: average number of receiving parties and receiving party repetition rate Time regularity: continuous sending day proportion and night sending proportion 8. The dynamic short message pricing method based on customer behavior characteristics according to claim 7, characterized in that, In step 3, the basic sending quantity step model is calculated according to the following formula: ； wherein is the base transmission amount billing amount; is the first is the number of messages transmitted by the customer for the step, is the step number, as is the number of messages transmitted for the 0-10,000 message interval, is the number of messages transmitted for the 1-50,000 message interval, is the number of messages transmitted for the 50,000 message interval and above; is the first is the unit rate for messages transmitted for the step. The time period discount model is calculated according to the following formula: ； wherein is the billing amount after time-of-day discounting; is the time-of-day discount factor, which is = 0.8 for off-peak hours, = 1.0 for busy hours, and = 1.2 for peak hours. The behavior reward model is calculated according to the following formula: ； wherein is the final billing amount; is the delivery rate reward rebate ratio, when the delivery rate > 99% = 0.05, otherwise = 0; is the low complaint rate reward rebate ratio, when the complaint rate < 0.1% = 0.03, otherwise = 0.

9. The dynamic short message pricing method based on customer behavior characteristics according to claim 8, characterized in that, Step 4 specifically comprises the following steps: for each candidate in the valuation model library , calculate a match score with the customer characteristics : ； in Candidate solutions Match score; Customer behavior feature vectors set according to specific requirements The first in The weight coefficients of each input feature, and satisfying ; Candidate solutions In the Standard values ​​for each input feature dimension; Then, the candidate scheme with the minimum score is selected as the optimal charging scheme, and the calculation formula is as follows: ； wherein is the sequence number of the optimal charging scheme.

10. The dynamic short message pricing method based on customer behavior characteristics according to claim 9, wherein, In step 5, the detailed bill specifically comprises a customer ID, a charging period, a bill generation time, a final amount to be paid, and a bill state.

Images