A digital marketing management system and method for telecommunications services

Abstract

The present invention relates to the technical field of Internet of Things, and specifically to a digital marketing management system and method for telecommunication services. The system includes: a user behavior monitoring module, a path trajectory extraction module, a dynamic work and rest modeling module, a behavior cycle analysis module, and a user portrait generation module. In the present invention, through the group feature modeling of cellular communication data, the change rate of behavior trends is accurately calculated, the group behavior characteristics are dynamically depicted, and the extraction of the high-frequency area node sequence transforms the analysis of behavior trajectories from a single-point discrete state into a path association mode, realizing the recognition of group activity patterns. Based on the cross-analysis of group residence duration statistics and multi-dimensional behavior indicators, a time distribution model is constructed to identify stable behavior intervals and typical patterns, improving the analytical ability of group behavior characteristics. Combining holiday feature recognition with hot spot area distribution characteristics, the marketing strategy matching is optimized, and the accuracy of the group recommendation strategy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of the Internet of Things, and in particular, to a digital marketing management system and method for telecommunications services. Background Art

[0002] The technical field of the Internet of Things includes systems that achieve device interconnection, information collection, and analysis through intelligent devices, network communication, and data processing technologies. The core content of this technical field includes a perception layer, a network layer, and an application layer. Among them, the perception layer is mainly composed of devices such as sensors, RFID, and cameras, and is responsible for collecting environmental data and device status information in the physical world; the network layer realizes communication between devices and servers through wired or wireless communication technologies, such as cellular networks, Wi-Fi, Bluetooth, LoRa, etc.; the application layer relies on cloud computing, big data analysis, and artificial intelligence technologies to complete data storage, analysis, and application services, such as smart homes, industrial automation, and smart cities. The systematic development of the overall technical field of the Internet of Things involves multiple links such as data collection, transmission, processing, and application, aiming to build an intelligent and automated digital world and improve resource utilization and operation efficiency.

[0003] Among them, a digital marketing management system for telecommunications services refers to a system that realizes precise marketing and customer relationship management based on the telecommunications network environment, combined with data analysis and intelligent computing capabilities. This system mainly involves cluster analysis of group behavior characteristics, precise positioning of target customer groups, dynamic generation and optimization of marketing content, and implementation of multi-channel marketing plans. Specific methods include feature mining based on group behavior statistical models, prediction of group preference trends using machine learning models, generation of categorized marketing content through natural language processing technology, and distribution through channels such as text messages, social media, and APP push. In addition, this system also optimizes subsequent marketing strategies through real-time collection and analysis of feedback data, and improves the intelligent level of marketing management.

[0004] The prior art relies on static data statistics and lacks calculation of time series distribution characteristics, resulting in the limitation of user feature characterization to fixed patterns. Behavioral trajectory analysis stays at discrete position points and fails to associate continuous state migrations, affecting the recognition of group behavior patterns. The modeling of time distribution characteristics does not combine multi-dimensional statistical indicators, making it difficult to accurately judge typical behavior periods. Behavioral prediction relies on fixed time windows and does not consider periodic change offsets, reducing the tracking ability of irregular behavior patterns. Marketing strategies are based on rule matching and fail to optimize in combination with dynamic behavior characteristics, affecting accuracy and user interaction effects. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art, and to propose a digital marketing management system and method for telecommunications services.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A digital marketing management system for telecommunications services includes:

[0007] The user behavior monitoring module obtains cellular communication records, collects event types, signal strengths, communication frequencies, and residence durations, sorts them according to timestamps, and calculates the mean maximum fluctuation rate by partition to obtain the event intensity change rate value within the continuous section of the user;

[0008] The path trajectory extraction module determines the trajectory state based on the event intensity change rate value, collects base station numbers, access status timestamps, sorts them to generate a node sequence, and counts the handover frequency to obtain a high-frequency path node sequence;

[0009] The dynamic work and rest modeling module detects the stable residence duration based on the high-frequency path node sequence, collects multi-region residence data within the same period, counts the downlink request and uplink event frequencies, combines the frequency values, determines the stable interval, filters out the repeatedly occurring periods, arranges the daily distribution, accumulates the number of days of occurrence, constructs a work and rest model, and obtains the core work and rest period distribution value;

[0010] The behavior cycle analysis module calls the continuous daily work and rest time according to the core work and rest period distribution value, detects the period offset, filters out the offset periods, and arranges the work and rest nodes to obtain the dynamic prediction result of the work and rest cycle offset.

[0011] As a further solution of the present invention, the event intensity change rate value within the continuous section of the user includes dynamic indicators, extreme value indicators, and fluctuation indicators. The high-frequency path node sequence includes core nodes, time series associations, and handover identifiers. The core work and rest period distribution value includes main period identifiers, distribution characteristics, and concentration indicators. The dynamic prediction result of the work and rest cycle offset includes cycle offset, offset amplitude, and dynamic trend.

[0012] As a further solution of the present invention, the user behavior monitoring module includes:

[0013] The cellular communication data acquisition sub-module obtains cellular communication record data, collects event types, signal strengths, communication frequencies, and residence durations, and sorts all data according to timestamps to obtain time-series cellular communication data;

[0014] The event signal analysis sub-module calculates the mean and maximum value for the signal strength, communication frequency, and residence duration within the same time interval based on the time-series cellular communication data, divides the multiple data into time series, and identifies the change trend within the differential interval to obtain the interval signal statistical characteristics;

[0015] The continuous section intensity change calculation sub-module calls the interval signal statistical characteristics, divides the data according to timestamps, and uses the formula:

[0016] ;

[0017] Calculate the rate of change of signal strength within a continuous section to obtain the user's signal change rate value within the continuous section;

[0018] Among them, represents the rate of change of signal strength, represents the signal strength at the th time point, represents the total number of data points within the time window of the selected section, represents the total duration of the time window, represents the difference in signal strength between adjacent time points, represents the time point index.

[0019] As a further solution of the present invention, the path trajectory extraction module includes:

[0020] The trajectory state determination sub-module calculates the rate change difference between adjacent time points based on the event strength change rate value, determines whether the trajectory state has changed. If the rate change exceeds the set threshold, it is marked as a trajectory switching point, the switching point set is called, and the corresponding timestamps are matched to establish a trajectory switching time series;

[0021] The node sequence generation sub-module calls the trajectory switching time series, extracts the base station number and access status in chronological order, matches the timestamps and sorts them to generate an initial node sequence, filters the continuously accessed base station numbers, removes duplicate and abnormal data, and constructs a trajectory node sequence;

[0022] The high-frequency path analysis sub-module calls the trajectory node sequence, counts the base station switching frequency, calculates the connection strength between nodes according to the frequency, and uses the formula:

[0023] ;

[0024] Perform operations to obtain the influence weight between nodes, filter high-frequency nodes above the threshold, and establish a high-frequency path node sequence;

[0025] Among them, represents the node influence weight, represents the number of switches, p , q represents the node index, and r represents the index of other nodes.

[0026] As a further solution of the present invention, the dynamic work and rest modeling module includes:

[0027] The residence period screening sub-module detects the multi-region stable residence duration based on the high-frequency path node sequence, collects the residence data of multiple regions within the same period, calculates the residence time ratio of multiple regions, and screens the periods with the residence time ratio exceeding the set threshold to obtain the high-residence ratio periods;

[0028] The stable interval judgment sub-module counts the downlink request frequency and uplink event frequency of the corresponding period based on the high-residence ratio periods, calculates the combined frequency value, judges whether the period belongs to the stable interval, and screens the periods that repeatedly appear in multiple different days to obtain the stable residence interval distribution;

[0029] The core daily routine construction sub-module sorts out the distribution of multiple periods within the day based on the stable residence interval distribution, accumulates the number of days when the same period appears, and uses the formula:

[0030] ;

[0031] Performs operations to obtain the core daily routine period distribution value, and sorts out the time interval within the day to obtain the core daily routine period distribution;

[0032] Wherein, represents the core daily routine period distribution value, represents the cumulative duration of the stable residence period on the s-th day, represents the combined frequency value of the period on the s-th day, represents the total residence duration of the whole day on the s-th day, represents the number of repeated occurrences of the period on the s-th day, represents the statistical number of days, and s represents the day index.

[0033] As a further solution of the present invention, the behavior cycle analysis module includes:

[0034] The period offset screening sub-module calls the consecutive-day daily routine time based on the core daily routine period distribution value, calculates the period intervals of multiple time points, obtains the daily routine frequency change value of the differential time period, screens the data points with significant period changes, extracts the offset amplitude and direction of multiple daily routine time points, and calculates the standardized value and cumulative offset rate of the offset period to obtain the offset period data;

[0035] The daily routine node sorting sub-module calls the offset period data, detects the daily routine time points with high frequency, calculates the offset trend between adjacent time points, judges the stability of the time nodes, screens the time points with a lower fluctuation range, and re-divides the daily routine cycle nodes to obtain the stable daily routine cycle nodes;

[0036] The cycle offset prediction sub-module calls the stable daily routine cycle nodes, calculates the offset trend and fluctuation range of multiple time points, establishes a cycle offset calculation model, and uses the formula:

[0037] ;

[0038] Calculate the cycle offset composite index, obtain the offset trend vector, analyze the offset change rate in combination with past data, and generate the dynamic prediction result of the rest cycle offset;

[0039] Among them, represents the cycle offset composite index, represents the offset on the t-th day, represents the offset direction coefficient on the t-th day, represents the stability coefficient on the t-th day, represents the total number of days within the statistical cycle, represents the expected value of the offset, represents the regularization factor, and t represents the cycle day index.

[0040] As a further solution of the present invention, the system further includes:

[0041] The user portrait generation module calls the dynamic prediction result of the rest cycle offset, identifies the holiday pattern label, collects the high-frequency path nodes and the core stop characteristics, statistically analyzes the spatio-temporal distribution, generates the user behavior portrait, and obtains the dynamic behavior portrait parameter set;

[0042] The dynamic behavior portrait parameter set includes pattern labels, path characteristics, and spatio-temporal parameters.

[0043] As a further solution of the present invention, the user portrait generation module includes:

[0044] The rest activity variability analysis sub-module calls the dynamic prediction result of the rest cycle offset, calculates the activity offset value of the user in different time periods, and uses the formula:

[0045] ;

[0046] Operate to obtain the daytime activity variability index and obtain the rest offset distribution parameter;

[0047] Among them, represents the daytime activity variability index, represents the activity record at the -th time point, represents the total number of measured time points, represents the average activity time of all time points, represents the standard deviation of all time points, and u represents the time point index;

[0048] The holiday mode recognition sub-module compares the user's work and rest differences between weekdays and holidays based on the work and rest offset distribution parameters, counts the behavior change trends in the differential time periods, calculates the work and rest offset amounts in multiple time periods, screens the holiday behavior characteristic values according to the work and rest offset mean and standard deviation, screens the time periods with the most obvious changes, and obtains the holiday mode labels.

[0049] The spatio-temporal behavior statistics sub-module calls the holiday mode labels, collects the user's high-frequency path nodes and core stop locations, counts the stop times of multiple path nodes, calculates the proportion of the stop time in the differential area, calculates the spatio-temporal activity distribution according to the dispersion degree of the stop time, generates the user behavior portrait, and obtains the dynamic behavior portrait parameter set.

[0050] A digital marketing management method for telecommunications services, which is executed based on the above digital marketing management system for telecommunications services, includes the following steps:

[0051] S1: Obtain cellular communication records, collect event types, signal strengths, communication frequencies, and residence durations, sort them according to timestamps, and calculate the mean, maximum value, and fluctuation rate by partition to obtain the event intensity change rate values within the user's continuous section.

[0052] S2: Based on the event intensity change rate values within the user's continuous section, determine the trajectory state, collect the base station numbers, access status timestamps, sort them to generate a node sequence, and count the handover frequencies to obtain a high-frequency path node sequence.

[0053] S3: Call the high-frequency path node sequence, detect the stable residence duration, collect the residence data in multiple regions within the same time period, count the downlink request and uplink event frequencies, merge the frequency values, screen the stable intervals according to the residence duration, screen the repeatedly occurring time periods, sort out the daily distribution, accumulate the number of days of occurrence, and establish the core work and rest period distribution values.

[0054] S4: According to the core work and rest period distribution values, call the continuous-day work and rest times, detect the time period offsets, screen the offset time periods, and sort out the work and rest nodes to obtain the dynamic prediction results of the work and rest cycle offsets.

[0055] S5: Call the dynamic prediction results of the work and rest cycle offsets, combine with the telecommunications service push data, screen the active time periods of the target users, match the adaptive marketing push times, and obtain the optimized values of the user marketing reach times.

[0056] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0057] In the present invention, through the modeling of the group characteristics of cellular communication data, the change rate of behavioral trends can be accurately calculated, and the group behavioral characteristics can be dynamically characterized. The extraction of the high-frequency region node sequence transforms the analysis of behavioral trajectories from a single-point discrete state to a path association mode, enabling the accurate identification of group activity patterns. Based on the cross-analysis of group residence duration statistics and multi-dimensional behavioral indicators, a time distribution model is constructed to identify stable behavioral intervals and typical patterns, enhancing the analytical ability of group behavioral characteristics. The dynamic prediction of periodic offsets effectively tracks irregular behavioral trends. By combining holiday feature recognition and hot spot area distribution characteristics, the marketing strategy matching is optimized, and the accuracy of the group recommendation strategy is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is the system flowchart of the present invention;

[0059] Figure 2 is the flowchart of the user behavior monitoring module of the present invention;

[0060] Figure 3 is the flowchart of the path trajectory extraction module of the present invention;

[0061] Figure 4 is the flowchart of the dynamic work and rest modeling module of the present invention;

[0062] Figure 5 is the flowchart of the behavior cycle analysis module of the present invention;

[0063] Figure 6 is the flowchart of the user portrait generation module of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0064] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0065] In the description of the present invention, it should be understood that the orientation or positional relationships indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. are based on the orientation or positional relationships shown in the accompanying drawings. These are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention. In addition, in the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0066] The entire process of data acquisition, calculation, inference, and application involved in the present invention strictly adheres to current laws, regulations, and ethical norms, ensuring that the data source is legal, transparent, and effectively authorized, without infringing on personal privacy or the rights and interests of third parties; the calculation and inference processes use algorithm models that meet industry standards, and ensure the security and privacy of data processing through secure encryption and anonymization technologies, avoiding any form of bias or discriminatory output; the final application scenarios are clearly limited to legal and compliant fields, fully evaluating and avoiding potential social risks, ensuring that the technical achievements serve the public well-being and the needs of social development, and all links are subject to independent ethical review and legal compliance verification to comprehensively safeguard social morality and public interests.

[0067] Embodiment 1

[0068] Please refer to Figure 1 , the present invention provides a technical solution: A digital marketing management system for telecommunications services includes:

[0069] The user behavior monitoring module obtains cellular communication records, collects event types, signal strength, communication frequency, and residence duration, sorts them according to the time stamp, and calculates the mean maximum fluctuation rate by partition to obtain the event intensity change rate value within the continuous section of the user;

[0070] The path trajectory extraction module determines the trajectory state based on the event intensity change rate value, collects the base station number access status time stamp, sorts them to generate a node sequence, and counts the handover frequency to obtain a high-frequency path node sequence;

[0071] The dynamic work and rest modeling module detects the stable residence duration based on the high-frequency path node sequence, collects the residence data in multiple regions within the same period, counts the downlink request and uplink event frequencies, combines the frequency values, determines the stable interval, filters out the repeatedly occurring periods, arranges the daily distribution, accumulates the number of days of occurrence, constructs a work and rest model, and obtains the core work and rest period distribution value;

[0072] The behavior cycle analysis module calls the continuous daily work and rest time according to the core work and rest period distribution value, detects the time period deviation, filters out the deviated time periods, and arranges the work and rest nodes to obtain the dynamic prediction result of the work and rest cycle deviation;

[0073] The user portrait generation module calls the dynamic prediction result of the work and rest cycle deviation, identifies the holiday mode label, collects the high-frequency path nodes and core stop characteristics, counts the spatio-temporal distribution, generates a user behavior portrait, and obtains a set of dynamic behavior portrait parameters.

[0074] The rate of change of event intensity values within the user's continuous section includes dynamic indicators, extreme value indicators, and fluctuation indicators. The high-frequency path node sequence includes core nodes, temporal correlations, and switching identifiers. The distribution values of the core rest periods include main period identifiers, distribution characteristics, and centrality indicators. The prediction results of the rest cycle deviation dynamic include cycle deviation, deviation amplitude, and dynamic trends. The dynamic behavior portrait parameter set includes pattern labels, path characteristics, and spatio-temporal parameters.

[0075] Please refer to Figure 2 , the user behavior monitoring module includes:

[0076] The cellular communication data acquisition sub-module obtains cellular communication record data, collects event types, signaling strengths, communication frequencies, and residence durations, and sorts all data according to timestamps to obtain time-series cellular communication data;

[0077] Multiple acquisition devices need to be called to monitor the cellular communication behavior of users, specifically including base station devices, terminal devices, and signal relay devices. These devices respectively record the communication behavior data of users at different time points, including event types, signaling strengths, communication frequencies, and residence durations. During the data acquisition process, each base station will record the timestamp when the user accesses, and at the same time store the signaling strength, communication frequency, and residence duration information, and use a synchronous clock to align the data of different base stations in terms of time. After data acquisition, the system will sort the data according to timestamps to ensure the integrity of the time series. For example, at a certain moment T1, base station A records that the user's signaling strength is -85dBm, the residence duration is 120s, and the communication frequency is 3 times per minute, while base station B records the signaling strength as -80dBm, the residence duration as 100s, and the communication frequency as 4 times per minute at moment T2. By comparing the data of multiple base stations, the communication behavior of users in different regions can be restored, and finally complete time-series cellular communication data can be obtained.

[0078] Based on the time-series cellular communication data, the event signal analysis sub-module calculates the mean and maximum values for the signaling strength, communication frequency, and residence duration within the same time interval, and divides the multiple data into time series to identify the change trends within the differentiated intervals, obtaining the interval signal statistical characteristics;

[0079] First, the system filters the data within the selected time interval, extracts all signal strength values, communication frequency values, and residence duration values for this time period, and then calculates the mean and maximum values of these data. For example, assume that within a certain time interval T1 - T2, the collected signal strength data is [-85, -83, -80, -78, -75] dBm, the communication frequency data is [3, 5, 4, 6, 7] times per minute, and the residence duration data is [120, 110, 95, 130, 140] seconds. Then the calculated means are signal strength -80.2 dBm, communication frequency 5 times per minute, and residence duration 119 seconds, and the maximum values are signal strength -75 dBm, communication frequency 7 times per minute, and residence duration 140 seconds. Next, the data is divided into multiple sub-intervals according to the time series, and the change trend of adjacent intervals is calculated. For example, the time interval is divided into sub-intervals such as T1 - T3, T3 - T5, etc., and the mean change of each interval is calculated. If the mean signal strength in a certain interval decreases by more than 10%, the communication frequency increases by more than 30%, and the residence duration decreases by more than 20%, then this interval is identified as a significant signal change interval, and finally the interval signal statistical characteristics are obtained.

[0080] The continuous section intensity change calculation sub-module calls the interval signal statistical characteristics, divides the data according to the time stamp, and uses the formula:

[0081] ;

[0082] Calculate the signal strength change rate within the continuous section to obtain the user's continuous section signal change rate value;

[0083] Among them, represents the signal strength change rate, represents the signal strength at the th time point, represents the total number of data points within the time window of the selected section, represents the total duration of the time window, represents the difference in signal strength between adjacent time points, represents the time point index.

[0084] First, obtain all signal strength values within the selected time interval, calculate the signal strength change amount ΔSSG between adjacent time points, simultaneously count the total duration of the time window ΔTWD, and calculate the average value of all signal strength values within this time period, using the formula:

[0085] ;

[0086] Assume that within a certain time window T1 - T5, the signal strength measurement values are -85 dBm, -83 dBm, -80 dBm, -78 dBm, -75 dBm in sequence, then the calculation of ΔSSG is as follows:

[0087] ;

[0088] The total time window duration ΔTWD = 10 minutes, and the total number of data points NTP = 5, so the summation term is calculated as follows:

[0089] ;

[0090] ;

[0091] ;

[0092] ;

[0093] The result shows that the signal strength change rate during this time period is 1.9dB / minute. If the baseline value is set to 1.5dB / minute, the current change rate exceeds the baseline value, indicating that the signal changes in this time period are more drastic, which can be used to identify changes in user mobility status.

[0094] Table 1. Sample table of collected data

[0095]

[0096] As shown in Table 1, the collected signaling strength, communication frequency, and dwell time data are used to calculate the signal change trend and signal strength change rate.

[0097] See also Figure 3 , the path trajectory extraction module includes:

[0098] The trajectory state determination submodule calculates the rate change difference between adjacent time points based on the event intensity change rate value to determine whether the trajectory state has changed. If the rate change exceeds the set threshold, it is marked as a trajectory switching point, the switching point set is called, and the corresponding timestamps are matched to establish a trajectory switching time series.

[0099] First, obtain the event intensity change rate values of the device at different time points, calculate the rate changes at adjacent time points, and obtain the rate change difference set. The rate calculation formula can be expressed as:

[0100] ;

[0101] in, Indicates time The rate at Indicates time The intensity of the event, For time intervals, for example, if a device The event intensity at is 120, The event intensity at is 135 and the time interval is 2 seconds, then the rate ; For rate change, define:

[0102] ;

[0103] If the rate at the previous time point is 6.5, then , each value in the calculated rate change set is compared with the set threshold Perform comparisons, such as setting thresholds , if the calculated rate changes , then mark the current time point As the trajectory switching point, the time point Add it to the switching point set and match the timestamp to build a time series in the database to obtain the complete trajectory switching time series.

[0104] The node sequence generation submodule calls the trajectory switching time sequence, extracts the base station number and access status in chronological order, matches the timestamps and sorts them to generate the initial node sequence, filters the base station numbers of consecutive accesses, removes duplicate and abnormal data, and constructs the trajectory node sequence;

[0105] For example, if a device frequently switches between base stations A and B within a certain time period (e.g., 5 times within 3 seconds), it can be determined that the switching is affected by external factors, such as signal interference or device failure. In this case, this part of the data needs to be eliminated. For the sorting of base station numbers, they are arranged in ascending order by timestamp. For example, if the time series is {12:00, 12:02, 12:05, 12:10}, the corresponding base station number sequence is {A, B, B, C}. Then, redundant data is removed. For example, if the device is connected to base station B between 12:02 and 12:05, only one B is retained, and the trajectory node sequence {A, B, C} is finally constructed.

[0106] The high-frequency path analysis submodule calls the trajectory node sequence, counts the base station switching frequency, and calculates the connection strength between nodes based on the frequency using the formula:

[0107] ;

[0108] Calculate the influence weights between nodes, filter out high-frequency nodes above the threshold, and establish a high-frequency path node sequence;

[0109] in, Represents the node influence weight, represents the number of switching times, p , q represents the node index, and r represents the index of other nodes.

[0110] formula:

[0111] ;

[0112] Among them and represent the number of handovers between nodes p and q. If p, q, and r represent base stations A, B, and C respectively, the number of handovers is , , , then:

[0113] ;

[0114] For node pairs with influence weights greater than the threshold (for example, setting ), high-frequency nodes that meet the conditions are screened out, such as {A, B}, and finally a high-frequency path node sequence {A, B} is constructed.

[0115] Table 2 High-frequency path analysis data table:

[0116] As shown in Table 2, the connection strength between A and B in the high-frequency path analysis is the highest and meets the screening threshold. Therefore, A and B form a high-frequency path node sequence.

[0117] Please refer to Figure 4 , the dynamic work and rest modeling module includes:

[0118] The residence period screening sub-module, based on the high-frequency path node sequence, detects the multi-region stable residence duration, collects the residence data of multiple regions within the same period, calculates the residence time occupancy ratio of multiple regions, and screens out the periods with a residence time occupancy ratio exceeding the set threshold to obtain high-residence occupancy ratio periods;

[0119] First, detect the multi-region stable residence duration. During this process, the system traverses the data set, calculates the residence time of each location point, and determines whether it exceeds the set residence time threshold. The set threshold refers to the statistical historical behavior data. If the threshold is 10 minutes, the points with a residence time less than 10 minutes will be excluded, and only the points with a high residence time will be retained. Further, it is screened whether the residence regions within adjacent time periods are continuous. If the continuous residence duration reaches the set requirement, the time periods are merged and the residence information of this region is recorded. Subsequently, after obtaining the multi-region residence data, the system calculates the multi-region residence time occupancy ratio. The ratio calculation uses the formula:

[0120] ;

[0121] Among them, represents the residence occupancy ratio of region , is the total residence time of this region, is the total residence time of all regions. Subsequently, the residence ratio is screened, and the threshold is set to 0.3, that is, the regions with a residence time ratio greater than 30% will be marked as high-residence regions. Combining with the example, assuming that the user stays in regions A, B, and C for 100 minutes, 50 minutes, and 30 minutes respectively, and the total residence time is 180 minutes, then the residence ratio of each region is as follows:

[0122] ;

[0123] According to the set threshold of 0.3, only region A is screened into the high-residence ratio period, and finally the time period with a high-residence ratio is obtained and the relevant data is recorded.

[0124] Based on the high-residence ratio period, the stable interval judgment sub-module counts the downlink request frequency and uplink event frequency in the corresponding period, calculates the combined frequency value, judges whether the period belongs to the stable interval, screens the periods that appear repeatedly in multiple different days, and obtains the stable residence interval distribution;

[0125] First, traverse the selected residence time period, count the number of user downlink requests and uplink events in this period respectively, and calculate the combined frequency value. The calculation method of the combined frequency value is as follows:

[0126] ;

[0127] Among them, represents the combined frequency value of the period , is the downlink request frequency of this period, is the uplink event frequency of this period. Assuming that the number of downlink requests in a certain period is 40 and the number of uplink events is 60, then the combined frequency value is:

[0128] ;

[0129] Subsequently, the system sets the reference value of the combined frequency. Referring to historical data statistics, the threshold is set to 50, that is, if the combined frequency value of a certain period exceeds 50, it is determined that it belongs to the stable interval. Continue to screen the periods that appear repeatedly in multiple different days, and use the window sliding method to calculate whether the stable residence period of each day matches other days. If it appears repeatedly in at least 70% of the days, then it is finally determined that this period is the stable residence interval. If the statistical number of days is 10 days, then a certain period that appears more than 7 days is selected into the stable residence interval, and the interval information is stored.

[0130] Based on the stable residence interval distribution, the core daily routine construction sub-module arranges the distribution of multiple periods within the day, accumulates the number of days when the same period appears, and uses the formula:

[0131] ;

[0132] Calculate to obtain the core rest period distribution value, and organize it in the time intervals within a day to obtain the core rest period distribution;

[0133] Among them, represents the core rest period distribution value, represents the cumulative duration of the stable stay period on the s-th day, represents the combined frequency value of the period on the s-th day, represents the total duration of the all-day stay on the s-th day, represents the number of repeated occurrences of the period on the s-th day, represents the number of statistical days, and s represents the day index.

[0134] The formula is as follows:

[0135] ;

[0136] Among them, is the cumulative duration of the stable stay period on the th day, represents the combined frequency value of the period on the th day, is the total duration of the all-day stay on the th day, is the number of repeated occurrences of the period on the th day, represents the number of statistical days. Assuming that in the data of 5 days of statistics, the cumulative duration of stay, combined frequency value, total duration of all-day stay, and number of repeated occurrences per day are as follows in the table:

[0137] Table 3 Core rest parameter data table

[0138] Substitute into the calculation formula:

[0139] ;

[0140] Calculate each item:

[0141] ;

[0142] The final core rest period distribution value is 123.29. After organizing the time intervals within each day, the final distribution result of the core rest period is obtained.

[0143] Please refer to Figure 5 , the behavior cycle analysis module includes:

[0144] Based on the core rest period distribution value, the time period offset screening sub-module calls the consecutive-day rest time, calculates the time period intervals at multiple time points, obtains the rest frequency change values in different time periods, screens the data points with significant time period changes, extracts the offset amplitudes and directions at multiple rest time points, calculates the standardized value and cumulative offset rate of the offset time period, and obtains the offset time period data;

[0145] First, call the rest time records for multiple consecutive days, convert the rest time points of each day into the standard time format, extract the start time, end time, and intermediate key time nodes of the rest for each day, calculate the time differences between the corresponding time points of every two adjacent days, and obtain the offset situation of each time point. For example, assume that a person's wake-up times for 7 consecutive days are 6:30, 6:45, 6:40, 6:50, 6:35, 6:55, 6:30 respectively. Then calculate the daily offset: The second day is offset by 15 minutes compared to the first day, the third day is offset by -5 minutes compared to the second day, and so on, to obtain the time offset array within the entire cycle. , standardize this time offset array to eliminate the influence of time units on the calculation. Then, statistically analyze the offset data at different time points, obtain the offset mean, standard deviation, and cumulative offset rate of each time period, and set the screening threshold for significant changes. For example, set 2 times the standard deviation as the judgment criterion for significant changes. If the calculated standard deviation is 10 minutes, then the time points with an absolute offset value greater than 20 minutes are regarded as significant change points. Screen and record the data of these time points. For example, if the calculated offsets of 6:55 and 6:30 are +20 minutes and -25 minutes respectively, meeting the standard of significant offset, then these two time points will be screened and recorded, and used as the characteristic points of the rest time period offset. In this way, the data of the time points with significant changes within the entire rest time period are obtained, and finally, the offset time period data set is formed.

[0146] The rest node sorting sub-module calls the offset time period data, detects the rest time points that appear frequently, calculates the offset trend between adjacent time points, judges the stability of the time nodes, screens the time points with a lower fluctuation range, and re-divides the rest cycle nodes to obtain the stable rest cycle nodes;

[0147] First, extract the occurrence frequencies of all time points within the statistical cycle, calculate the cumulative occurrence times of the high-frequency time points, and set the frequency threshold. For example, if a certain time point appears 25 times within a 30-day statistical cycle, then the occurrence frequency of this time point is , if the high-frequency threshold is set to 0.8, then this time point is considered a high-frequency rest time point and marked. Then calculate the offset trend between adjacent high-frequency time points. First, calculate the average offset amplitude between adjacent high-frequency time points, that is , for example, if the offset data of the 6:30 time point are [-5, 10, -10, 5, -5], then the average offset amplitude is In minutes, the overall offset trend vector is calculated by combining multiple high-frequency time points. By comparing the offset situations at adjacent time points, it is determined whether it is stable, and the stability threshold of the offset fluctuation range is set. For example, if the offset fluctuation range at a certain time point is less than 5 minutes, then this time point is considered relatively stable. The time points with a lower offset fluctuation range are screened, and the rest cycle is re-divided based on these stable time points. For example, if the offset fluctuation ranges of 6:30, 12:00, and 22:30 are all less than 5 minutes, then these time points can be used as the stable nodes of the new rest cycle, and finally the stable rest cycle time points are output.

[0148] The cycle offset prediction sub-module calls the stable nodes of the rest cycle, calculates the offset trends and fluctuation ranges at multiple time points, establishes a cycle offset calculation model, and uses the formula:

[0149] ;

[0150] Calculate the cycle offset composite index, obtain the offset trend vector, analyze the offset change rate in combination with past data, and generate the dynamic prediction result of the rest cycle offset;

[0151] Among them, represents the cycle offset composite index, represents the offset at the t-th day, represents the offset direction coefficient at the t-th day, represents the stability coefficient at the t-th day, represents the total number of days within the statistical cycle, represents the expected value of the offset, represents the regularization factor, and t represents the cycle day index.

[0152] First, set the values of each parameter: Let the number of days within the statistical cycle be , and the offset per day be Adopt the aforementioned calculation method. For example, for the value array , the offset direction coefficient is assigned according to the positive or negative sign of the offset. For example, a positive offset is set to 1, and a negative offset is set to -1. The stability coefficient adopts the stability scores of the time points of each day. For example, the scoring standard is set as follows: the fluctuation range < 5 minutes, the score is 1; the fluctuation range is 5 - 10 minutes, the score is 0.8; the fluctuation range > 10 minutes, the score is 0.5. Then the stability score array of a certain time point may be , calculate the expected offset minutes, let the regularization factor be , and substitute it into the formula:

[0153] ;

[0154] Calculate the specific value:

[0155] ;

[0156] ;

[0157] ;

[0158] ;

[0159] ;

[0160] ;

[0161] ;

[0162] The sum is .

[0163] This result indicates that the offset fluctuation in this period is relatively large. By combining the data of previous periods to calculate the offset change rate, use to calculate the offset change rate of the previous and subsequent periods. If the change rate is greater than the set threshold, the offset trend of the rest cycle is significant, and finally a dynamic prediction result of the rest cycle offset is generated. See Table 4.

[0164] Table 4 Data Table for Calculating Rest Cycle Offset

[0165] As shown in Table 4, the time points with higher offset calculation values indicate larger fluctuations, which can be used for adjusting the rest cycle prediction in the future.

[0166] Please refer to Figure 6 , the user profile generation module includes:

[0167] The rest activity variability analysis sub-module calls the dynamic prediction result of the rest cycle offset, calculates the activity offset value of the user in different time periods, and uses the formula:

[0168] ;

[0169] Perform operations to obtain the daytime activity variability index and obtain the rest offset distribution parameter;

[0170] Among them, represents the daytime activity variability index, represents the activity record at the th time point, represents the total number of time points measured, represents the average activity time of all time points, represents the standard deviation of all time points, and u represents the time point index;

[0171] Formula:

[0172] ;

[0173] Among them, represents the daily activity variability index, represents the activity record at the th time point, is the total number of time points, is the average activity time of all time points, is the standard deviation of all time points. This process involves multiple calculation steps. First, the activity data of the user within 24 hours of a certain day is collected, and the activity status (such as walking, stationary, exercise intensity, etc.) is recorded every 15 minutes, for a total of 96 time points. For example, the activity data of a certain user at some time points is as follows:

[0174] Table 5 User Activity Data Record

[0175] As shown in Table 5, the activity data of the user at different time points is recorded and stored in the array . Then, calculate the mean value of these data:

[0176] ;

[0177] Next, calculate the standard deviation :

[0178] ;

[0179] ;

[0180] Then, calculate the exponential offset value of each time point:

[0181] ;

[0182] ;

[0183] And so on, calculate the offset values of all time points and find the mean value to obtain :

[0184] ;

[0185] This value represents the daily activity variability of the user. If it indicates that the daily routine activities are highly unstable. If it indicates that the daily routine is relatively regular. Combining with practical applications, this data can be used to predict the offset trend of the user's daily routine and provide parameter support for subsequent daily routine pattern recognition.

[0186] The holiday mode recognition sub-module compares the user's work and holiday schedules based on the rest schedule deviation distribution parameters, counts the behavior change trends in the differentiated time periods, calculates the rest schedule deviation for multiple time periods, screens the holiday behavior characteristic values according to the mean and standard deviation of the rest schedule deviation, selects the time period with the most obvious changes, and obtains the holiday mode label;

[0187] First, select the data for 7 consecutive days and calculate the :

[0188] ;

[0189] Calculate the mean by comparing weekdays (the first four days) and holidays (the last three days):

[0190] ;

[0191] ;

[0192] Calculate the standard deviation:

[0193] ;

[0194] ;

[0195] Using the and deviation as the screening criterion, when , it is determined that the user has significant rest schedule differences during holidays and is marked as "holiday mode".

[0196] The spatio-temporal behavior statistics sub-module calls the holiday mode label, collects the user's high-frequency path nodes and core stop locations, counts the stop times at multiple path nodes, calculates the proportion of stop times in the differentiated areas, calculates the spatio-temporal activity distribution based on the dispersion degree of the stop times, generates a user behavior portrait, and obtains a set of dynamic behavior portrait parameters.

[0197] First, select the stay time data of a user at different locations:

[0198] Table 6 User Spatio-Temporal Behavior Statistics

[0199] As shown in Table 6, calculate the proportion of stay time in each area, evaluate the distribution dispersion degree according to the standard deviation, and calculate the spatio-temporal activity index:

[0200] ;

[0201] ;

[0202] ;

[0203] When it indicates that the user's spatio-temporal behavior is relatively scattered. If then the user's activity area is relatively concentrated. This parameter can be used to generate a user's dynamic behavior portrait, extract high-frequency paths and core stopping positions, and combine with the changes in holiday patterns to finally form a spatio-temporal behavior feature parameter set.

[0204] A digital marketing management method for telecommunications services. The digital marketing management method for telecommunications services is executed based on the above-mentioned digital marketing management system for telecommunications services, and includes the following steps:

[0205] S1: Obtain cellular communication records, collect event types, signaling strengths, communication frequencies, and residence durations, sort them according to timestamps, calculate the mean, maximum value, and fluctuation rate by partition, and obtain the event intensity change rate value within the user's continuous section;

[0206] S2: Based on the event intensity change rate value within the user's continuous section, determine the trajectory state, collect the base station number access status timestamps, sort them to generate a node sequence, and count the handover frequency to obtain a high-frequency path node sequence;

[0207] S3: Call the high-frequency path node sequence, detect the stable residence duration, collect multi-region residence data within the same period, count the downlink request and uplink event frequencies, merge the frequency values, screen the stable interval according to the residence duration, screen the repeatedly occurring periods, sort out the daily distribution, accumulate the number of days of occurrence, and establish a core rest period distribution value;

[0208] S4: According to the core rest period distribution value, call the continuous day rest time, detect the period offset, screen the offset periods, and sort out the rest nodes to obtain the rest cycle offset dynamic prediction result;

[0209] S5: Call the rest cycle offset dynamic prediction result, combine with the telecommunications service push data, screen the active periods of target users, match the adaptive marketing push time, and obtain the optimized value of the user marketing reach period.

[0210] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A digital marketing management system for telecommunications services, characterized in that, The system includes: The user behavior monitoring module obtains cellular communication records, collects event types, signaling strengths, communication frequencies, and residence durations, sorts them according to timestamps, calculates the mean, maximum value, and fluctuation rate for each partition, and obtains the event intensity change rate values within consecutive user segments; The path trajectory extraction module determines the trajectory status based on the event intensity change rate values, collects base station numbers and access statuses, matches timestamps, sorts them to generate a node sequence, counts the handover frequencies, and obtains a high-frequency path node sequence; The dynamic daily routine modeling module detects the stable residence duration based on the high-frequency path node sequence, collects multi-region residence data within the same period, counts the frequencies of downlink requests and uplink events, combines the frequency values, determines the stable intervals, filters out the repeatedly occurring periods, organizes the daily distribution, accumulates the number of days of occurrence, constructs a daily routine model, and obtains the core daily routine period distribution values; The behavior cycle analysis module calls the consecutive-day daily routine times according to the core daily routine period distribution values, detects the period offsets, filters out the offset periods, and organizes the daily routine nodes to obtain the dynamic prediction results of the daily routine cycle offsets; The behavior cycle analysis module includes: The period offset filtering sub-module calls the consecutive-day daily routine times based on the core daily routine period distribution values, converts the daily routine time points of each day into the standard time format, extracts the start time, end time, and intermediate key time nodes of the daily routine, calculates the time differences between corresponding time points of every two adjacent days, obtains the offset situations of the time points, statistically analyzes the offset data, obtains the offset mean, standard deviation, and cumulative offset rate of the time periods, sets the screening threshold for significant changes, filters and records the data of the time points as the characteristic points of the daily routine period offsets, and obtains the offset period data; The daily routine node organizing sub-module calls the offset period data, detects the frequently occurring daily routine time points, calculates the offset trends between adjacent time points, determines the stability of the time nodes, filters out the time points with a lower fluctuation range, and re-divides the daily routine cycle nodes to obtain the stable daily routine cycle nodes; The cycle offset prediction sub-module calls the stable daily routine cycle nodes, calculates the offset trends and fluctuation ranges of multiple time points, establishes a cycle offset calculation model, and uses the formula: ； Calculates the cycle offset composite index, obtains the offset trend vector, combines the past data to analyze the offset change rate, and generates the dynamic prediction results of the daily routine cycle offsets; Among them, represents the period offset composite index, represents the offset on the t-th day, represents the offset direction coefficient on the t-th day, represents the stability coefficient on the t-th day, represents the total number of days within the statistical period, represents the expected value of the offset, represents the regularization factor, and t represents the period day index; Calls the dynamic prediction results of the daily routine cycle offsets, combines the telecom service push data, filters out the active periods of the target users, and matches the adaptive marketing push times to obtain the optimized values of the user marketing reach times.

2. The digital marketing management system for telecommunications services according to claim 1, wherein The event intensity change rate values within the consecutive user segments include dynamic indicators, extreme value indicators, and fluctuation indicators. The high-frequency path node sequence includes core nodes, temporal correlations, and handover identifiers. The core daily routine period distribution values include main period identifiers, distribution characteristics, and concentration indicators. The dynamic prediction results of the daily routine cycle offsets include cycle offsets, offset amplitudes, and dynamic trends.

3. The digital marketing management system for telecommunications services according to claim 2, wherein The user behavior monitoring module includes: The cellular communication data acquisition sub-module obtains cellular communication record data, collects event types, signaling strength, communication frequency, and residence duration, and sorts all data according to the time stamp to obtain time-series cellular communication data; The event signal analysis sub-module calculates the mean and maximum values for the signaling strength, communication frequency, and residence duration within the same time interval based on the time-series cellular communication data, and performs time series partitioning on multiple data items to identify the change trend within the differential interval, obtaining the interval signal statistical features; The continuous section intensity change calculation sub-module calls the interval signal statistical features, divides the data according to the time stamp, and uses the formula: ； Calculate the signal strength change rate within the continuous section to obtain the user's continuous section signal change rate value; Among them, represents the signal strength change rate, represents the signal strength at the th time point, represents the total number of data points within the time window of the selected section, represents the total duration of the time window, represents the difference in signal strength between adjacent time points, represents the time point index.

4. The digital marketing management system for telecommunications services according to claim 3, characterized in that, The path trajectory extraction module includes: The trajectory state determination sub-module calculates the rate change difference between adjacent time points based on the event intensity change rate value, determines whether the trajectory state has changed. If the rate change exceeds the set threshold, it is marked as a trajectory switching point, calls the switching point set, and matches the corresponding time stamp to establish a trajectory switching time series; The node sequence generation sub-module calls the trajectory switching time series, extracts the base station number and access status in chronological order, matches the time stamp and sorts them to generate an initial node sequence, filters the continuously accessed base station numbers, removes duplicate and abnormal data, and constructs a trajectory node sequence; The high-frequency path analysis sub-module calls the trajectory node sequence, counts the base station switching frequency, calculates the connection strength between nodes according to the frequency, and uses the formula: ； Perform operations to obtain the influence weight between nodes, filter the high-frequency nodes above the threshold, and establish a high-frequency path node sequence; Among them, represents the influence weight between nodes , represents the number of switches, p , q represents the node index, and r represents the index of other nodes.

5. The digital marketing management system for telecommunications services according to claim 4, wherein The dynamic work and rest modeling module includes: The residence period screening sub-module detects the stable residence duration in multiple regions based on the high-frequency path node sequence, collects the residence data in multiple regions during the same period, calculates the residence time ratio in multiple regions, and screens the periods with a residence time ratio exceeding the set threshold to obtain high-residence ratio periods; The stable interval judgment sub-module counts the downlink request frequency and uplink event frequency during the corresponding period based on the high-residence ratio periods, calculates the combined frequency value, determines whether the period belongs to the stable interval, and screens the periods that repeatedly appear in multiple different days to obtain the stable residence interval distribution; The core work and rest construction sub-module organizes the distribution of multiple periods within a day based on the stable residence interval distribution, accumulates the number of days when the same period appears, and uses the formula: ； Perform operations to obtain the core work and rest period distribution value, and organize the time interval within a day to obtain the core work and rest period distribution; Among them, represents the distribution value of the core rest period, represents the cumulative duration of the stable residence period on the s-th day, represents the combined frequency value of the period on the s-th day, represents the total residence duration of the whole day on the s-th day, represents the recurrence times of the period on the s-th day, represents the number of statistical days, and s represents the day index.

6. The digital marketing management system for telecommunications services according to claim 1, wherein The system further includes: The user portrait generation module calls the dynamic prediction result of the work and rest cycle offset, identifies the holiday mode label, collects the high-frequency path nodes and core stop characteristics, counts the spatio-temporal distribution, generates a user behavior portrait, and obtains a dynamic behavior portrait parameter set; The dynamic behavior portrait parameter set includes mode labels, path characteristics, and spatio-temporal parameters.

7. The digital marketing management system for telecom services according to claim 6, characterized in that, The user portrait generation module includes: The work and rest activity variability analysis sub-module calls the dynamic prediction result of the work and rest cycle offset, calculates the activity offset value of the user in different time periods, and uses the formula: ； Operate to obtain the daily activity variability index and get the rest schedule offset distribution parameter; Among them, represents the daytime activity variability index, represents the activity record at the th time point, represents the total number of time points measured, represents the average activity time at all time points, represents the standard deviation at all time points, and u represents the time point index; The holiday mode recognition sub-module, based on the rest schedule offset distribution parameter, compares the rest schedules of the user on weekdays and holidays, counts the behavior change trends in the differential time periods, calculates the rest schedule offsets in multiple time periods, screens the holiday behavior characteristic values according to the mean and standard deviation of the rest schedule offsets, screens the time periods with the most obvious changes, and obtains the holiday mode label; The spatio-temporal behavior statistics sub-module calls the holiday mode label, collects the user's high-frequency path nodes and core stop positions, counts the stop times of multiple path nodes, calculates the proportion of the stop time in the differential area, calculates the spatio-temporal activity distribution according to the dispersion degree of the stop time, generates the user behavior portrait, and obtains the dynamic behavior portrait parameter set.

8. A digital marketing management method for telecommunications services, characterized in that, Execute according to the digital marketing management system for telecommunications services described in any one of claims 1-7, including the following steps: S1: Obtain the cellular communication records, collect the event type, signal strength, communication frequency, and residence duration, sort them according to the time stamp, and calculate the mean, maximum value, and fluctuation rate in partitions to obtain the event intensity change rate value in the user's continuous section; S2: Based on the event intensity change rate value in the user's continuous section, determine the trajectory state, collect the base station number and access status, match the time stamp, sort to generate the node sequence, and count the handover frequency to obtain the high-frequency path node sequence; S3: Call the high-frequency path node sequence, detect the stable residence duration, collect the residence data in multiple regions in the same time period, count the downlink request and uplink event frequencies, merge the frequency values, screen the stable intervals according to the residence duration, screen the repeatedly occurring time periods, sort out the daily distribution, accumulate the number of days of occurrence, and establish the core rest schedule time period distribution value; S4: According to the core rest schedule time period distribution value, call the continuous day rest schedule time, detect the time period offset, screen the offset time periods, and sort out the rest schedule nodes to obtain the rest schedule cycle offset dynamic prediction result; S5: Call the rest schedule cycle offset dynamic prediction result, combine with the telecommunications service push data, screen the active time periods of the target users, match the adaptive marketing push time, and obtain the optimized value of the user marketing reach time period.

Images