An advertising information management system and method

Abstract

The application is specifically an advertising information management system and method, relating to the technical fields of advertising management and sensor data processing, comprising: in response to an advertising display or interaction event of a terminal device, locking a time window with the event occurrence time as the center, and obtaining historical sensor data within the time window based on a ring buffer. In the application, unique characteristics of human interaction are mined from the physical layer, such characteristics cannot be simulated by cheating behaviors such as simulator volume brushing, API click injection, real machine group control, etc., significantly improving the reliability of anti-cheating and covering single machine and cluster cheating scenarios comprehensively; at the same time, the original sensor data is processed by the terminal local DSP / NPU, ensuring that the data does not leave the device, meeting the privacy compliance requirements of GDPR, personal information protection law, etc., avoiding the delay and security risks of cloud data transmission, and providing core protection for advertisers to accurately calculate the delivery effect and reduce budget waste.

Description

Technical Field

[0001] This invention relates to the fields of advertising distribution management and sensor data processing technology, and in particular to an information management system and method for advertising distribution. Background Technology

[0002] With the rapid development of the mobile internet, mobile advertising has become a core carrier of digital marketing. However, the problem of false exposure is common in the advertising process, such as emulator click fraud, API-injected clicks, and real device group control operations, which leads to wasted advertisers' budgets and an inability to accurately evaluate the effectiveness of advertising.

[0003] Existing anti-fraud technologies primarily rely on software-level behavioral feature detection, such as analyzing click frequency, user activity patterns, and device fingerprints. However, these methods have significant drawbacks: Cheating can evade detection by simulating normal operating procedures or altering device fingerprints; The false alarm rate is high, and a single software feature cannot distinguish between normal operations and cheating behavior in special scenarios. Unable to counter cheating by using multiple real phones, these groups operate by fixing hundreds of real phones on metal frames. Their software features are indistinguishable from those of normal users, making them difficult to identify using traditional methods.

[0004] Therefore, there is an urgent need for an information management method for advertising that can break through the bottleneck of traditional software-layer detection, build a reliable anti-fraud system from the physical layer, and take into account low power consumption, high precision, and privacy compliance. Summary of the Invention

[0005] The purpose of this invention is to provide an information management system and method for advertising publication in order to solve the above-mentioned problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: An information management method for advertising placement, comprising: In response to advertising displays or interactive events on terminal devices, a time window centered on the moment the event occurs is locked, and historical sensor data within the time window is obtained based on a circular buffer. Collect multimodal sensor data from the terminal device within a time window. The multimodal sensor data includes at least acceleration data and angular velocity data, and perform spatiotemporal alignment on the data. Physical feature analysis is performed on multimodal sensor data to extract frequency domain features reflecting the motion state of the equipment and time domain features reflecting the interactive mechanical response; The frequency domain features and time domain features are input into a preset fusion scoring model to calculate the authenticity probability, and the authenticity probability is used to determine whether the ad display or interaction event is a valid exposure.

[0007] Preferably, the step of responding to an advertisement display or interactive event on a terminal device, locking a time window centered on the moment the event occurs, and acquiring historical sensor data within the time window based on a circular buffer, specifically includes: During the SDK initialization phase, a contiguous, fixed-size memory region is allocated as a level-one cache pool. The cache pool is logically structured as a circular queue with a capacity of 2000 frame nodes. Each frame node contains: timestamp, raw accelerometer reading, raw gyroscope reading, ambient light level, and marker bits; Sampling strategy: Status A: When the current application is in the background, or when no ad container is loaded in the foreground; Status B: When the preset lifecycle callback method of the ad container is called and the visible area ratio exceeds the preset ratio; State C: The instant a finger touch is detected on the screen or an advertisement video begins to play.

[0008] Preferably, the method further includes: When the event is triggered It happened, and the timestamp is Immediately acquire a read-write lock on the circular buffer to pause the overwrite of new data; Calculate the target start time ; Locating the timestamp in the circular buffer is close to Frame index; In the circular buffer array, the corresponding Starting from the index of the data frame, read the buffer data clockwise until the index points to... The corresponding current frame; Copy this data to a separate authentication session object.

[0009] Preferably, the acquisition terminal device collects multimodal sensor data within a time window. The multimodal sensor data includes at least acceleration data and angular velocity data, and the data is spatiotemporally aligned, specifically including: Request configuration of sensor hardware parameters, including accelerometer configuration, gyroscope configuration, and special processing of light sensor and barometer; The collected raw data is preprocessed, including deburring and zero-bias calibration; and the processed data is aligned on the time axis and spatial axis. Specifically, time axis alignment establishes a unified time axis, mapping all sensor values ​​onto the axis to generate a synchronized matrix. .

[0010] Preferably, the step of performing physical feature analysis on multimodal sensor data to extract frequency domain features reflecting the motion state of the equipment and time domain features reflecting the interactive mechanical response specifically includes: Biological micro-tremor analysis based on power spectral density: Input the aligned sequence of acceleration magnitudes; Construct a high-pass filter and determine its cutoff frequency and transfer function; Output dynamic acceleration sequence ; right Applying a Hanning window results in a windowing signal. ; right Perform a fast Fourier transform to calculate the one-sided power spectral density.

[0011] Preferably, the method further includes: Define the following metrics: DC / Low Frequency Band : Corresponds to the active swaying of the human body; Biological frequency band : Corresponds to muscle tremors; Mechanical high frequency band : Corresponds to motor vibration or environmental noise; Calculate the proportion of bioenergy ; Judgment Logic: like < It is determined to be a static support / simulator; It is the sum of the energies of all frequency bands in the spectrum. like >0.15, and the spectrum has no sharp single peak: determined to be held by a real person; If there is a spike in the spectrum at 50Hz: it is determined to be mains interference / group control rack resonance.

[0012] Preferably, the method further includes click mechanical response analysis based on jerk: Calculate the first derivative of acceleration ; At the time when the click event occurs Nearby windows [ -50ms, Search within +50ms] maximum value ; Existence verification: If <Sensor noise floor threshold This was determined to be an API injection attack. If the existence verification passes, extract all local maxima points within a preset time after the click and fit an exponentially decaying envelope. ; Material verification: If the attenuation coefficient If the value exceeds the preset threshold: it is determined to be rigidly fixed; like If the value falls within the preset handheld range: it is determined as a handheld click; If there is no attenuation: it is determined to be interference from the vibration motor.

[0013] Preferably, the method further includes: Multi-sensor causality and cross-correlation tests: The illumination and acceleration sequences are normalized, and the normalized light rate of change sequence is calculated. With acceleration magnitude sequence Cross-correlation function, search delay time Maximum correlation coefficient in the range [-100ms, +100ms] It performs timing logic decisions, involving optical flow characteristics and synchronization characteristics; Anti-group control detection based on group coherence: Select a segment of acceleration data during the stationary period and calculate the phase spectrum; The server receives phase data from K devices under the same IP subnet and calculates the phase difference consistency between the devices at key frequencies. ; like If the difference between 1 and 1 is less than the preset threshold, it indicates that multiple devices are vibrating at the same frequency.

[0014] Preferably, the step of inputting frequency domain features and time domain features into a preset fusion scoring model, calculating the authenticity probability, and determining whether an ad display or interaction event is a valid exposure based on the authenticity probability specifically includes: Determine the current device attitude based on the DC component of the accelerometer: And define the state: State A: Handheld mode; State B: Desktop mode; State C: Stand mode; Based on the corresponding state, determine the weight factors of each parameter in the physical feature weight vector; Calculate the probability score of authenticity: ; in, In order to target the Dynamic weighting coefficients for each physical feature; For the first Likelihood ratio of each feature; Preset effective ad exposure threshold, if If the exposure exceeds the preset effective exposure threshold, it is considered an effective exposure.

[0015] Preferably, the advertising publishing information management system includes: The multimodal sensor data acquisition module is configured based on a three-state finite state machine and a circular buffer to respond to advertising display / interaction events and realize multimodal sensor data acquisition and time window data extraction. The multi-source data standardization preprocessing module is configured to address the fragmentation problem in the mobile sensor ecosystem by configuring hardware parameters, heartbeat interpolation, data cleaning, and spatiotemporal alignment. The multi-dimensional physical feature analysis module is configured to extract frequency domain biological micro-tremor features and time domain click mechanical response features, and combine multi-sensor causal cross-correlation test and population coherence analysis to construct a physical feature recognition system covering single-machine and cluster cheating. The dynamic fusion decision and exposure management module is configured to dynamically adjust feature weights based on device posture and calculate the probability of authenticity through a likelihood ratio scoring model.

[0016] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention mines unique human interaction features from the physical layer. These features cannot be simulated by fraudulent behaviors such as simulator click fraud, API injection clicks, and real device group control, significantly improving the reliability of anti-fraud measures and comprehensively covering both single-machine and cluster fraud scenarios. At the same time, it uses the terminal's local DSP / NPU to process raw sensor data, ensuring that the data does not leave the device and complies with privacy compliance requirements such as GDPR and personal information protection laws. This not only avoids the latency and security risks of cloud data transmission, but also provides a core guarantee for advertisers to accurately calculate the effectiveness of their campaigns and reduce budget waste.

[0017] 2. This invention achieves low-power, high-real-time acquisition of sensor data through a collaborative design of a three-state finite state machine and a circular buffer. The memory pre-allocation strategy avoids CPU jitter or GC stuttering caused by instantaneous memory allocation during advertising interactions, balancing power consumption and data accuracy requirements, and adapting to the battery life requirements of mobile devices. At the same time, it constructs a unified standardized sensor data pipeline, and effectively solves the problem of inconsistent data quality caused by the fragmentation of the mobile sensor ecosystem by means of fine-tuning hardware parameters, heartbeat interpolation, spatiotemporal alignment, etc., improving the accuracy of physical feature analysis and enhancing the compatibility of the technical solution with multiple terminals such as smartphones and VR / AR headsets. Attached Figure Description

[0018] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 This is a structural diagram of the method of the present invention; Figure 2 This is a system flowchart of the present invention. Detailed Implementation

[0019] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0020] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0021] Example 1

[0022] Its specific implementation method is combined with the appendix Figure 1 and attached Figure 2 Please provide a detailed explanation.

[0023] Appendix Figure 1 The flowchart of an advertising publishing information management system and method provided in this embodiment of the invention illustrates the complete steps from responding to advertising display or interactive events on terminal devices to calculating the probability of authenticity.

[0024] Appendix Figure 2 This invention provides a structural block diagram of an advertising publishing information management system and method, which shows the connection relationship between the multimodal sensor data acquisition module and the dynamic fusion decision and exposure management module, and marks the main functional interaction flow of each module.

[0025] In this embodiment, it includes: Before detailing the steps, the execution environment must be defined; This invention operates on intelligent terminal devices that include microelectromechanical systems (MEMS) sensors. These terminal devices include, but are not limited to, smartphones, tablets, wearable devices (such as VR / AR headsets), and intelligent in-vehicle control screens. Sensor module: Integrated with a six-axis or nine-axis inertial measurement unit (IMU), including a three-axis micromachined accelerometer and a three-axis micromachined gyroscope. Optional accessories include a magnetometer, ambient light sensor (ALS), barometer, and capacitive touchscreen. The terminal must have digital signal processing (DSP) capabilities or an NPU (neural network processing unit) to execute local feature extraction algorithms, ensuring that raw sensor data does not leave the device and meeting privacy compliance requirements.

[0026] In response to advertising displays or interactive events on terminal devices, a time window centered on the moment the event occurs is locked, and historical sensor data within the time window is obtained based on a circular buffer. Specifically, it includes: When the system starts, it first initializes a sensor management module in the memory heap of the terminal device; Pre-allocation of memory: In order to avoid CPU jitter or GC (garbage collection) lag caused by memory allocation (malloc / new) at the moment of ad click, the system requests a contiguous, fixed-size memory region as a level 1 cache pool during the SDK initialization phase. The cache pool is logically structured as a circular queue, with a set queue capacity. It has 2000 frame nodes; if the sampling rate is 100Hz, this capacity is sufficient to store historical data from the past 20 seconds, far exceeding the required 1-second time window, to provide ample safety redundancy; Each frame node contains: a nanosecond-level timestamp based on system startup time, raw accelerometer readings, raw gyroscope readings, ambient light level, and a flag bit (used to indicate data validity or hardware malfunction). Each field can be retrieved from the device.

[0027] The system maintains two atomic operation pointers: Pointing to the latest write position, Points to the oldest data location; The write strategy follows the overwrite logic: when catch up hour, Automatically move forward and discard the oldest frame of data; Sampling strategy: To resolve the trade-off between power consumption and accuracy, this invention designs a three-state finite state machine to control the sensor's sampling frequency: State A: Sleep monitoring state: Triggering condition: When the current application is in the background, or when no ad container is loaded in the foreground; Execution logic: Unregister all high-frequency sensor listeners, and only retain large-amplitude accelerometer interrupts or UI touch callback interfaces of the operating system.

[0028] State B: Preheating-ready state: Triggering condition: When the preset lifecycle callback method (onAttachedToWindow) of the ad container (AdView) is called, and the visible area ratio exceeds the preset ratio (30%); onAttachedToWindow is a lifecycle callback method for the View class in Android development.

[0029] Execution logic: The system sends instructions to the Hardware Abstraction Layer (HAL) to preheat the IMU sensor.

[0030] Set the sampling frequency to intermediate frequency mode (e.g., 20Hz). The purpose of this stage is to fill the historical data segment of the circular buffer to ensure that there is enough preceding data when a click occurs.

[0031] Note: No feature calculations are performed at this time; only a simple data writing operation is performed.

[0032] State C: Full-speed capture state: Triggering conditions: The instant a finger touches the screen or an advertisement video begins to play; Execution logic: Immediately increase the sensor sampling rate to high-frequency mode (100Hz or the maximum frequency supported by the hardware). Lock the current timestamp. Activate the countdown latch mechanism and set the subsequent recording duration. .

[0033] When the event is triggered It happened, and the timestamp is At this time, the system executes the following key algorithm steps, which are typically performed in a separate background thread (WorkerThread) in the actual code implementation to avoid blocking the UI main thread: Immediately acquire a read-write lock on the circular buffer to suspend the overwrite of new data (or redirect new data to a temporary queue) to prevent data races during the reading process; Calculate the target start time ; Using binary search within the circular buffer, the system quickly locates the timestamp closest to the nearest one. The frame index; this step solves the performance loss caused by simple array traversal and ensures microsecond-level positioning in a large buffer; In the circular buffer array, corresponding to the backtracking start time Starting from the index of the data frame, read the buffer data clockwise until the index points to... The corresponding current frame; Copy this data (representing the approach and press phases) to a separate validation session object; Backward time-delay recording: Release the buffer lock; Start a high-precision timer for 500ms. During this period, newly acquired sensor data is not only written to the circular buffer, but also directly appended to the aforementioned verification session object.

[0034] Collect multimodal sensor data from the terminal device within a time window. The multimodal sensor data includes at least acceleration data and angular velocity data, and perform spatiotemporal alignment on the data. Specifically, it includes: This addresses data quality issues at the hardware physical level. Due to the severe fragmentation of the Android / iOS ecosystem and the significant differences in the underlying sensor drivers from different manufacturers, a unified data standardization pipeline must be established at the application layer.

[0035] Specific configuration requests are made for sensor hardware parameters, including accelerometer configuration, gyroscope configuration, and special processing for light sensors and barometers; Accelerometer configuration details: Range: The preferred setting is ±4g. Although users rarely exceed 2g in daily operations, the instantaneous acceleration generated by the impact force when hitting a hard surface can reach more than 3g. Setting it to the default ±2g will cause waveform clipping, resulting in the loss of crucial high-frequency characteristics. Bandwidth: Set half of the sampling rate of the internal low-pass filter (DLPF) cutoff frequency, such as 50Hz, to prevent high-frequency noise aliasing.

[0036] Gyroscope configuration details: Measurement range: The preferred setting is ±2000 dps (degrees per second). This covers all scenarios, from minor vibrations to violent shaking of the phone.

[0037] Special treatment for light sensors and barometers: Since most mobile phones' ambient light sensors (ALS) and barometers are environmental sensors, their driving strategy is usually on-change rather than continuous. That is, data is only called back when the value changes.

[0038] Design a heartbeat interpolation mechanism: if no light sensor callback is received within 100ms, the system will automatically copy the previous reading and insert it into the current time point to forcibly construct a continuous time series.

[0039] 100ms is the maximum duration for which the system allows the light sensor to remain silent.

[0040] The collected raw data is preprocessed, including deburring and zero-bias calibration; and the processed data is aligned on the time axis and spatial axis. Specifically, time axis alignment establishes a unified time axis, mapping all sensor values ​​onto this axis to generate a synchronized matrix. ; Abnormal data cleaning and preprocessing: The collected raw data often contains noise and outliers, and needs to undergo the following cleaning process: De-glitching: Detecting pulse noise in sensor readings. For example, an acceleration abruptly changes from 1g to 10g and immediately recovers in the next frame. This is usually due to low-level register read errors or electrostatic interference. A median filter is used for processing with a window size of 3 frames.

[0041] Zero-bias calibration: While factory-level calibration cannot be performed on user phones, relative changes can be utilized. Before feature extraction, the mean value for the first 100ms of the time window is calculated, and this mean value is subtracted from all subsequent data. This eliminates fixed biases caused by sensor aging or temperature drift.

[0042] Spatiotemporal alignment of multi-source heterogeneous data: Time axis alignment: Different sensors have independent clock crystals and reporting delays, and Lagrange interpolation or monotonic cubic spline interpolation is used.

[0043] Spatial axis alignment: The coordinate system of a mobile phone sensor is relative to the screen. Users hold the phone in various postures (landscape, portrait, flat, 45-degree tilt). In order to extract uniform physical features, the data must be projected onto a world coordinate system.

[0044] The above spatiotemporal alignment process is a direct reference to existing technology and will not be elaborated upon here.

[0045] Physical feature analysis is performed on multimodal sensor data to extract frequency domain features reflecting the motion state of the equipment and time domain features reflecting the interactive mechanical response; Specifically, it includes: When human muscles maintain isometric contraction, due to the time delay of the spinal reflex circuit, physiological tremors with a frequency concentrated in the range of 8-12 Hz are generated. In contrast, mechanical supports exhibit a highly rigid static state (0 Hz), while electric treadmills or vibration motors exhibit strong harmonics at specific frequencies.

[0046] Biological micro-tremor analysis based on power spectral density: Input aligned acceleration magnitude sequence ; Since gravity g≈9.8m / The DC component will mask the weak jitter signal. The system constructs a 4th-order Butterworth high-pass filter and determines: Cutoff frequency Transfer function (Z-domain) ; Output dynamic acceleration sequence ; To reduce spectrum leakage caused by data truncation, Apply Hanning window: Receive windowing signal , ;in, This is the index of the current sampling point; 0.5 is the total window length; 0.5 is the normalization coefficient. right Perform a Fast Fourier Transform: , ; Calculate the one-sided power spectral density ; The sampling frequency; Define three key frequency band energies: DC / Low Frequency Band (0.5-3Hz): Corresponds to active shaking of the human body; Biological frequency band (7-13Hz): Corresponds to muscle tremors; Mechanical high frequency band (15-40Hz): Corresponds to motor vibration or environmental noise; Calculate the proportion of bioenergy : , To prevent the denominator from being zero, the minimum value is obtained. Judgment Logic: like < It is determined to be a static support / simulator; It is the sum of the energy of all frequency bands (from 0Hz to 20Hz) in the spectrum, representing the total intensity of the phone's vibration at that instant.

[0047] like >0.15, and the spectrum has no sharp single peak: determined to be held by a real person; If there is a spike in the spectrum at 50Hz: it is determined to be mains interference / group control rack resonance.

[0048] It also includes click mechanics response analysis based on jerk: The physical process of tapping the screen follows the law of conservation of momentum. The device receives a reverse impulse the instant a finger strikes the screen. Because a phone is not a rigid body (due to the phone case and hand muscle damping), the acceleration is not abrupt but rather exhibits a damped oscillating waveform. Software simulations of tapping do not capture this physical process.

[0049] To highlight the impact force at the moment of impact, the first derivative of the acceleration is calculated. (i.e., jerk); a five-point center difference formula is used to improve noise immunity: , The sampling interval; At the time when the click event occurs Nearby windows [ -50ms, Search within +50ms] maximum value ; Existence verification: If <Sensor noise floor threshold This indicates that there was no physical disturbance at the moment of clicking, which is determined to be an API injection attack; If the existence verification passes, further analysis of the waveform after the click is performed. The actual physical vibration should conform to the response of a second-order underdamped system. The existing formula is used; all local maxima points within a preset time (200ms) after a click are extracted, and an exponentially decaying envelope is fitted. ; The attenuation coefficient in the response formula of a second-order underdamped system Merging, ; The envelope of the oscillation is obtained as follows , Initial amplitude represents the amount of force applied when clicking the screen.

[0050] Material verification: If the attenuation coefficient If the value exceeds the preset threshold (waveform disappears instantly): it is determined to be rigidly fixed (such as placed on a table). Although it may be a real person, it needs to be combined with the weight reduction processing in subsequent steps. like If the value is within the preset handheld range (palm flesh damping): it is determined as a handheld click; If there is no attenuation (continuous oscillation): it is determined to be interference from the vibration motor.

[0051] Also includes: Multi-sensor causality and cross-correlation tests: To eliminate the dimensional differences between illumination and acceleration, the illumination and acceleration sequences are normalized, and the normalized light rate of change sequence is calculated. With acceleration magnitude sequence Cross-correlation function: , Search delay time Maximum correlation coefficient in the range [-100ms, +100ms] , To perform sequential logic decisions: Optical flow characteristics: inspection At any given moment, does the light exhibit a negative pulse (darkening)? Note: If the ambient light is too dim (<10 Lux), this step will be automatically skipped to avoid misjudgment; Synchronicity characteristic: If The preset threshold (0.6) and optimal latency ≈0 indicates that the vibration and occlusion are caused by the same event (finger tap); Reverse teaching (abnormal situation): If a regular change in light (such as flickering) is detected, but the accelerometer does not respond, it is determined to be a physical attack that uses screen flickering to interfere with the light sensor; Anti-group control detection based on group coherence: This is an advanced defense against swarm control of real mobile phones. The swarm control studio fixed hundreds of real mobile phones to metal frames. Although these phones are real, their background noise has a very strong spatial correlation.

[0052] Select a period of acceleration data during a stationary phase (non-click period); Calculate the phase spectrum , It is the imaginary part; For real part; The server receives phase data from K devices under the same IP subnet and calculates the phase difference consistency between the devices at key frequencies. : , Number of time windows / number of observation points; Phase difference; This is Euler's formula; Judgment Logic: like If the difference between 1 and 1 is less than the preset threshold, it means that multiple devices are vibrating at the same frequency (for example, when a person walks across the floor of the server room, all the mobile phones on the shelf generate micro-vibrations at the same time). That is, close to 1; Conclusion: Even if this phone passes all the above individual tests, if it resonates with other phones on the same network segment, it will be directly identified as a group control device and added to the blacklist.

[0053] The frequency domain features and time domain features are input into a preset fusion scoring model to calculate the authenticity probability, and the authenticity probability is used to determine whether the ad display or interaction event is a valid exposure. Specifically, it includes: A single feature (such as micro-vibration or light sensitivity) may fail under certain conditions.

[0054] For example, a user might tap the phone while it's flat on a table (no slight tremor from holding it) or in a dark environment (no change in light sensitivity). Relying solely on a single feature would lead to an extremely high false positive rate. Therefore, a scene-aware, dynamically weighted fusion network is designed to dynamically adjust the confidence weights of each feature based on the current physical environment.

[0055] Before fusion, the system first determines the current device attitude based on the DC component of the accelerometer (gravity vector g): Calculate the tilt angle: Calculate the angle between the Z-axis and the direction of gravity. ; And define the state: A-mode handheld mode: ∈[30°, 150°] and acceleration variance > ; This is the preset minimum value; Status B Desktop Mode: ≈0 or 180° and acceleration variance Approaching 0; Status C support mode: It remains constant (constant value) but exhibits high-frequency mechanical vibration; Based on the corresponding state, determine the physical feature weight vector. The weighting factors of each parameter in the equation. For biological micro-tremor weights; Click bounce weight; Cross-correlation weights; For group control resonance weight; Calculate the probability score of authenticity: ; in, In order to target the Dynamic weighting coefficients for physical characteristics (such as micro-vibration, rebound, etc.); For the first The likelihood ratio of a feature represents the ratio of the probability of that feature appearing in real human samples to the probability of it appearing in cheating samples. For prior probability, The preset average real rate of all users on the network. The preset average cheating rate for all users across the network; Preset effective ad exposure threshold, if If the exposure exceeds the preset effective ad exposure threshold, it is considered an effective exposure. like If the exposure is less than the preset effective exposure threshold, it is considered an invalid exposure and will be processed accordingly.

[0056] Example 2

[0057] Please see Figure 2 An information management system for advertising distribution includes the following components: The multimodal sensor data acquisition module is configured based on a three-state finite state machine and a circular buffer to respond to advertising display / interaction events, and to achieve low-power, high-real-time acquisition of multimodal sensor data (acceleration, angular velocity, etc.) and accurate extraction of time window data. The multi-source data standardization preprocessing module is configured to address the fragmentation problem in the mobile sensor ecosystem. Through fine-tuning of hardware parameters, heartbeat interpolation, data cleaning, and spatiotemporal alignment, it constructs a unified data standardization pipeline to ensure data quality. The multi-dimensional physical feature analysis module is configured to extract frequency domain biological micro-tremor features and time domain click mechanical response features, and combine multi-sensor causal cross-correlation test and population coherence analysis to construct a physical feature recognition system covering single-machine and cluster cheating. The dynamic fusion judgment and exposure management module is configured to dynamically adjust feature weights based on device posture and calculate the probability of authenticity through a likelihood ratio scoring model, thereby achieving accurate judgment of effective exposure and anomaly handling for ad display / interaction events.

[0058] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0059] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0060] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0061] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0062] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0063] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0064] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0065] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0066] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0067] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. An information management method for advertising publication, characterized in that, include: In response to advertising displays or interactive events on terminal devices, a time window centered on the moment the event occurs is locked, and historical sensor data within the time window is obtained based on a circular buffer. Collect multimodal sensor data from the terminal device within a time window. The multimodal sensor data includes at least acceleration data and angular velocity data, and perform spatiotemporal alignment on the data. Physical feature analysis is performed on multimodal sensor data to extract frequency domain features reflecting the motion state of the equipment and time domain features reflecting the interactive mechanical response; The frequency domain features and time domain features are input into a preset fusion scoring model to calculate the authenticity probability, and the authenticity probability is used to determine whether the ad display or interaction event is a valid exposure.

2. The advertising publishing information management method according to claim 1, characterized in that, In response to advertising displays or interactive events on terminal devices, a time window centered on the moment the event occurs is locked, and historical sensor data within the time window is acquired based on a circular buffer, specifically including: During the SDK initialization phase, a contiguous, fixed-size memory region is allocated as a level-one cache pool. The cache pool is logically structured as a circular queue with a capacity of 2000 frame nodes. Each frame node contains: timestamp, raw accelerometer reading, raw gyroscope reading, ambient light level, and marker bits; Sampling strategy: Status A: When the current application is in the background, or when no ad container is loaded in the foreground; Status B: When the preset lifecycle callback method of the ad container is called and the visible area ratio exceeds the preset ratio; State C: The instant a finger touch is detected on the screen or an advertisement video begins to play.

3. The advertising publishing information management method according to claim 2, characterized in that, Also includes: When the event is triggered It happened, and the timestamp is Immediately acquire a read-write lock on the circular buffer to pause the overwrite of new data; Calculate the target start time ; Locating the timestamp in the circular buffer is close to Frame index; In the circular buffer array, the corresponding Starting from the index of the data frame, read the buffer data clockwise until the index points to... The corresponding current frame; Copy this data to a separate authentication session object.

4. The advertising publishing information management method according to claim 1, characterized in that, The system collects multimodal sensor data from the terminal device within a time window. This multimodal sensor data includes at least acceleration and angular velocity data, and performs spatiotemporal alignment on the data. Specifically, this includes: Request configuration of sensor hardware parameters, including accelerometer configuration, gyroscope configuration, and special processing of light sensor and barometer; The collected raw data is preprocessed, including deburring and zero-bias calibration; and the processed data is aligned on the time axis and spatial axis. Specifically, time axis alignment establishes a unified time axis, mapping all sensor values ​​onto the axis to generate a synchronized matrix. .

5. The advertising publishing information management method according to claim 1, characterized in that, Physical feature analysis is performed on multimodal sensor data to extract frequency domain features reflecting the motion state of the equipment and time domain features reflecting the interactive mechanical response. Specifically, this includes: Biological micro-tremor analysis based on power spectral density: Input the aligned sequence of acceleration magnitudes; Construct a high-pass filter and determine its cutoff frequency and transfer function; Output dynamic acceleration sequence ; right Applying a Hanning window, a windowing signal is obtained. ; right Perform a fast Fourier transform to calculate the one-sided power spectral density.

6. The advertising publishing information management method according to claim 5, characterized in that, Also includes: Define the following metrics: DC / Low Frequency Band : Corresponds to the active swaying of the human body; Biological frequency band : Corresponds to muscle tremors; Mechanical high frequency band : Corresponds to motor vibration or environmental noise; Calculate the proportion of bioenergy ; Judgment Logic: like < It is determined to be a static support / simulator; It is the sum of the energies of all frequency bands in the spectrum. like >0.15, and the spectrum has no sharp single peak: determined to be held by a real person; If there is a spike in the spectrum at 50Hz: it is determined to be mains interference / group control rack resonance.

7. The advertising publishing information management method according to claim 6, characterized in that, It also includes click mechanics response analysis based on jerk: Calculate the first derivative of acceleration ; At the time when the click event occurs Nearby windows [ -50ms, Search within +50ms] maximum value ; Existence verification: If <Sensor noise floor threshold This was determined to be an API injection attack. If the existence verification passes, extract all local maxima points within a preset time after the click and fit an exponentially decaying envelope. ; Material verification: If the attenuation coefficient If the value exceeds the preset threshold: it is determined to be rigidly fixed; like If the value falls within the preset handheld range: it is determined as a handheld click; If there is no attenuation: it is determined to be interference from the vibration motor.

8. The advertising publishing information management method according to claim 7, characterized in that, Also includes: Multi-sensor causality and cross-correlation tests: The illumination and acceleration sequences are normalized, and the normalized light rate of change sequence is calculated. With acceleration magnitude sequence Cross-correlation function, search delay time Maximum correlation coefficient in the range [-100ms, +100ms] It performs timing logic decisions, involving optical flow characteristics and synchronization characteristics; Anti-group control detection based on group coherence: Select a segment of acceleration data during the stationary period and calculate the phase spectrum; The server receives phase data from K devices under the same IP subnet and calculates the phase difference consistency between the devices at key frequencies. ; like If the difference between 1 and 1 is less than the preset threshold, it indicates that multiple devices are vibrating at the same frequency.

9. The advertising publishing information management method according to claim 1, characterized in that, The frequency domain features and time domain features are input into a pre-defined fusion scoring model to calculate the authenticity probability. Based on the authenticity probability, it is determined whether the ad display or interaction event is a valid exposure. Specifically, this includes: Determine the current device attitude based on the DC component of the accelerometer: And define the state: State A: Handheld mode; State B: Desktop mode; State C: Stand mode; Based on the corresponding state, determine the weight factors of each parameter in the physical feature weight vector; Calculate the probability score of authenticity: ； in, In order to target the Dynamic weighting coefficients for each physical feature; For the first Likelihood ratio of each feature; Preset effective ad exposure threshold, if If the exposure exceeds the preset effective exposure threshold, it is considered an effective exposure.

10. An advertising publishing information management system, comprising an advertising publishing information management method according to any one of claims 1-9, characterized in that, include: The multimodal sensor data acquisition module is configured based on a three-state finite state machine and a circular buffer to respond to advertising display / interaction events and realize multimodal sensor data acquisition and time window data extraction. The multi-source data standardization preprocessing module is configured to address the fragmentation problem in the mobile sensor ecosystem by configuring hardware parameters, heartbeat interpolation, data cleaning, and spatiotemporal alignment. The multi-dimensional physical feature analysis module is configured to extract frequency domain biological micro-tremor features and time domain click mechanical response features, and combine multi-sensor causal cross-correlation test and population coherence analysis to construct a physical feature recognition system covering single-machine and cluster cheating. The dynamic fusion decision and exposure management module is configured to dynamically adjust feature weights based on device posture and calculate the probability of authenticity through a likelihood ratio scoring model.

Images