Ott large screen advertisement intelligent monitoring and broadcasting system and method based on multi-modal agent

Abstract

The application relates to the field of advertisement technology, in particular to an OTT large-screen advertisement intelligent monitoring and broadcasting system and method based on a multi-modal intelligent agent, which comprises a time window analysis module, a verification time window is determined based on page browsing time sequence of an OTT platform and terminal running data; an operation offset compensation module, an operation offset prediction model is established based on historical running data, and an advertisement verification opportunity is predicted; an operation path planning module, a menu tree structure of the OTT platform is constructed in real time, a visual understanding model is used to automatically identify a clickable area and an operation hotspot, an optimal execution path from a current position to a target advertisement position is calculated, and a verification start time is calculated; a multi-dimensional verification execution module, multi-dimensional real-time advertisement verification is carried out on the same advertisement on different types of devices based on the verification start time, verification results of different types of devices are compared, and device difference data is obtained; and a verification strategy adjustment module, which adjusts the advertisement verification strategy according to an advertisement delivery plan and the device difference data.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of advertising technology, in particular to an OTT large-screen advertisement intelligent monitoring and broadcasting system and method based on a multi-modal intelligent agent. BACKGROUND

[0002] With the rapid evolution of the digital marketing ecosystem, OTT (Over-The-Top) content platforms have become one of the important channels for brand advertisement precise placement. In the face of the trend of increasingly personalized user behavior and highly complex interface interaction structure, advertisers and placement platforms have higher requirements for intelligent monitoring, active verification and strategy management of advertisement placement effects.

[0003] The existing advertisement monitoring and broadcasting mechanism generally has the following technical bottlenecks: first, it lacks active monitoring and broadcasting capability based on platform state, and can only record after the advertisement is passively triggered, resulting in low monitoring and broadcasting efficiency; second, it lacks an intelligent agent cooperation mechanism and cannot dynamically adjust the monitoring and broadcasting strategy according to the interface level, content layout or user behavior of different platforms; third, existing solutions rely on single modal data for identification and cannot accurately understand advertisement content, brand identification or context semantics, resulting in insufficient monitoring and broadcasting precision; fourth, it lacks the ability to adapt to personalized recommendation logic and changes in the interactive interface of OTT platforms, making it difficult to support comprehensive supervision and optimization of advertisement strategy effects.

[0004] Therefore, how to build a cooperation mechanism between an HID intelligent agent and a multi-modal large model to realize intelligent navigation of multi-level menus and active triggering of advertisement positions in OTT platforms is a problem to be solved.

[0005] Therefore, an OTT large-screen advertisement intelligent monitoring and broadcasting system and method based on a multi-modal intelligent agent are proposed. SUMMARY

[0006] The present application aims to provide an OTT large-screen advertisement intelligent monitoring and broadcasting system and method based on a multi-modal intelligent agent to realize intelligent navigation of multi-level menus and active triggering of advertisement positions in OTT platforms.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0008] The OTT large-screen advertisement intelligent monitoring and broadcasting system based on a multi-modal intelligent agent comprises:

[0009] A time window analysis module determines a verification time window based on the page browsing time sequence of the OTT platform and the terminal running data analysis of the advertisement exposure time interval;

[0010] A running offset compensation module establishes a running offset prediction model based on historical running data, and predicts the advertisement verification time according to the current running state and the verification time window;

[0011] An operation path planning module constructs a menu tree structure of the OTT platform in real time, automatically identifies a clickable area and an operation hotspot based on a visual understanding model, calculates an optimal execution path from a current position to a target ad slot, and optimizes an ad verification timing; and calculates a verification start time according to the optimized ad verification timing.

[0012] A multi-dimensional verification execution module performs multi-dimensional real-time ad verification on the same ad on different models of devices based on the verification start time, and obtains verification results; and obtains device difference data by comparing the verification results of different models of devices.

[0013] A verification strategy adjustment module intelligently adjusts an ad verification strategy according to an ad delivery plan and the device difference data.

[0014] Preferably, the terminal running data includes device CPU usage, memory occupancy, network delay time, video buffering state, platform response time, and interface loading speed.

[0015] The process of determining the verification time window includes: analyzing a page browsing time sequence of the OTT platform, extracting ad display nodes and display duration information; calculating a device performance index and a network stability coefficient based on the terminal running data; calculating an exposure time interval of an expected ad occurrence according to the ad display nodes and display duration information, in combination with the device performance index; and adjusting a start time and a duration of the exposure time interval based on the network stability coefficient, to obtain the verification time window.

[0016] Preferably, the running offset prediction model includes: a historical running data collection layer, a time sequence feature extraction layer, a multivariate regression analysis layer, and a real-time correction layer; the historical running data collection layer collects historical running data of different time periods and different platforms, including historical device CPU usage, historical memory occupancy, historical network delay time, historical video buffering state, historical platform response time, and historical interface loading speed; the time sequence feature extraction layer extracts periodic variation rules and trend features of the historical running data; the multivariate regression analysis layer establishes a running offset prediction function based on device status, network quality, and platform load, to predict a running offset; and the real-time correction layer real-time corrects a prediction result of the running offset according to a current running state, to predict an ad verification timing in combination with the verification time window.

[0017] Preferably, the menu tree structure includes: a main interface node, an application classification node, a sub-menu node, an ad slot node, and an operation path edge of the OTT platform; and each node contains position coordinates, clickable state, and hierarchical relationship information.

[0018] The verification starting time acquisition process comprises: analyzing a current interface state in real time through the visual understanding model to identify position coordinates of clickable areas and operation hotspots; calculating an operation step sequence and a path distance from a current position to a target advertisement position based on the menu tree structure to obtain an optimal execution path; determining a total execution time length of the optimal execution path in combination with terminal running data; optimizing an advertisement verification timing according to the advertisement verification timing and the total execution time length of the optimal execution path; and back-calculating an accurate time point at which a verification operation needs to be started based on the optimized advertisement verification timing as the verification starting time.

[0019] Preferably, the visual understanding model comprises: an interface element detection layer for identifying positions and types of UI components; a clickable area analysis layer for judging interaction attributes and operation priorities of interface elements; a platform feature identification layer for identifying interface styles and layout features of different OTT platforms; and an operation hotspot positioning layer for calculating center coordinates and operation ranges of clickable areas.

[0020] Preferably, the device difference data acquisition process comprises: starting advertisement verification based on the verification starting time, image acquisition of display interfaces of the same advertisement on different types of devices, calculation of multi-dimensional verification parameters of the advertisement on different devices to obtain verification results; the multi-dimensional verification parameters comprise image quality parameters, position parameters, content integrity and advertisement time length; the image quality parameters comprise image quality definition and color saturation; the position parameters comprise size proportion and position offset parameters; and analysis of difference values of the multi-dimensional verification parameters between different devices to establish a device performance difference matrix as the device difference data to quantify display effect deviations between devices.

[0021] Preferably, the adjustment process of the advertisement verification strategy comprises: extracting strategy parameters based on an advertisement delivery plan; identifying key difference factors affecting verification accuracy based on the device difference data to generate a device performance grading table and a compensation coefficient matrix; allocating verification precision levels for different priority advertisements in combination with the strategy parameters and the device performance grading table; adjusting verification thresholds of different devices according to the compensation coefficient matrix; and updating the advertisement verification strategy based on the adjusted verification thresholds and the precision levels.

[0022] Preferably, the OTT large-screen advertisement intelligent monitoring method based on a multi-modal intelligent agent comprises:

[0023] determining a verification time window by analyzing an exposure time interval of an advertisement preset based on a program time axis of an OTT platform in combination with terminal running data;

[0024] establishing a running offset prediction model based on historical monitoring data, and predicting an advertisement verification timing according to a current running state and the verification time window;

[0025] Real-time construction of a menu tree structure of the OTT platform, automatic recognition of clickable areas and operation hotspots based on a visual understanding model, calculation of an optimal execution path from a current position to a target ad position, and optimization of the ad verification timing; calculation of a verification start time according to the optimized ad verification timing;

[0026] Multi-dimensional real-time ad verification of the same ad on different models of devices based on the verification start time, to obtain verification results; comparison of the verification results of different models of devices to obtain device difference data;

[0027] Intelligent adjustment of an ad verification strategy according to an ad delivery plan and device difference data.

[0028] Compared with the prior art, the present application has the following beneficial effects:

[0029] 1. The present application can accurately predict the timing of the appearance of an ad and compensate for system running deviation through the collaborative work of the time window analysis module and the running deviation compensation module, significantly improving the accuracy of monitoring and broadcasting. Based on OTT platform page browsing timing analysis, combined with terminal running data to establish a running deviation prediction model, the timing of ad verification can be predicted in advance, avoiding the monitoring and broadcasting lag problem caused by traditional passive waiting. At the same time, through the time sequence feature extraction and multivariate regression analysis of historical running data, an accurate prediction function is established, realizing intelligent compensation for factors such as network fluctuations and device performance changes.

[0030] 2. The present application realizes the technical leap from passive monitoring to active verification through the operation path planning module and the multi-dimensional verification execution module. The visual understanding model can analyze the OTT platform interface state in real time, automatically recognize clickable areas and operation hotspots, dynamically construct a menu tree structure, and calculate the optimal execution path from the current position to the target ad position. This intelligent path planning not only improves the operation efficiency, but also adapts to the interface differences of different OTT platforms, supporting the automatic monitoring and broadcasting of mainstream OTT platforms. The multi-dimensional verification execution module can monitor multiple dimensions of the ad, such as picture quality parameters, position parameters, content integrity, and playback duration, forming a comprehensive verification system.

[0031] 3、The application effectively solves the display difference problem between different models of devices through the device difference data acquisition and verification strategy adjustment module, and establishes an intelligent adaptation mechanism. The system can automatically analyze the multi-dimensional verification parameter differences of the same advertisement on different devices, establish a device performance difference matrix, and quantify the display effect deviation between devices. Based on these difference data, the verification strategy adjustment module can dynamically adjust the verification threshold and judgment standard for different devices to ensure the consistency and comparability of the monitoring results. At the same time, through the establishment of the compensation coefficient matrix, the system can automatically compensate for the performance differences of different devices, improving the fairness and accuracy of monitoring. This cross-device compatibility mechanism enables the application to adapt to mainstream smart TV brands and models in the market, supports unified monitoring management of different manufacturer devices, provides a technical foundation for large-scale advertising monitoring, and effectively reduces system deployment and maintenance costs. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 A structure schematic diagram of an OTT large-screen advertisement intelligent monitoring system based on a multi-modal intelligent agent is provided for an embodiment of the application.

[0033] Figure 2 A structure schematic diagram of a running offset prediction model is provided for an embodiment of the application.

[0034] Figure 3 A flowchart for obtaining a verification start time is provided for an embodiment of the application.

[0035] Figure 4 A flowchart of an OTT large-screen advertisement intelligent monitoring method based on a multi-modal intelligent agent is provided for an embodiment of the application. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0037] The application proposes an OTT large-screen advertisement intelligent monitoring system and method based on a multi-modal intelligent agent, which realizes intelligent navigation of multi-level menus of an OTT platform and active triggering of an advertisement position. In order to illustrate that the method of the application can realize intelligent navigation of multi-level menus of an OTT platform and active triggering of an advertisement position, the effectiveness of the application will be illustrated from two embodiments below.

[0038] Embodiment one:

[0039] In the embodiments of the present application, the monitoring and broadcasting scene of a certain advertising company putting a slide screen advertisement on a mainstream OTT platform is taken as an example for detailed description. The company needs to monitor whether the slide screen advertisement put in the small window of the program list page of the platform is accurately displayed according to the contract requirements in real time.

[0040] Figure 1 The specific structure diagram of the OTT large screen advertisement intelligent monitoring and broadcasting system based on the multi-modal intelligent agent of the present application comprises a time window analysis module, a running offset compensation module, an operation path planning module, a multi-dimensional verification execution module and a verification strategy adjustment module. The following is described according to the contents of Figure 1

[0041] The time window analysis module determines the verification time window based on the page browsing time sequence of the OTT platform, in combination with the terminal running data analysis of the exposure time interval of the advertisement.

[0042] The terminal running data comprises device CPU usage, memory occupancy, network delay time, video buffer state, platform response time and interface loading speed.

[0043] The process of determining the verification time window comprises: analyzing the page browsing time sequence of the OTT platform, extracting the advertisement display node and display time length information; calculating the device performance index and network stability coefficient based on the terminal running data; calculating the exposure time interval of the advertisement expected to appear according to the advertisement display node and display time length information, in combination with the device performance index; adjusting the start time and duration of the exposure time interval based on the network stability coefficient to obtain the verification time window.

[0044] Specifically, the page browsing time sequence is the operation time sequence of the user in the OTT platform interface, including page loading time, slide operation time, stay time, etc.; the exposure time interval is the preset display time period of the advertisement, which is usually determined by the advertisement putting plan; and the verification time window is the time range of the system executing the advertisement verification monitoring and broadcasting operation, which needs to cover the complete display period of the advertisement.

[0045] The terminal running data is collected in real time through the system monitoring API.

[0046] The process of determining the verification time window comprises:

[0047] The system identifies the occurrence time and duration of the advertisement container element through page DOM structure analysis, for example, in the program list page of a certain OTT platform, the system identifies the slide screen advertisement container with ID “ad-banner”, and the preset display time length is 5 seconds.

[0048] ​The device performance index is calculated based on terminal running data, and the network stability coefficient is calculated based on the device performance index. The device performance index is calculated by weighted average: device performance index = 0.3 x (100-CPU usage rate) + 0.3 x (100-memory occupancy rate) + 0.2 x interface loading speed + 0.2 x platform response time; the network stability coefficient is calculated based on the variance of network delay: stability coefficient = 1 / (1+network delay variance);

[0049] According to the advertisement display node and the display time length information, the exposure time interval of the advertisement is calculated in combination with the device performance index. For example, a certain sliding screen advertisement is preset to appear within 3-8 seconds after page loading, in combination with the current device performance index of 0.8, the system calculates that the actual exposure time interval is 3.5-8.5 seconds.

[0050] The starting time and duration of the exposure time interval are adjusted based on the network stability coefficient to obtain a verification time window. When the network stability coefficient is 0.9, the system sets the verification time window to 3.0-9.0 seconds, ensuring that the possible network fluctuation influence is covered.

[0051] By comprehensively considering the platform characteristics and the device state, the occurrence time of the advertisement is accurately predicted, and the omission problem caused by the traditional fixed time monitoring is avoided, and the monitoring accuracy is improved. By introducing terminal running data such as CPU, memory, and network delay, the device performance and the network stability are calculated, the verification window is dynamically realized, and the context awareness is realized. The system is no longer blindly checked at the preset time point, but can perceive the current load and network state of the terminal, so as to dynamically adjust the expected interval of the advertisement exposure. The probability of capturing the advertisement on a weak network or a low-performance device is improved, and the initial success rate of verification is improved.

[0052] Further, as shown in Figure 2 , a running offset compensation module is run, a running offset prediction model is established based on historical running data, and an advertisement verification time is predicted according to a current running state and a verification time window. The "running offset" refers to the deviation between the actual running state of the system and the expected state, including operation delay, processing delay, transmission delay, etc. The "verification time" refers to the best time point at which the system starts the verification operation.

[0053] The operation offset prediction model comprises a historical operation data collection layer, a time sequence feature extraction layer, a multivariate regression analysis layer and a real-time correction layer. The historical operation data collection layer collects historical operation data of different time periods and different platforms, including historical device CPU usage, historical memory occupancy, historical network delay time, historical video buffer status, historical platform response time and historical interface loading speed. The time sequence feature extraction layer extracts the periodic variation law and trend characteristics of the historical operation data. The multivariate regression analysis layer establishes an operation offset prediction function based on device status, network quality and platform load to predict the operation offset. The real-time correction layer corrects the prediction result of the operation offset in real time according to the current operation state, and predicts the advertisement verification timing in combination with the verification time window.

[0054] Specifically, the historical operation data collection layer maintains a time sequence database containing 30 days of historical data, and the data sampling frequency is once per second.

[0055] The time sequence feature extraction layer uses a time series analysis method to identify the differences in usage patterns between weekdays and weekends, and between daytime and nighttime. For example, it is found that the network peak period is from 7 pm to 10 pm, and the delay increases by about 15%.

[0056] The multivariate regression analysis layer uses a multivariate linear regression model: predicted offset = a x CPU usage + β x memory occupancy + γ x network delay + δ x platform load + ε, where a, β, γ and δ are coefficients obtained by training historical data, and ε is a bias term.

[0057] The real-time correction layer dynamically adjusts the prediction result of the operation offset by using a Kalman filter algorithm. According to the dynamically adjusted operation offset and the verification time window, the advertisement verification timing is predicted.

[0058] Through historical data learning and real-time correction, the system operation offset is effectively compensated, the verification timing prediction error is reduced, and the timeliness of monitoring and broadcasting is improved. By establishing a prediction model including historical data collection and multivariate regression analysis, the "predictive compensation" of the advertisement verification timing is realized. It not only considers the current operation state, but also predicts the inherent and periodic operation delay of the system according to the historical data pattern. This makes the prediction of the verification timing upgrade from "passive adaptation" to "active prediction", improving the accuracy of performing verification operations at the precise exposure moment of the advertisement.

[0059] Further, the operation path planning module constructs a menu tree structure of the OTT platform in real time, automatically identifies clickable areas and operation hotspots based on a visual understanding model, calculates the optimal execution path from the current position to the target advertisement position, and optimizes the advertisement verification timing. The verification start time is calculated according to the optimized advertisement verification timing.

[0060] The menu tree structure comprises: a main interface node of an OTT platform, an application classification node, a sub-menu node, an advertising position node and an operation path edge; each node contains position coordinates, a clickable state and hierarchical relationship information; taking a certain OTT platform as an example, the main interface node contains first-level menus such as "home page", "TV series" and "movie", the "TV series" node contains second-level sub-menus such as "hot broadcast" and "new series", and the advertising position node is located in the right recommendation area of the program list page.

[0061] The flowchart of the acquisition process of the verification start time is as shown in the figure, comprising: Figure 3

[0062] The visual understanding model is used to analyze the current interface state in real time, and identify the position coordinates of the clickable area and the operation hot spot; for example, the "TV series" button is identified to be located at coordinates (200, 150) on the home page of a certain OTT platform, and the clickable area is 150x50 pixels. The visual understanding model comprises: an interface element detection layer, which uses the YOLO target detection algorithm to identify the position and type of UI components, such as buttons, menu items, text boxes and other interface elements; a clickable area analysis layer, which judges the interaction attributes and operation priority of the interface elements, and determines the clickable property by analyzing the visual features (color, border, shadow, etc.) and semantic information of the elements; a platform feature recognition layer, which identifies the interface style and layout features of different OTT platforms, establishes a platform feature library, and supports automatic adaptation of mainstream OTT platforms; an operation hot spot positioning layer, which calculates the center coordinates and operation range of the clickable area, and uses pixel-level accurate positioning to ensure the operation success rate.

[0063] The operation step sequence and path distance from the current position to the target advertising position are calculated based on the menu tree structure, and the optimal execution path is obtained; for example, the path from the home page to the advertising position is: home page→TV series→hot broadcast→program list→advertising position, which requires a total of 4 steps of operation.

[0064] The total execution time of the optimal execution path is determined in combination with terminal running data; the operation time of each step includes a click time (50ms), an interface response time (dynamically adjusted according to the platform response time), and a page loading time (calculated according to the network status), totaling about 2.5 seconds.

[0065] The advertising verification timing is optimized according to the total execution time of the optimal execution path; if the predicted advertising appears after 8 seconds, the path execution requires 2.5 seconds, and the optimized verification timing is adjusted to start execution at 5.5 seconds.

[0066] Based on the optimized advertising verification timing, the accurate time point at which the verification operation needs to be started is calculated by backstepping, as the verification start time; considering the system processing delay of 100ms, the final verification start time is determined to be 5.4 seconds. ​

[0067] By constructing a menu tree and calculating an optimal path, structured understanding and efficient navigation of the OTT complex UI are achieved, which can adapt to interface differences of different platforms, so that the agent can stably find the target advertising position in different layouts and dynamically changing interfaces. At the same time, the starting time is deduced by calculating the path time consumption, ensuring the "on-time" delivery of the operation, and improving the operation success rate and the automation degree of monitoring and broadcasting.

[0068] Further, the multi-dimensional verification execution module performs multi-dimensional real-time advertisement verification on the same advertisement on different models of devices based on the verification starting time to obtain verification results; and compares the verification results of different models of devices to obtain device difference data.

[0069] The device difference data acquisition process includes:

[0070] Based on the verification starting time, the advertisement verification is started, and the display interfaces of the same advertisement on different models of devices are image collected; the system simultaneously controls different brands and models of smart televisions to collect advertisement display screens at the same time point;

[0071] The multi-dimensional verification parameters of the advertisement on different devices are calculated to obtain verification results; the multi-dimensional verification parameters include quality parameters, position parameters, content integrity, and advertisement duration. The quality parameters include picture clarity and color saturation; the picture clarity is calculated by the SSIM (Structural Similarity) algorithm, and the numerical range is 0-1, and the closer to 1 indicates that the picture is clearer; the color saturation is calculated by the saturation component of the HSV color space. The position parameters include size ratio and position offset parameters; the size ratio is obtained by calculating the ratio of the advertisement area to the total screen area; the position offset parameter is obtained by calculating the Euclidean distance between the advertisement center point and the preset position. The content integrity is detected by OCR (text recognition) and image feature matching to detect the integrity of the advertisement text, logo, and image elements, and the integrity score range is 0-1. The advertisement duration is accurately measured by video frame analysis.

[0072] The difference values of the multi-dimensional verification parameters between different devices are analyzed, and a device performance difference matrix is established as the device difference data, which quantifies the display effect deviation between devices. For example, the picture clarity of A brand television is 0.95, the picture clarity of B brand television is 0.88, and the picture clarity of C brand television is 0.92, forming a difference matrix for subsequent compensation calculation.

[0073] By establishing a standardized, multi-dimensional advertising display quality evaluation system, the advertising is monitored and compared across devices in multiple dimensions. The final output of the "device performance difference matrix" converts the vague "display effect is not good" problem into measurable, traceable, and comparable structured data, which comprehensively evaluates the advertising delivery effect, provides data support for precise monitoring, and provides an objective basis for advertising delivery optimization.

[0074] Further, the verification strategy adjustment module intelligently adjusts the advertising verification strategy according to the advertising delivery plan and device difference data. The adjustment process of the advertising verification strategy includes:

[0075] Based on the advertising delivery plan, the strategy parameters are extracted, including advertising priority (three levels of high, medium, and low), delivery frequency (per hour / day), time requirement (real-time / delayed), budget level, etc.

[0076] Based on the device difference data, the key difference factors affecting verification accuracy are identified, and a device performance classification table and a compensation coefficient matrix are generated. The devices are classified into A (high-end), B (mid-end), and C (low-end) based on performance, and the compensation coefficients of each level of device are calculated.

[0077] Combined with the strategy parameters and the device performance classification table, the verification precision levels are allocated for different priority advertisements. High priority advertisements are allocated high precision monitoring (error <1%), medium priority advertisements are allocated standard precision (error <3%), and low priority advertisements are allocated basic precision monitoring (error <5%).

[0078] According to the compensation coefficient matrix, the verification threshold of different devices is adjusted. For example, the picture quality threshold of C-level devices is adjusted from 0.9 to 0.8, and the position offset threshold is adjusted from 5 pixels to 8 pixels.

[0079] Based on the adjusted verification threshold and precision level, the advertising verification strategy is updated.

[0080] By actively using device difference data, the verification strategy is adjusted in reverse, such as setting a more relaxed picture quality threshold for devices with poor performance, or allocating more stringent verification standards for high priority advertisements. This enables the system to continuously evolve during use, realizes intelligent adjustment and personalized configuration of the strategy, improves the adaptability and precision of monitoring, and reduces the false positive rate.

[0081] Through the cooperative work of the five modules, the automated advertisement verification with high precision, high robustness and self-adaptive optimization capability is realized. First, the accurate timing of advertisement exposure is accurately predicted by combining real-time terminal performance and historical data model, which solves the verification failure problem caused by device delay and network fluctuation in traditional monitoring and broadcasting, and improves the monitoring and broadcasting efficiency; second, through the visual understanding and path planning capability, the system can operate the complex and variable OTT interface, effectively overcoming the problem of script failure caused by UI change; on this basis, the system can not only confirm the existence of the advertisement, but also carry out multi-dimensional quantitative evaluation and find the display difference across devices; finally, it can use these difference data for self-learning, intelligently adjust the subsequent verification strategy, so that the whole monitoring and broadcasting system becomes more and more accurate and efficient with the growth of running time.

[0082] Embodiment two:

[0083] In embodiment one, the method proposed by the application successfully realizes intelligent navigation of the multi-level menu of the OTT platform and active triggering of the advertising position. In order to further verify the effectiveness of the application, the OTT large-screen advertisement intelligent monitoring and broadcasting method based on multi-modal intelligent agent is also proposed in the embodiment of the application to monitor and broadcast the advertisements of another advertisement placement company on the OTT platform. Referring to Figure 4 , Figure 4 The specific flowchart of the method of the application is as follows.

[0084] Based on the page browsing time sequence of the OTT platform, the exposure time interval of the advertisement is analyzed by combining terminal running data, and the verification time window is determined; the terminal running data includes device CPU usage, memory occupancy, network delay time, video buffer state, platform response time and interface loading speed;

[0085] The process of determining the verification time window includes: analyzing the page browsing time sequence of the OTT platform, extracting the advertisement display node and display time information; calculating the device performance index and network stability coefficient based on the terminal running data; calculating the exposure time interval of the expected appearance of the advertisement according to the advertisement display node and display time information, and combining the device performance index; based on the network stability coefficient, the start time and duration of the exposure time interval are adjusted to obtain the verification time window.

[0086] Further, a running offset prediction model is established based on historical running data, and the advertisement verification timing is predicted according to the current running state and the verification time window;

[0087] The operation offset prediction model comprises a historical operation data collection layer, a time sequence feature extraction layer, a multiple regression analysis layer and a real-time correction layer; the historical operation data collection layer collects historical operation data of different time periods and different platforms, including historical device CPU usage, historical memory occupancy, historical network delay time, historical video buffer status, historical platform response time and historical interface loading speed; the time sequence feature extraction layer extracts periodic variation rules and trend features of the historical operation data; the multiple regression analysis layer establishes an operation offset prediction function based on device status, network quality and platform load to predict operation offset; and the real-time correction layer corrects the prediction result of the operation offset in real time according to the current operation status, and predicts the advertisement verification timing in combination with the verification time window.

[0088] The real-time correction layer further comprises an abnormal state sensing unit that monitors whether the current terminal operation data has a sudden abnormality that deviates significantly from the historical time sequence features (such as system pop-up windows and network instantaneous congestion); when an abnormality is detected, the operation offset compensation module temporarily increases the predicted value of the operation offset, and issues an instruction to the operation path planning module to re-plan or delay execution. The system is provided with the ability to cope with sudden conditions, and the running robustness and risk avoidance ability of the system in a real and complex environment are enhanced. Through abnormal state sensing, the system can handle sudden events (such as system pop-up windows and APP crash precursors) that cannot be predicted by the historical model, and immediately take avoidance measures (such as delaying verification and re-planning). This enables the system to work stably in a chaotic reality from a prediction model running in an ideal environment.

[0089] Further, a menu tree structure of the OTT platform is constructed in real time, the visual understanding model is used to automatically identify clickable areas and operation hotspots, the optimal execution path from the current position to the target advertisement position is calculated, and the advertisement verification timing is optimized; and the verification start time is calculated according to the optimized advertisement verification timing.

[0090] The menu tree structure comprises a main interface node, an application classification node, a sub-menu node, an advertisement position node and an operation path edge of the OTT platform; each node contains position coordinates, clickable state and hierarchical relationship information;

[0091] The verification start time acquisition process comprises: analyzing the current interface state in real time through the visual understanding model to identify the position coordinates of the clickable areas and operation hotspots; calculating the operation step sequence and path distance from the current position to the target advertisement position based on the menu tree structure to obtain the optimal execution path; determining the total execution time of the optimal execution path in combination with the terminal operation data; optimizing the advertisement verification timing according to the advertisement verification timing and the total execution time of the optimal execution path; and based on the optimized advertisement verification timing, the precise time point at which the verification operation needs to be started is calculated reversely to serve as the verification start time.

[0092] The visual understanding model comprises: an interface element detection layer that identifies the position and type of a UI component; a clickable area analysis layer that determines the interaction attribute and operation priority of the interface element; a platform feature identification layer that identifies the interface style and layout features of different OTT platforms; and an operation hotspot positioning layer that calculates the center coordinates and operation range of the clickable area.

[0093] Further, based on the verification start time, multi-dimensional real-time advertisement verification is performed on the same advertisement on different models of devices to obtain verification results; and by comparing the verification results of different models of devices, device difference data is obtained.

[0094] The device difference data is obtained by: starting advertisement verification based on the verification start time, image acquisition of the display interface of the same advertisement on different models of devices, calculation of multi-dimensional verification parameters of the advertisement on different devices to obtain verification results; the multi-dimensional verification parameters include image quality parameters, position parameters, content integrity and advertisement duration; the image quality parameters include image clarity and color saturation; the position parameters include size ratio and position offset parameters; and by analyzing the difference values of the multi-dimensional verification parameters between different devices, a device performance difference matrix is established as the device difference data, and the display effect deviation between devices is quantified.

[0095] The device performance difference matrix is a multi-dimensional tensor structure; in addition to the device model dimension, it also includes the OTT application version dimension and the operating system version dimension; the system can analyze the changes in advertisement display effects on the same device caused by application or system updates, thereby deepening the adjustment accuracy of the verification strategy from the device level to the application and system version level. The granularity of the strategy optimization is deepened from "device" to "advertisement material" itself, and a fine-grained verification strategy is achieved. The system can identify the performance problems of specific types of advertisement creatives on specific devices and dynamically weight the strategies accordingly. This makes verification and optimization no longer a one-size-fits-all approach, but rather a precise targeting and solution to specific problems caused by the incompatibility of advertisement materials and devices.

[0096] Further, based on the advertisement delivery plan and the device difference data, the advertisement verification strategy is intelligently adjusted.

[0097] The adjustment process of the advertisement verification strategy comprises: extracting strategy parameters based on the advertisement delivery plan; identifying key difference factors that affect verification accuracy based on the device difference data to generate a device performance classification table and a compensation coefficient matrix; assigning verification precision levels to different priority advertisements in combination with the strategy parameters and the device performance classification table; adjusting the verification threshold of different devices according to the compensation coefficient matrix; and updating the advertisement verification strategy based on the adjusted verification threshold and precision level.

[0098] The adjustment process of the advertisement verification strategy also includes feature analysis of the advertisement material itself. The system extracts the material features of the advertisement to be verified (such as video advertisements, picture advertisements, and unique identifiers of the advertisements in the creative library); when the display effects of different devices on a specific type of advertisement material are significantly different (for example, a specific video encoding is prone to screen flashing on a certain type of device), the compensation coefficient matrix will be dynamically weighted for the combination of the advertisement material features and the device model, achieving more targeted strategy adjustment for specific problems. A "version-level" deep problem attribution capability is provided. By extending the difference matrix to the application version and the system version, when an advertisement display problem occurs, the system can not only locate which device, but also accurately locate the problem that occurs after the device updates a certain App or OS version. This provides extremely valuable and highly actionable intelligence for platform developers to fix bugs, shortening the problem troubleshooting and repair cycle.

[0099] While embodiments of the present application have been shown and described with reference to a few embodiments, it will be understood by those skilled in the art that various changes in form and details can be made therein without departing from the spirit and scope of the application which is defined by the appended claims and their equivalents.

Claims

1. An OTT large-screen advertising intelligent monitoring system based on multimodal intelligent agents, characterized in that: include: The time window analysis module analyzes the exposure time range of advertisements based on the page browsing time sequence of the OTT platform and combined with terminal operation data to determine the verification time window; The runtime offset compensation module establishes a runtime offset prediction model based on historical runtime data, and predicts the timing of ad verification based on the current runtime status and verification time window. The operation path planning module constructs the menu tree structure of the OTT platform in real time, automatically identifies clickable areas and operation hotspots based on a visual understanding model, calculates the optimal execution path from the current location to the target ad position, and optimizes the timing of ad verification. Calculate the verification start time based on the optimized ad verification timing; The multi-dimensional verification execution module performs multi-dimensional real-time ad verification on the same ad on different models of devices based on the verification start time, and obtains the verification results; it compares the verification results of different models of devices to obtain device difference data. The verification strategy adjustment module intelligently adjusts the advertising verification strategy based on advertising campaign plans and device differences data.

2. The OTT large-screen advertising intelligent monitoring system based on multimodal intelligent agents according to claim 1, characterized in that: The terminal operation data includes device CPU utilization, memory usage, network latency, video buffering status, platform response time, and interface loading speed. The process of determining the verification time window includes: parsing the page browsing sequence of the OTT platform and extracting the ad display nodes and display duration information; calculating the device performance index and network stability coefficient based on the terminal operation data; calculating the expected exposure time interval of the ad based on the ad display nodes and display duration information and the device performance index; and adjusting the start time and duration of the exposure time interval based on the network stability coefficient to obtain the verification time window.

3. The OTT large-screen advertising intelligent monitoring system based on multimodal intelligent agents according to claim 1, characterized in that: The operational offset prediction model includes: a historical operational data acquisition layer, a time-series feature extraction layer, a multivariate regression analysis layer, and a real-time correction layer. The historical operational data acquisition layer collects historical operational data from different time periods and platforms, including historical device CPU utilization, historical memory usage, historical network latency, historical video buffering status, historical platform response time, and historical interface loading speed. The time-series feature extraction layer extracts the periodic variation patterns and trend characteristics of the historical operational data. The multivariate regression analysis layer establishes an operational offset prediction function based on device status, network quality, and platform load to predict operational offsets. The real-time correction layer corrects the predicted operational offsets in real time based on the current operational status and combines this with the verification time window to predict the timing of ad verification.

4. The OTT large-screen advertising intelligent monitoring system based on multimodal intelligent agents according to claim 1, characterized in that: The menu tree structure includes: main interface nodes of the OTT platform, application category nodes, sub-menu nodes, ad slot nodes, and operation path edges; each node contains location coordinates, clickable status, and hierarchical relationship information; The process of obtaining the verification start time includes: analyzing the current interface state in real time through the visual understanding model to identify the location coordinates of clickable areas and operation hotspots; calculating the sequence of operation steps and path distance from the current location to the target ad position based on the menu tree structure to obtain the optimal execution path; determining the total execution time of the optimal execution path by combining terminal operation data; optimizing the ad verification timing based on the ad verification timing and the total execution time of the optimal execution path; and calculating the precise time point at which the verification operation needs to be started based on the optimized ad verification timing, as the verification start time.

5. The OTT large-screen advertising intelligent monitoring system based on multimodal intelligent agents according to claim 4, characterized in that: The visual understanding model includes: an interface element detection layer to identify the position and type of UI components; a clickable area analysis layer to determine the interaction attributes and operation priorities of interface elements; a platform feature recognition layer to identify the interface style and layout features of different OTT platforms; and an operation hotspot positioning layer to calculate the center coordinates and operation range of the clickable area.

6. The OTT large-screen advertising intelligent monitoring system based on multimodal intelligent agents according to claim 1, characterized in that: The process of acquiring the device difference data includes: initiating ad verification based on the verification start time, capturing images of the display interface of the same ad on different models of devices, calculating multi-dimensional verification parameters of the ad on different devices, and obtaining verification results; the multi-dimensional verification parameters include image quality parameters, position parameters, content integrity, and ad duration; image quality parameters include image clarity and color saturation; position parameters include size ratio and position offset parameters; analyzing the differences in the multi-dimensional verification parameters between different devices, establishing a device performance difference matrix, and using it as the device difference data.

7. The OTT large-screen advertising intelligent monitoring system based on multimodal intelligent agents according to claim 1, characterized in that: The adjustment process of the advertising verification strategy includes: extracting strategy parameters based on the advertising campaign plan; identifying key difference factors affecting verification accuracy based on the device difference data, and generating a device performance grading table and a compensation coefficient matrix; assigning verification accuracy levels to ads of different priorities by combining the strategy parameters and the device performance grading table; adjusting the verification thresholds for different devices according to the compensation coefficient matrix; and updating the advertising verification strategy based on the adjusted verification thresholds and accuracy levels.

8. A method for intelligent monitoring of OTT large-screen advertisements based on multimodal intelligent agents, characterized in that: include: Based on the page browsing time sequence of the OTT platform, combined with terminal operation data analysis, the exposure time interval of advertisements is determined to identify the verification time window; A runtime offset prediction model is established based on historical runtime data to predict the timing of ad verification based on the current runtime status and verification time window. The menu tree structure of the OTT platform is constructed in real time, and clickable areas and operation hotspots are automatically identified based on a visual understanding model. The optimal execution path from the current location to the target ad position is calculated, and the timing of ad verification is optimized. Calculate the verification start time based on the optimized ad verification timing; Based on the verification start time, multi-dimensional real-time ad verification is performed on the same ad on different models of devices to obtain verification results; the verification results of different models of devices are compared to obtain device difference data. The system intelligently adjusts ad verification strategies based on ad delivery plans and device differences.

Images