Cross-screen interactive advertisement effect evaluation system based on multi-modal intelligent agent driving

Abstract

The application relates to the technical field of advertisement effect evaluation, in particular to a cross-screen interactive advertisement effect evaluation system based on a multi-modal intelligent agent driver, which comprises the following steps: collecting advertisement playing data of an advertisement on a target device, obtaining a login account and a geographic position of the target device, performing account verification and distance verification within an advertisement delivery time, and identifying a cross-screen associated device of the target device; monitoring a running state of the cross-screen associated device to obtain user behavior data; constructing a user cross-screen behavior link according to a time sequence of the user behavior data; performing analysis based on the advertisement playing data of the target device and the user cross-screen behavior link of the cross-screen associated device to obtain cross-screen associated behavior data; performing user conversion analysis of different advertisement position types based on the cross-screen associated behavior data; and calculating an advertisement cross-screen interaction index. The application can accurately evaluate the advertisement effect through identification of user cross-screen behavior.

Description

Technical Field

[0001] This invention relates to the field of advertising effectiveness evaluation technology, specifically to a cross-screen interactive advertising effectiveness evaluation system driven by a multimodal intelligent agent. Background Technology

[0002] With the booming development of the digital economy and the widespread adoption of various smart terminals, consumers' shopping journeys are becoming increasingly complex. Users are no longer limited to obtaining information and purchasing goods through a single device, but exhibit interactive behaviors across devices and multiple touchpoints. For example, consumers may first encounter product advertisements on smart TVs (OTT large screens), then search for products through mobile devices, and finally complete the transaction on their personal computers. This fragmented and interconnected user behavior path poses a significant challenge to traditional marketing effectiveness evaluation.

[0003] Existing marketing effectiveness evaluation methods, such as metrics based on single-device click-through rates, impressions, or conversion rates, generally struggle to fully capture and analyze user interaction data across different digital terminals (such as OTT screens, mobile devices, and personal computers). This makes it difficult for advertisers to accurately assess the actual impact of initial marketing outreach (especially on emerging channels like OTT screens) on subsequent user purchasing behavior, and also prevents them from effectively establishing a causal relationship between marketing campaigns and final sales figures.

[0004] Specifically, traditional methods struggle to fully identify and attribute cross-device user behavior conversions, potentially underestimating the ROI of marketing campaigns. For instance, even if a user doesn't directly click on a product ad seen on a smart TV, they might later actively search for and purchase it on another device. Within the current technological framework, accurately identifying and attributing such cross-platform conversions remains a significant challenge, limiting marketing decision-makers' ability to fully understand the true performance of their marketing strategies.

[0005] To address this, a cross-screen interactive advertising effectiveness evaluation system based on multimodal intelligent agents is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a cross-screen interactive advertising effectiveness evaluation system driven by multimodal intelligent agents, which can accurately evaluate the advertising effectiveness by recognizing users' cross-screen behavior.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A cross-screen interactive advertising effectiveness evaluation system driven by multimodal intelligent agents includes:

[0009] The advertising delivery module collects advertising playback data on the target device; the advertising playback data includes advertising placement type, advertising delivery time, and user feedback data.

[0010] The cross-screen association recognition module obtains the login account and geographical location of the target device, performs account verification and distance verification during the advertising period, and identifies the cross-screen associated devices of the target device.

[0011] The cross-screen behavior recognition module monitors the operating status of cross-screen associated devices to obtain user behavior data; and constructs a user cross-screen behavior chain based on the time sequence of the user behavior data.

[0012] The cross-screen attribution analysis module analyzes the cross-screen behavior data based on the ad playback data of the target device and the user cross-screen behavior links of cross-screen related devices within the associated behavior time period; the associated behavior time period is dynamically adjusted based on ad type and product characteristics.

[0013] The advertising quantitative evaluation module analyzes user conversion for different ad placement types based on a time decay attribution model of cross-screen related behavior data, obtains cross-screen interaction metrics, and recommends ad placement combinations for multiple ads.

[0014] The ad slot type refers to the playback format of the ad on the target device, including device startup ad, homepage recommendation ad, pre-video ad, mid-video ad, post-video ad, video pause ad, and device screensaver ad.

[0015] The advertising placement time refers to the time period during which the advertising is placed, including the start and end times of the placement.

[0016] The user feedback data refers to the viewing feedback from users on the target device, including ad viewing duration, ad skipping actions, ad clicking actions, and user feedback text.

[0017] The account verification process involves identifying the user's login account on the target device during the advertising period; and filtering based on the login account on the target device to obtain devices with the same account on the target device during the advertising period.

[0018] The distance verification involves obtaining the geographical locations of the target device and devices with the same account, calculating the straight-line distance between the devices, and performing interval distance verification based on the time data of the straight-line distance to determine the cross-screen associated devices.

[0019] The process of performing interval distance verification based on time series data of linear distance includes:

[0020] Continuously acquire the geographical location information of the target device and devices with the same account, and calculate the straight-line distance between them in real time to construct time series data of the straight-line distance; based on the preset distance threshold, distance change rate and geographical location co-staying pattern, analyze the time series data to determine the devices with the same account that meet the verification criteria as cross-screen associated devices.

[0021] The process of obtaining the user cross-screen behavior chain includes:

[0022] Obtain ad playback data from users on the target device;

[0023] User operation data on cross-screen connected devices is collected as user behavior data; the user behavior data includes search behavior, browsing behavior, purchase behavior and sharing behavior on cross-screen connected devices;

[0024] The collected target device ad playback data and user behavior data from multiple cross-screen related devices are arranged in chronological order to construct a timeline of user behavior on cross-screen related devices, thus obtaining the user's cross-screen behavior chain.

[0025] The process of acquiring the cross-screen related behavior data includes:

[0026] The time window is matched between the ad playback data of the target device and the cross-screen behavior links of users on cross-screen related devices to determine the time period of related behaviors after the ad playback.

[0027] Extract user feedback data and user behavior data from cross-screen related devices within the time period of the associated behavior;

[0028] Based on semantic analysis of ad content, semantic analysis of user feedback data, and semantic analysis of user behavior, the correlation between ad content and user behavior is calculated; user behaviors with a correlation greater than a preset threshold are filtered to construct cross-screen related behavior data.

[0029] The dynamic adjustment process of the associated behavior time period includes:

[0030] Determine the industry category and average order value based on product characteristics; for different ad placement types, industry categories and average order values, collect and maintain historical conversion time data of product ads, and use Bayesian optimization algorithms to dynamically search and determine the optimal time window boundary for related behaviors with user conversion rate as the optimization goal.

[0031] Semantic analysis of advertising content: Natural language processing and computer vision analysis are performed on the advertising content in the advertising playback data to extract the keywords, themes, sentiments and visual element features of the advertisement and construct the semantic vector of the advertising content;

[0032] Semantic analysis of user feedback data: Natural language processing is performed on the user feedback text in the user feedback data to extract user comments on advertisements and construct user feedback semantic vectors;

[0033] User behavior semantic analysis: Perform semantic analysis on user cross-screen behavior, including search behavior, browsing behavior, purchase behavior and sharing behavior, including search keywords, browsed product topics, purchased product categories and shared content, and construct user behavior semantic vectors;

[0034] The similarity between the semantic vectors of ad content, user feedback, and user behavior is calculated to obtain the correlation between ad content and user behavior.

[0035] The cross-screen advertising interaction metrics include user cross-screen purchase conversion rate and path contribution coefficient; the path contribution coefficient is obtained through a time decay attribution model and multiple advertising conversion attribution analyses, including:

[0036] Obtain the timeline of user behavior across screens, combine it with geolocation and device type to identify and construct cross-screen context tags; calculate the conversion rate of ad placement sequences under different cross-screen contexts, compare it with the global average conversion rate, and obtain context adjustment weights;

[0037] The time between the ad placement and the conversion is normalized based on the ad playback data, and the ad placement is assigned a corresponding attribution weight. Combined with the attribution model based on Markov chain, the baseline contribution coefficient of different ad placements in the user conversion path is analyzed.

[0038] The path contribution coefficient is calculated by fusing the baseline contribution coefficient with the context adjustment weight.

[0039] In the case of multiple ad placement combinations, ad placement combinations are recommended for different cross-screen scenarios based on the user cross-screen purchase conversion rate of multiple ads and the path contribution coefficient of different ad placements in the user conversion path.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] 1. This invention solves the problem of cross-device user identification by identifying cross-screen associated devices through login account and geolocation. By using account matching and geolocation-assisted verification, it effectively improves the accuracy of user identity association, ensures the reliability of subsequent user cross-screen behavior link construction, and avoids user behavior interruption caused by device dispersion.

[0042] 2. This invention limits user behavior within the scope of advertising influence by setting a time window, and uses semantic analysis to deeply explore the intrinsic correlation between advertising content and user behavior, filtering out behaviors that are truly influenced by the advertising, avoiding interference from irrelevant behaviors, and ensuring the accuracy of the evaluation results.

[0043] 3. This invention quantifies the correlation between advertising content and user behavior through deep semantic understanding of multimodal data, enabling the system to identify whether the intent behind user behavior is related to advertising content, thereby more accurately judging the effectiveness of advertising and improving the precision of correlation calculation. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the cross-screen interactive advertising effect evaluation system based on multimodal intelligent agent driving according to the present invention;

[0045] Figure 2 This is a logical schematic diagram of the cross-screen interactive advertising effect evaluation system based on multimodal intelligent agent driving according to the present invention;

[0046] Figure 3 This is a schematic diagram of the cross-screen associated device verification process of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Example 1:

[0049] This invention proposes a cross-screen interactive advertising effectiveness evaluation system based on multimodal intelligent agents. The system structure is as follows: Figure 1 As shown, the system's logical flow is as follows: Figure 2 As shown, it includes:

[0050] The advertising delivery module collects advertising playback data on the target device; the advertising playback data includes advertising placement type, advertising delivery time, and user feedback data.

[0051] The cross-screen association recognition module obtains the login account and geographical location of the target device, performs account verification and distance verification during the advertising period, and identifies the cross-screen associated devices of the target device.

[0052] The cross-screen behavior recognition module monitors the operating status of cross-screen associated devices to obtain user behavior data; and constructs a user cross-screen behavior chain based on the time sequence of the user behavior data.

[0053] The cross-screen attribution analysis module analyzes the cross-screen behavior data based on the ad playback data of the target device and the user cross-screen behavior links of cross-screen related devices within the associated behavior time period; the associated behavior time period is dynamically adjusted based on ad type and product characteristics.

[0054] The advertising quantitative evaluation module analyzes user conversion for different ad placement types based on a time decay attribution model of cross-screen related behavior data, obtains cross-screen interaction metrics, and recommends ad placement combinations for multiple ads.

[0055] Preferably, the above-mentioned device association and user data acquisition are conducted in an experimental environment.

[0056] Once a device is identified as a cross-screen associated device, and with the user's authorization and consent, the system collects the user's browsing, searching, and purchasing behavior by listening to specific events in the application (such as the loading event of the product details page, the submission event of the search box, and the callback event of successful payment). The collected raw data is encapsulated into a JSON object containing fields such as behavior type, time, application name, product ID / search keywords, etc., and uploaded to the server for unified cleaning and standardization processing.

[0057] Preferably, the type of ad placement is the playback format of the ad on the target device, including device startup ad, homepage recommendation ad, pre-video ad, mid-video ad, post-video ad, pause ad, and device screensaver ad.

[0058] The advertising placement time refers to the time period during which the advertising is placed, including the start and end times of the placement.

[0059] The user feedback data refers to the viewing feedback from users on the target device, including ad viewing duration, ad skipping actions, ad clicking actions, and user feedback text.

[0060] Specifically, the device boot-up ad is a full-screen ad displayed when the device starts up; the homepage recommendation ad is an ad displayed on the device's main interface or in the recommended section of the application's homepage; the pre-video ad is an ad that appears before the video content plays; the mid-video ad is an ad inserted during the video content playback; the post-video ad is an ad that appears after the video content plays; the pause ad is an ad that pops up when the video playback is paused; and the device screensaver ad is an ad displayed when the device enters screensaver mode. This categorization of ad placements helps the system to analyze the impact of different ad formats on user behavior conversion in a more refined manner.

[0061] Furthermore, the ad viewing duration refers to the actual length of time a user watches the ad; the ad skipping operation refers to the user's active choice to skip the ad; the ad clicking operation refers to the user's action of clicking the ad; and the user feedback text refers to the user's comments or text feedback on the ad content.

[0062] User feedback data provides multi-dimensional, quantitative, and qualitative information that can directly reflect users' level of interest in advertising, willingness to interact, and emotional inclinations. It is an important basis for evaluating the attractiveness and effectiveness of advertising.

[0063] This invention refines the information dimensions of advertising expression forms, enabling subsequent analysis to distinguish the differences in advertising effectiveness across different advertising formats and time periods. Meanwhile, user feedback data provides direct, multimodal interaction information between users and advertising content, offering richer data support for advertising content optimization and performance attribution.

[0064] Preferably, the account verification identifies the user's login account on the target device during the advertising period; the login account includes email account, mobile phone number and third-party account; based on the login account of the target device, devices with the same account on the target device during the advertising period are filtered.

[0065] The distance verification involves obtaining the geographical locations of the target device and devices with the same account, calculating the straight-line distance between the devices, and performing interval distance verification based on the time data of the straight-line distance to determine the cross-screen associated devices.

[0066] The verification process for cross-screen associated devices is as follows: Figure 3 As shown.

[0067] During account verification, the system will identify the user's login account on the target device and filter out other devices that log in with the same account during the advertising period as devices with the same account.

[0068] During distance verification, the system continuously acquires the geographical location information of the target device and devices with the same account, and calculates the straight-line distance between them in real time, constructing time-series data of the straight-line distance. Based on preset distance thresholds, distance change rates, and shared geographical location patterns, the system analyzes the time-series data to identify devices with the same account that meet the verification criteria as cross-screen associated devices.

[0069] Preferably, the account verification also includes ID mapping and fuzzy matching algorithms to identify non-identical account combinations belonging to the same user;

[0070] The ID mapping algorithm achieves this by establishing a unique user ID association table between different types of accounts (such as email, mobile phone number, and third-party account). This association table can be pre-built and dynamically updated based on the same personal information (such as name and ID number) filled in by users when registering on different platforms or the association information obtained when logging in with third-party authorization.

[0071] The fuzzy matching algorithm uses a semantic matching model based on deep learning to extract features from user historical login records, device fingerprints (such as unique or near-unique identifiers like operating system version, browser type, IP address, and MAC address), geographic location trajectories, and application usage habits (such as a list of frequently used applications, application usage duration, and active time periods). It then uses unsupervised or semi-supervised clustering algorithms (such as DBSCAN and Gaussian Mixture Model GMM) to divide users into groups and identify device combinations with highly similar behavioral patterns but whose login accounts are not completely identical.

[0072] For example, if user A logs in using their email account on a smart TV and their phone number on their mobile phone, but both accounts show similar geographical locations and app usage habits within the same time period, the system will make a probabilistic association and confirm the account. ID mapping and fuzzy matching algorithms can further improve the accuracy of account matching.

[0073] This invention solves the problem of cross-device user identification by identifying cross-screen associated devices through login accounts and geolocation. By using account matching and geolocation-assisted verification, it effectively improves the accuracy of user identity association, ensures the reliability of subsequent user cross-screen behavior link construction, and avoids user behavior interruption caused by device dispersion.

[0074] Preferably, the process of performing interval distance verification based on time series data of linear distance includes:

[0075] Continuously acquire the geographical location information of the target device and devices with the same account, and calculate the straight-line distance between them in real time to construct time series data of the straight-line distance; based on the preset distance threshold, distance change rate and geographical location co-staying pattern, analyze the time series data to determine the devices with the same account that meet the verification criteria as cross-screen associated devices.

[0076] In distance verification, the distance threshold can be set to 50 meters, and this range can be confirmed by verifying historical distance data of cross-screen associated devices.

[0077] The rate of change of distance refers to the change in the straight-line distance between two devices per unit time. Its threshold can be set to less than 5 meters per second, indicating that the devices are in a relatively stationary or slowly moving state.

[0078] The criteria for determining a shared geographic location pattern are: within a preset time window, the average Euclidean distance between the geographic coordinates of the two devices is less than a preset distance, and at least 70% of the geographic locations are within the same preset geofence (such as a home or office area). This determination will identify the shared location area using K-Means clustering or the DBSCAN algorithm.

[0079] Once the distance threshold, distance change rate, and geographic location all meet the same dwell mode, it is identified as a cross-screen associated device.

[0080] This invention further refines the distance verification process by introducing a combined stay mode of straight-line distance time series data, preset distance threshold, distance change rate, and geographical location. Through multi-dimensional and dynamic location data analysis, it significantly improves the accuracy and robustness of cross-screen associated device identification, especially effectively excluding devices that are geographically close by chance, accurately locking the devices used by the same user in different physical spaces, avoiding misjudgments, and ensuring the quality of data association.

[0081] Preferably, the process of obtaining the user cross-screen behavior link includes:

[0082] User operation data on cross-screen connected devices is collected as user behavior data; the user behavior data includes search behavior, browsing behavior, purchase behavior and sharing behavior on cross-screen connected devices;

[0083] The collected target device ad playback data and user behavior data from multiple cross-screen related devices are arranged in chronological order to construct a timeline of user behavior on cross-screen related devices, thus obtaining the user's cross-screen behavior chain.

[0084] Specifically, the cross-screen behavior recognition module monitors the operational status of identified cross-screen connected devices and collects user operation data on these devices, including search, browsing, purchasing, and sharing behaviors, as user behavior data. This user behavior data is arranged chronologically to construct a timeline of user behavior on cross-screen connected devices, forming a complete user cross-screen behavior chain. During the construction of the user cross-screen behavior chain, the system performs data cleaning, removing duplicate, abnormal, or invalid behavior data, and interpolating or filling in missing time points to ensure the completeness and accuracy of the behavior chain.

[0085] The identification of duplicate data is based on the complete consistency of behavior type, timestamp, and behavior content.

[0086] Anomalies are identified using statistical methods, such as Z-score analysis of user behavior duration and operation frequency, marking outliers exceeding 3 standard deviations as anomalies; or using anomaly detection algorithms based on isolated forests to identify anomalous points in behavioral patterns.

[0087] Invalid data includes data with incomplete behavior records, incorrect format, or missing key fields due to system errors, data transmission interruptions, etc. This type of data is filtered directly.

[0088] For missing time points, interpolation or padding is performed. Specifically, linear interpolation is used to pad missing data for short time intervals. For example, if a device has no operation records for a short period of time but is still active, it is inferred from the preceding and following behaviors. For missing data for long time intervals, such as when the device is turned off, no data padding is performed for that period. Instead, the behavior chain is marked as interrupted to ensure the integrity and accuracy of the behavior chain.

[0089] This invention constructs a complete cross-device user behavior trajectory, integrating fragmented user behavior data into a coherent timeline, enabling the system to track the complete path of users from ad exposure to final conversion, providing a comprehensive and clear data foundation for understanding user decision-making processes and conversion attribution.

[0090] Preferably, the process of acquiring the cross-screen related behavior data includes:

[0091] The time window is matched between the ad playback data of the target device and the cross-screen behavior links of users on cross-screen related devices to determine the time period of related behaviors after the ad playback.

[0092] Extract user feedback data and user behavior data from cross-screen related devices within the time period of the associated behavior;

[0093] Based on semantic analysis of ad content, semantic analysis of user feedback data, and semantic analysis of user behavior, the correlation between ad content and user behavior is calculated; user behaviors with a correlation greater than a preset threshold are filtered to construct cross-screen related behavior data.

[0094] The cross-screen attribution analysis module matches the ad playback data of the target device with the user cross-screen behavior links of related devices within a time window. The related behavior time period is set as a specified period from the start to the end of ad playback. This time period is the golden conversion cycle derived from user behavior data analysis and can be dynamically adjusted according to ad type and product characteristics.

[0095] The dynamic adjustment process includes: for different ad placement types (such as boot-up ads and pre-roll ads), different industry categories (such as e-commerce and education), and different product average order values ​​(such as high-priced luxury goods and low-priced daily necessities), the system will maintain its historical conversion time data separately, and use Bayesian optimization algorithms to dynamically search and determine the optimal associated behavior time window boundary in the specific scenario with user conversion rate as the optimization goal.

[0096] The process of obtaining the golden conversion period involves collecting data on the time interval between a user viewing an advertisement and completing a conversion action (such as purchasing, searching, or browsing). Statistical analysis is used to identify the peak interval of the probability distribution of user conversion time, thereby determining an initial, global golden conversion window. This golden conversion period can be dynamically adjusted according to the type of advertisement and product characteristics. This dynamic adjustment process is ongoing to adapt to market changes and the evolution of user behavior.

[0097] This invention limits user behavior within the scope of advertising influence by setting a time window, and uses semantic analysis to deeply explore the intrinsic correlation between advertising content and user behavior, filtering out behaviors that are truly influenced by the advertising, avoiding interference from irrelevant behaviors, and ensuring the accuracy of the evaluation results.

[0098] Preferably, semantic analysis of advertising content is performed: natural language processing and computer vision analysis are conducted on the advertising content in the advertising playback data. For advertising text, a pre-trained BERT model is used to extract keywords, themes, and sentiment; for advertising images and videos, ResNet and 3D-CNN models are used to extract visual element features, such as products, brand logos, and scenes, and to construct multimodal semantic vectors of the advertising content.

[0099] Semantic analysis of user feedback data: Perform natural language processing on the user feedback text in the user feedback data, such as using sentiment analysis models to identify positive, negative or neutral comments from users on advertisements, and construct semantic vectors for user feedback.

[0100] Semantic analysis of user behavior involves performing semantic analysis on user cross-screen behavior, including search behavior (search keywords), browsing behavior (browsing product topics), purchase behavior (purchasing product categories), and sharing behavior (sharing content). For example, Word2Vec or Sentence-BERT can be used to convert keywords and product topics into semantic vectors to construct semantic vectors for user behavior.

[0101] After obtaining the semantic vectors mentioned above, the semantic vectors of different modalities are fused through a fusion network to calculate the relevance between the advertising content and the user. The relevance calculation adopts weighted cosine similarity.

[0102] This invention quantifies the correlation between advertising content and user behavior through deep semantic understanding of multimodal data, enabling the system to identify whether the intent behind user behavior is related to advertising content, thereby more accurately judging the effectiveness of advertising and improving the precision of correlation calculation.

[0103] Preferably, for different ad placement types, the proportion of users who complete a purchase through cross-screen linked devices within the time range after the ad is placed is calculated, and the cross-screen purchase conversion rate of users under different ad placement types is obtained.

[0104] Based on this cross-screen related behavioral data, we analyze user conversion rates for different ad placement types and calculate cross-screen ad interaction metrics.

[0105] Specifically, for different ad placement types, the system will calculate the proportion of users who complete a purchase through cross-screen linked devices within the time range after the ad is launched, and calculate the user cross-screen purchase conversion rate under different ad placement types.

[0106] Preferably, in addition to user cross-screen purchase conversion rate, this system will also calculate the following cross-screen advertising interaction metrics:

[0107] User cross-screen browsing conversion rate: refers to the proportion of users who view the advertisement and then browse related products or services through cross-screen connected devices; User cross-screen search conversion rate: refers to the proportion of users who view the advertisement and then search for related keywords or products through cross-screen connected devices; User cross-screen sharing conversion rate: refers to the proportion of users who view the advertisement and then share the advertisement content or related products through cross-screen connected devices.

[0108] Furthermore, based on the evaluation of ad placement effectiveness, it is also possible to evaluate the effectiveness of multiple ad placement combinations.

[0109] The cross-screen advertising interaction metrics include user cross-screen purchase conversion rate and path contribution coefficient; the path contribution coefficient is obtained through a time decay attribution model and multiple advertising conversion attribution analyses, including:

[0110] Obtain the timeline of user behavior across screens, combine it with geolocation and device type to identify and construct cross-screen context tags; calculate the conversion rate of ad placement sequences under different cross-screen contexts, compare it with the global average conversion rate, and obtain context adjustment weights;

[0111] While constructing the user's cross-screen behavior chain, we extract and integrate multi-dimensional information from the chain to give users real-time contextual tags.

[0112] Time context: Based on the timestamps of ad exposure and user behavior, determine whether it is a weekday daytime, a weekday evening, or a weekend.

[0113] Geographic context: Based on the analysis of shared stay patterns in geographic locations, determine whether the user is at home, in the office, or "on the go".

[0114] Device combination scenarios: Analyze currently active cross-screen device combinations, such as "smart TV + mobile phone" (a typical home audio-visual entertainment scenario) or "personal computer + mobile phone".

[0115] All conversion data are grouped according to the constructed "cross-screen contextual tags", and the conversion rate of a specific ad placement sequence is calculated for each context; taking the ad placement combination (boot-up ad - pre-roll ad) as an example;

[0116] The overall average conversion rate for the ad mix (boot-up ad - pre-roll ad) was 3%.

[0117] In the (home - weekend evening - home entertainment) scenario, the contextual conversion rate of the ad placement combination (boot-up ad - pre-roll ad) is 5%; in the (office - weekday daytime - computer work) scenario, the contextual conversion rate of the ad placement combination (boot-up ad - pre-roll ad) is only 1%; based on the conversion rate of the ad placement sequence in different scenarios, the contextual adjustment weight is obtained by comparing it with the global average conversion rate.

[0118] The time between the ad placement and the conversion is normalized based on the ad playback data, and the ad placement is assigned a corresponding attribution weight. Combined with the attribution model based on Markov chain, the baseline contribution coefficient of different ad placements in the user conversion path is analyzed.

[0119] The baseline contribution coefficient is fused with the context adjustment weight to calculate the path contribution coefficient;

[0120] In the case of multiple ad placement combinations, ad placement combinations are recommended for different cross-screen scenarios based on the user cross-screen purchase conversion rate of multiple ads and the path contribution coefficient of different ad placements in the user conversion path.

[0121] Recommendation strategies can directly guide advertising delivery systems. When a user is identified as being in a specific context, the system will prioritize using the ad placement sequence with the highest dynamic contribution coefficient in that context to reach the user, thereby maximizing the advertising effect.

[0122] This invention calculates the cross-screen purchase conversion rate of users under different ad placement types as a core indicator for evaluating advertising effectiveness. By quantifying the contribution of different ad placements to promoting cross-screen purchases, it helps advertisers identify efficient advertising channels and formats, optimize advertising budget allocation, and directly reflect the commercial value of advertising.

[0123] Example 2:

[0124] This invention proposes a cross-screen interactive advertising effectiveness evaluation system based on multimodal intelligent agents, comprising:

[0125] The ad delivery module collects ad playback data on target devices; the ad playback data includes ad placement type, ad delivery time, and user feedback data.

[0126] The cross-screen association recognition module obtains the login account and geographical location of the target device, performs account verification and distance verification during the advertising period, and identifies the cross-screen associated devices of the target device.

[0127] The cross-screen behavior recognition module monitors the operating status of cross-screen associated devices to obtain user behavior data; and constructs a user cross-screen behavior chain based on the time sequence of the user behavior data.

[0128] The cross-screen attribution analysis module analyzes the advertising playback data of the target device and the user cross-screen behavior links of cross-screen related devices to obtain cross-screen related behavior data.

[0129] The advertising quantitative evaluation module analyzes user conversion for different ad placement types based on cross-screen related behavioral data and calculates cross-screen interaction metrics for ads.

[0130] Take smart TV OTT large-screen advertising as an example;

[0131] The system obtains the login account of the smart TV, such as an email address. During the advertising period, the system identifies that the user logged into the same email address or a phone number / third-party account linked through an ID mapping algorithm on a smartphone and a personal computer within the same time period. The ID mapping algorithm is implemented by establishing a unique user ID association table between different types of accounts. This association table can be pre-built and dynamically updated based on the same personal information filled in by the user when registering on different platforms or the association information obtained when authorizing login through third parties. These devices are initially screened as devices with the same account.

[0132] The system continuously acquires the geographic location information of smart TVs, smartphones, and personal computers; it constructs time-series data on straight-line distances by calculating the straight-line distances between devices in real time. Assuming that during and for a period after the advertisement playback, the straight-line distance between the smart TV and smartphone remains within 10 meters, with a distance change rate of less than 1 meter / second, and both devices spend more than 15 minutes together within a preset "home geofence" (determined by GPS coordinates, Wi-Fi fingerprints, etc.); while the personal computer's geographic location is at another location, more than 500 meters away from the smart TV, and there is no shared dwell time pattern; then the smartphone that meets the distance verification criteria will be designated as the cross-screen associated device.

[0133] The system monitors the operational status of identified cross-screen connected devices (smartphones) and collects user operation data on smartphones as user behavior data.

[0134] Within the "golden conversion period" after ad playback (e.g., within 72 hours after ad playback), the system collects user behavior data on smartphones, including search, browsing, purchasing, and sharing behaviors. This collected user behavior data is arranged chronologically to construct a timeline of user behavior on smartphones, forming a complete cross-screen user behavior chain. During this construction process, the system performs data cleaning, such as identifying and removing duplicate, abnormal, or invalid behavior data. For missing data in short time intervals, linear interpolation is used to fill in the gaps; for missing data in long time intervals, no data filling is performed, but the behavior chain is marked as interrupted, ensuring the completeness and accuracy of the behavior chain.

[0135] The system matches the advertising playback data of smart TVs with the cross-screen user behavior of smartphones within a specific time window.

[0136] The time period for determining the associated behaviors after ad playback is a specified period from the start to the end of ad playback. This period is the golden conversion cycle derived from the analysis of a large amount of user behavior data.

[0137] Extract all user feedback data and user behavior data from smartphones within that time period.

[0138] Furthermore, a correlation analysis is conducted based on semantic analysis of advertising content, semantic analysis of user feedback data, and semantic analysis of user behavior, specifically including:

[0139] Semantic analysis of advertising content: Natural language processing and computer vision analysis are performed on the advertising content in the advertising playback data to extract the keywords, themes, sentiments and visual element features of the advertisement and construct the semantic vector of the advertising content;

[0140] Semantic analysis of user feedback data: Natural language processing is performed on the user feedback text in the user feedback data to extract user comments on advertisements and construct user feedback semantic vectors;

[0141] User behavior semantic analysis: Perform semantic analysis on user cross-screen behavior, including search behavior, browsing behavior, purchase behavior and sharing behavior, including search keywords, browsed product topics, purchased product categories and shared content, and construct user behavior semantic vectors;

[0142] The similarity between the semantic vectors of ad content, user feedback, and user behavior is calculated to obtain the correlation between ad content and user behavior.

[0143] Based on the constructed cross-screen related behavioral data, the system analyzes user conversion for different ad placement types and calculates cross-screen ad interaction metrics.

[0144] The cross-screen advertising interaction metrics include user cross-screen purchase conversion rate and path contribution coefficient; the path contribution coefficient is obtained through a time decay attribution model and multiple advertising conversion attribution analyses, including:

[0145] Obtain the timeline of user behavior across screens, combine it with geolocation and device type to identify and construct cross-screen context tags; calculate the conversion rate of ad placement sequences under different cross-screen contexts, compare it with the global average conversion rate, and obtain context adjustment weights;

[0146] The time between the ad placement and the conversion is normalized based on the ad playback data, and the ad placement is assigned a corresponding attribution weight. Combined with the attribution model based on Markov chain, the baseline contribution coefficient of different ad placements in the user conversion path is analyzed.

[0147] The path contribution coefficient is calculated by fusing the baseline contribution coefficient with the context adjustment weight.

[0148] In the case of multiple ad placement combinations, ad placement combinations are recommended for different cross-screen scenarios based on the user cross-screen purchase conversion rate of multiple ads and the path contribution coefficient of different ad placements in the user conversion path.

[0149] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cross-screen interactive advertising effectiveness evaluation system based on multimodal intelligent agents, characterized in that, include: The ad delivery module collects ad playback data on target devices, including ad placement type, ad delivery time, and user feedback data. The cross-screen association recognition module obtains the login account and geographical location of the target device, performs account verification and distance verification during the advertising period, and identifies the cross-screen associated devices of the target device. The cross-screen behavior recognition module monitors the operating status of cross-screen associated devices to obtain user behavior data; and constructs a user cross-screen behavior chain based on the time sequence of the user behavior data. The cross-screen attribution analysis module analyzes the cross-screen behavior data based on the advertising playback data of the target device and the cross-screen behavior links of users on cross-screen related devices within the time period of the associated behavior. The time period for the associated behavior is dynamically adjusted based on the advertising type and product characteristics; The process of acquiring the cross-screen related behavior data includes: The time window is matched between the ad playback data of the target device and the cross-screen behavior links of users on cross-screen related devices to determine the time period of related behaviors after the ad playback. Extract user feedback data and user behavior data from cross-screen related devices within the time period of the associated behavior; Based on semantic analysis of ad content, semantic analysis of user feedback data, and semantic analysis of user behavior, the correlation between ad content and user behavior is calculated; user behaviors with a correlation greater than a preset threshold are filtered to construct cross-screen related behavior data. The dynamic adjustment process of the associated behavior time period includes: Determine the industry category and average order value based on product characteristics; for different ad placement types, industry categories and average order values, collect and maintain historical conversion time data of product ads, and use Bayesian optimization algorithms to dynamically search and determine the optimal time window boundary for related behaviors with user conversion rate as the optimization goal; Semantic analysis of advertising content: Natural language processing and computer vision analysis are performed on the advertising content in the advertising playback data to extract the keywords, themes, sentiments and visual element features of the advertisement and construct the semantic vector of the advertising content; Semantic analysis of user feedback data: Natural language processing is performed on the user feedback text in the user feedback data to extract user comments on advertisements and construct user feedback semantic vectors; User behavior semantic analysis: Perform semantic analysis on user cross-screen behavior, including search behavior, browsing behavior, purchase behavior and sharing behavior, including search keywords, browsed product topics, purchased product categories and shared content, and construct user behavior semantic vectors; Calculate the similarity between the semantic vector of ad content, the semantic vector of user feedback, and the semantic vector of user behavior to obtain the correlation between ad content and user behavior; The advertising quantitative evaluation module analyzes user conversion for different ad placement types based on a time decay attribution model of cross-screen related behavior data, obtains cross-screen interaction metrics, and recommends ad placement combinations for multiple ads.

2. The cross-screen interactive advertising effect evaluation system based on multimodal intelligent agent driving according to claim 1, characterized in that: The ad slot type refers to the playback format of the ad on the target device, including device startup ad, homepage recommendation ad, pre-video ad, mid-video ad, post-video ad, video pause ad, and device screensaver ad. The advertising placement time refers to the time period during which the advertising is placed, including the start and end times of the placement. The user feedback data refers to the viewing feedback from users on the target device, including ad viewing duration, ad skipping actions, ad clicking actions, and user feedback text.

3. The cross-screen interactive advertising effect evaluation system based on multimodal intelligent agent driving according to claim 1, characterized in that: The account verification process involves identifying the user's login account on the target device during the advertising period; and filtering based on the login account on the target device to obtain devices with the same account on the target device during the advertising period. The distance verification involves obtaining the geographical locations of the target device and devices with the same account, calculating the straight-line distance between the devices, and performing interval distance verification based on the time data of the straight-line distance to determine the cross-screen associated devices.

4. The cross-screen interactive advertising effect evaluation system based on multimodal intelligent agent driving according to claim 3, characterized in that: The process of performing interval distance verification based on time series data of linear distance includes: Continuously acquire the geographical location information of the target device and devices with the same account, and calculate the straight-line distance between them in real time to construct time series data of the straight-line distance; based on the preset distance threshold, distance change rate and geographical location co-staying pattern, analyze the time series data to determine the devices with the same account that meet the verification criteria as cross-screen associated devices.

5. The cross-screen interactive advertising effect evaluation system based on multimodal intelligent agent driving according to claim 1, characterized in that: The process of obtaining the user cross-screen behavior chain includes: Obtain ad playback data from users on the target device; User operation data on cross-screen connected devices is collected as user behavior data; the user behavior data includes search behavior, browsing behavior, purchase behavior and sharing behavior on cross-screen connected devices; The collected target device ad playback data and user behavior data from multiple cross-screen related devices are arranged in chronological order to construct a timeline of user behavior on cross-screen related devices, thus obtaining the user's cross-screen behavior chain.

6. The cross-screen interactive advertising effect evaluation system based on multimodal intelligent agent driving according to claim 1, characterized in that: The cross-screen advertising interaction metrics include user cross-screen purchase conversion rate and path contribution coefficient; The path contribution coefficient is obtained through a time decay attribution model and multiple advertising conversion attribution analysis, including: Obtain the timeline of user behavior across screens, combine it with geolocation and device type to identify and construct cross-screen context tags; calculate the conversion rate of ad placement sequences under different cross-screen contexts, compare it with the global average conversion rate, and obtain context adjustment weights; The time between the ad placement and the conversion is normalized based on the ad playback data, and the ad placement is assigned a corresponding attribution weight. Combined with the attribution model based on Markov chain, the baseline contribution coefficient of different ad placements in the user conversion path is analyzed. The baseline contribution coefficient is fused with the context adjustment weight to calculate the path contribution coefficient; In the case of multiple ad placement combinations, ad placement combinations are recommended for different cross-screen scenarios based on the user cross-screen purchase conversion rate of multiple ads and the path contribution coefficient of different ad placements in the user conversion path.

Images