Artificial intelligence advertisement generation method and device based on user portrait, electronic equipment and storage medium

Abstract

The embodiment of the invention discloses an artificial intelligence advertisement generation method and device based on a user portrait, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence advertisement generation, the method comprises the steps that user behavior information and equipment state information are acquired, and the user behavior information comprises historical behavior data and real-time behavior data; determining a device type and a network state according to the device state information; determining attention state information of the user according to the user behavior information; generating a multi-dimensional user portrait according to the equipment type, the network state and the attention state information; determining an advertisement length type according to the user portrait, wherein the advertisement length type comprises an ultra-short type, a relatively short type and a complete type; and generating an advertisement according to the advertisement length type and the user portrait by using the advertisement editing model. The problem that a traditional advertisement generation method is insufficient in flexibility and low in accuracy is solved.

Description

Technical Field

[0001] The present application relates to the technical field of artificial intelligence advertising generation, and in particular to a method and device for generating artificial intelligence advertising based on user portraits. Background Art

[0002] With the widespread adoption of the internet and mobile devices, digital advertising has become a crucial tool for brand promotion and user engagement. Traditional ad generation methods often rely on static rules or simple user classification, making them difficult to adapt to the dynamic changes in user behavior and personalized needs. For example, existing technologies often use fixed ad length and content, unable to flexibly adjust to user attention, network environment, or device type, resulting in poor user experience and ineffective advertising.

[0003] Recent advances in artificial intelligence (AI) technology have brought new opportunities to ad generation. By analyzing user behavior data and device status, user profiles can be constructed and ad delivery optimized. However, existing methods still suffer from the following shortcomings: First, they lack dynamic assessment of user attention, resulting in a mismatch between ad cadence and user needs; second, ad length selection is too limited, failing to adjust based on user tolerance and real-time feedback; and third, ad creative editing lacks intelligence, making it difficult to accurately adapt to different scenarios. Therefore, a comprehensive, intelligent ad generation method is urgently needed to improve ad accuracy and user acceptance. Summary of the Invention

[0004] This application provides an artificial intelligence advertising generation method based on user portraits, which realizes the accurate generation of advertisements.

[0005] This application provides the following solutions: According to the first aspect, an artificial intelligence advertising generation method based on user portrait is provided, characterized in that the method includes: obtaining user behavior information and device status information, the user behavior information including historical behavior data and real-time behavior data; determining the device type and network status based on the device status information; determining the user's attention status information based on the user behavior information; generating a multi-dimensional user portrait based on the device type, network status and attention status information; determining the advertisement length type based on the user portrait, the advertisement length types including: ultra-short type, shorter type and full type; using an advertisement editing model to generate an advertisement based on the advertisement length type and the user portrait.

[0006] According to an achievable method in an embodiment of the present application, the real-time behavior data includes: the current page dwell time and the interaction frequency; the determination of the user's attention state information based on the user behavior information includes: judging the attention state as high attention, low attention or distraction state based on the current page dwell time and the interaction frequency; and combining the historical behavior data, correcting the attention state determination result by comparing the current behavior with the user's historical behavior pattern.

[0007] According to an achievable method in an embodiment of the present application, determining the advertisement length type based on the user portrait includes: determining the advertisement length type based on the user portrait using an advertisement length determination model obtained through pre-training; wherein, the pre-training includes: obtaining a training data set, the training data set including: multi-dimensional user portrait samples and their corresponding advertisement length type labels and advertisement viewing behavior data; training the advertisement length determination model using a supervised learning method, with the user portrait as input; the advertisement length determination model outputs a probability distribution for each length type, and determines the advertisement length type based on a preset threshold or maximum probability; during the training process, adding time dimension features, the time dimension features including: current time period and usage scenario information; the training objectives include: minimizing the difference between the advertisement length type and the advertisement length type label.

[0008] According to an achievable method in an embodiment of the present application, determining the advertisement length type based on the user portrait includes: for users whose attention status information is low attention state or low-speed network user, determining the advertisement length type as ultra-short; for users whose attention status information is medium attention state or medium-speed network user, determining the advertisement length type as shorter; for users whose attention status information is high attention state or high-speed network user, determining the advertisement length type as full; determining the user's historical advertisement tolerance, real-time feedback and / or the content currently browsed by the user based on the user behavior data; and dynamically adjusting the advertisement length type based on the user's historical advertisement tolerance, real-time feedback and / or the content currently browsed by the user.

[0009] According to an achievable method in an embodiment of the present application, dynamically adjusting the advertisement length type according to the user's historical advertisement tolerance and the real-time feedback includes: collecting the user's past advertisement viewing records, the advertisement viewing records including: the complete viewing time, skip time point and / or skip rate of each advertisement; calculating the user's advertisement tolerance threshold, wherein the tolerance threshold is determined by counting the user's average viewing time or the critical time for skipping advertisements; according to the tolerance threshold, classifying the user into low tolerance type, medium tolerance type, and high tolerance type; matching the corresponding advertisement length according to the user classification to obtain a first length value; during the advertisement playback process, monitoring the user's interactive behavior in real time to obtain real-time feedback, the interactive behavior including: whether to skip advertisements, perform mute operations, fast forward behaviors and / or switch applications; if the real-time feedback shows that the user's skip rate has increased or the tolerance has decreased, shortening the first length value; if the real-time feedback shows that the user has watched the advertisement completely and the frequency of interactive behaviors is high, extending the first length value; determining the advertisement length type according to the first length value.

[0010] According to an achievable method in an embodiment of the present application, the historical behavior data includes the user's historical preferred video clips, image styles and copy content; the use of the advertising editing model to generate advertisements according to the advertisement length type and the user portrait includes: based on the historical behavior data, screening materials that match the user portrait from the advertising material library; using computer vision technology to extract key pictures in the materials, and adjusting the editing rhythm and picture effects in combination with the attention state, the editing rhythm includes: picture switching frequency and information density; using natural language processing technology to optimize advertising copy and dubbing; generating personalized advertising content that meets user preferences through generative adversarial networks or text-to-video technology.

[0011] According to an achievable method in an embodiment of the present application, the use of an advertisement editing model to generate advertisements according to the advertisement length type and the user portrait includes: obtaining weather condition information, classifying weather conditions according to the weather condition information, and obtaining weather types, wherein the weather types include: suitable for going out and not suitable for going out; the advertisement editing model modifies the advertisement length type according to the weather type, and generates advertisements according to the advertisement length type and the user portrait.

[0012] According to the second aspect, an artificial intelligence advertising generation device based on user portrait is provided, characterized in that the device includes: an information acquisition unit, configured to acquire user behavior information and device status information, the user behavior information including historical behavior data and real-time behavior data; an information analysis unit, configured to determine the device type and network status based on the device status information; an attention status determination unit, configured to determine the user's attention status information based on the user behavior information; a user portrait generation unit, configured to generate a multi-dimensional user portrait based on the device type, network status and attention status information; a length type determination unit, configured to determine the advertisement length type based on the user portrait, the advertisement length types including: ultra-short type, shorter type and full type; an advertisement generation unit, configured to generate an advertisement based on the advertisement length type and the user portrait using an advertisement clipping model.

[0013] According to a third aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the steps of any one of the methods according to the first aspect are implemented.

[0014] According to a fourth aspect, an electronic device is provided, comprising: one or more processors; and a memory associated with the one or more processors, the memory being used to store program instructions, which, when read and executed by the one or more processors, execute the steps of the method described in any one of the above-mentioned first aspects.

[0015] According to the specific embodiments provided in this application, this application discloses the following technical effects: This application provides an artificial intelligence advertising generation method based on user behavior. By obtaining user behavior information and device status information, combining attention status, device type and network status to generate a multi-dimensional user portrait, and using an advertising editing model to dynamically generate personalized advertisements, it solves the problems of insufficient flexibility and low accuracy in traditional advertising generation methods.

[0016] Of course, any product implementing the present application does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 is a system architecture diagram applicable to the embodiments of the present application; Figure 2 This is a flowchart of the method for generating artificial intelligence advertisements based on user portraits provided in an embodiment of the present application; Figure 3 A structural block diagram of the user portrait-based artificial intelligence advertising generation device provided in an embodiment of the present application; Figure 4 A schematic block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of this application.

[0020] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "an", "the" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0021] It should be understood that the term "and / or" as used herein is merely a description of the relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0022] The word "if," as used herein, may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0023] Existing ad generation methods often rely on static rules or simple user classification, making them difficult to adapt to the dynamic changes in user behavior and personalized needs. In light of this, this application provides a new approach. To facilitate understanding of this application, we first describe the system architecture on which this application is based. Figure 1 An exemplary system architecture to which the embodiments of the present application can be applied is shown. Figure 1 As shown in , the system architecture may include: a user device and an artificial intelligence advertising generation device based on user portraits located on the server side.

[0024] User devices include, but are not limited to, smart mobile terminals, smart home devices, wearable devices, and personal computers (PCs). Smart mobile devices include mobile phones, tablets, laptops, PDAs (Personal Digital Assistants), and connected cars. Smart home devices include smart TVs and smart refrigerators. Wearable devices include smart watches, smart glasses, virtual reality devices, augmented reality devices, and mixed reality devices (i.e., devices that support both virtual reality and augmented reality).

[0025] The AI-powered ad generation device based on user portraits can be set up as an independent server, in a server group, or in a cloud server. A cloud server, also known as a cloud computing server or cloud host, is a host product in the cloud computing service system that solves the problems of difficult management and weak service scalability in traditional physical hosts and virtual private servers (VPS). Figure 1 In addition to the shown architecture, the user portrait-based artificial intelligence advertising generation device can also be set on a computer terminal with strong computing power.

[0026] As one possible implementation method, the artificial intelligence advertising generation device generates advertisements based on user behavior information and device status information, returns the advertisements to the user device through the network, and the user device displays the advertisements to the user. It should be understood that Figure 1 The user devices and artificial intelligence advertisement generating devices are merely illustrative. Any number of user devices and artificial intelligence advertisement generating devices may be provided as required.

[0027] Figure 2 This is a flowchart of the method for generating artificial intelligence advertisements based on user portraits provided in the embodiment of the present application. The method can be Figure 1 The artificial intelligence advertisement generating device based on user portrait in the system shown is executed. Figure 2 As shown in , the method may include the following steps: Step 201: Obtain user behavior information and device status information. The user behavior information includes historical behavior data and real-time behavior data.

[0028] Step 202: Determine the device type and network status based on the device status information.

[0029] Step 204: Generate a multi-dimensional user portrait based on device type, network status, and attention status information.

[0030] Step 205: Determine the advertisement length type according to the user portrait. The advertisement length types include: ultra-short, shorter, and full.

[0031] Step 206: Generate an advertisement based on the advertisement length type and user profile using the advertisement editing model.

[0032] It can be seen from the above process that this application provides an artificial intelligence advertising generation method based on user behavior. By obtaining user behavior information and device status information, combining attention status, device type and network status to generate a multi-dimensional user portrait, and using the advertising editing model to dynamically generate personalized advertisements, thereby solving the problems of insufficient flexibility and low accuracy in traditional advertising generation methods.

[0033] First, the above step 201, namely "obtaining user behavior information and device status information, where the user behavior information includes historical behavior data and real-time behavior data", is described in detail in conjunction with the embodiment.

[0034] This application aims to provide foundational support for subsequent user profiling and ad generation through multi-dimensional data collection. Specifically, "user behavior information" refers to the various types of interaction data generated by users when using devices or applications. It is divided into two subsets: "historical behavior data" and "real-time behavior data." "Historical behavior data" reflects user behavior patterns over a period of time, such as the total duration of ad viewing over the past 30 days (e.g., an average of 15 seconds per view), click-through rate (e.g., 5 ad clicks per day), ad skip frequency (e.g., 70% of ads skipped within 5 seconds), and preferred content types (e.g., a preference for sports videos). This data is typically accumulated over time through client-side logs or cloud-based databases and used to analyze users' long-term habits and tolerance. "Real-time behavior data" captures users' current, immediate behavior, such as time spent on the current page (e.g., 3 seconds on a news page), scrolling speed (e.g., 2 cm per second), interaction frequency (e.g., 2 clicks per minute), and usage scenarios (e.g., morning commuting, inferred by time and location). Meanwhile, "device status information" includes device hardware and network conditions, such as device model, operating system version, network type (Wi-Fi or 4G), and bandwidth speed (20Mbps).

[0035] By obtaining this information, the system can fully understand the user's historical preferences, current status and device environment, providing accurate basis for subsequent attention status judgment and advertising editing.

[0036] The above step 202, namely "determining the device type and network status according to the device status information," is described in detail below in conjunction with an embodiment.

[0037] "Device status information" refers to technical parameters and operational data collected from a user's device. "Device type" and "network status" are two core dimensions extracted from this information. Determining "device type" involves the device's hardware characteristics, such as model, screen size, and operating system version. This information is typically obtained through device APIs or client-side probes and is used to assess device performance and usage scenarios. For example, a high-end smartphone may support HD video playback, while a low-end device may be more suitable for lightweight content. "Network status" reflects the device's current connection quality, including network type (such as Wi-Fi, 4G, or 5G), real-time bandwidth (such as 20Mbps or 1Mbps), latency (such as 50ms), and packet loss rate (such as 2%). This data can be collected in real time through network monitoring tools to assess ad loading feasibility and user experience. For example, suppose user B uses a mid-range Android phone. The device status information displays its model, screen size, and operating system. Network monitoring indicates that it is using a 4G network with a bandwidth of 5Mbps and a latency of 80ms. Based on this data, the "device type" can be determined to be a "mid-range smartphone" and the "network status" to be a "medium-speed network."

[0038] The above step 203, namely "determining the user's attention state information based on the user's behavior information", is described in detail below in conjunction with an embodiment.

[0039] This method analyzes user behavior data to infer their current level of focus, thereby providing a basis for the length and editing rhythm of advertisements. The attention state represents the user's level of focus and can be categorized as "high attention," "low attention," or "distracted." It can also be given as a score.

[0040] Attention state is determined based on user behavior information. This can be based solely on real-time behavioral data, or by combining historical and real-time data for dual analysis, potentially aided by intelligent algorithm optimization. For example, the system may use a pre-trained machine learning model that takes behavioral data as input and outputs an attention state classification. Training data comes from labeled samples (such as user feedback or experimental observations). The model then refines its initial judgment by comparing current behavior with historical patterns. For example, user E's historical data indicates a high tendency to pay attention (often watching full ads), but real-time data indicates a current dwell time of only 3 seconds and frequent scrolling. After comprehensive analysis, the model may adjust the state to "distracted." Furthermore, if the device supports sensor data (such as camera detection of eye movements or heart rate changes), the results can be further verified. For example, focused eye movements indicate concentration, while scattered eye movements indicate distraction.

[0041] For example, suppose user A watches videos on a tablet at home. Historical data shows that they watch ads for an average of 20 seconds, with a low skip rate. Real-time data shows a 15-second dwell time on the current page, four interactions per minute, and a usage scenario of "evening leisure." Real-time data is used to preliminarily determine a user's "high attention state," which is then verified using historical tolerance data, ultimately outputting "high attention." Conversely, if user B uses their phone at work, their real-time dwell time is only two seconds, they frequently switch apps, and their historical skip rate is as high as 80%, the system determines the user is in a "distracted state." This approach, through multi-dimensional data analysis, ensures the accuracy and practicality of attention state determination, laying the foundation for subsequent ad generation.

[0042] As an implementable method, real-time behavior data includes the current page dwell time and interaction frequency; determining the user's attention state information based on the user behavior information includes: judging the attention state as high attention, low attention or distraction based on the current page dwell time and interaction frequency; optimizing the attention state judgment result in combination with historical behavior data, and correcting the attention state by comparing the current behavior with the user's historical behavior pattern.

[0043] Specifically, this embodiment determines the attention state based on the current page dwell time and interaction frequency. The "current page dwell time" refers to the length of time a user stays on a certain page or application interface, usually measured in seconds. For example, a user stays on a news page for 5 seconds or on a short video page for 2 seconds. The "interaction frequency" reflects the number of times a user interacts with the interface per unit time, such as 3 clicks or 2 swipes per minute. These indicators are directly related to the user's concentration level.

[0044] Based on the above real-time data, "judging the attention state as high attention, low attention or distracted state based on the current page dwell time and the interaction frequency" provides a preliminary judgment rule. Specifically, long dwell time and high interaction frequency usually correspond to a "high attention state", indicating that the user is willing to invest time and energy; short dwell time and low interaction frequency point to a "low attention state", indicating that the user's interest is limited; if the interaction frequency is irregular (such as fluctuating) or accompanied by rapid switching behavior, it may be a "distracted state", which means that the user's attention is scattered. Taking user C as an example, his current page dwell time is 15 seconds, and the interaction frequency is 4 times per minute. The system initially determines that it is a "high attention state"; user D's dwell time is only 2 seconds, there is no interaction, and it is determined to be a "low attention state"; user E's dwell time is 3 seconds, but the interaction frequency fluctuates (quickly sliding after clicking), and it is determined to be a "distracted state".

[0045] However, relying solely on real-time data may be inaccurate due to the variability of user behavior, so "combining historical behavior data to optimize the attention state judgment results, and correcting the attention state by comparing the current behavior with the user's historical behavior pattern" introduces a historical dimension to improve reliability. Historical behavior data can include the user's past ad viewing time, skip rate, and interaction habits. For example, user F watched ads for an average of 20 seconds in the past 30 days, with a skip rate of only 20%, indicating that his tolerance and attention tendency are relatively high. The correction process is adjusted by comparing the current behavior with historical patterns. For example, user F's current stay time is only 3 seconds, and the interaction frequency is low. It is initially judged to be a "low attention state", but its historical data shows a high attention tendency. The system may consider the current behavior to be a temporary anomaly and correct it to a "medium attention state". This method uses long-term patterns to correct instantaneous fluctuations and avoid misjudgment.

[0046] This step can be implemented in a variety of ways. For example, based on predefined rules, current behavior data can be compared with statistical indicators from historical behavior data to adjust the attention state. Specifically, key statistical indicators of historical behavior, such as average page dwell time, average interaction frequency, ad completion rate, and skip rate, are calculated. Current behavior data is compared with historical indicators and thresholds are set. The attention state is then adjusted based on the deviation. For example, if the current dwell time is significantly lower than the historical average, but the historical data shows a high attention tendency, "low attention" may be adjusted to "medium attention." Alternatively, a supervised learning model can be trained that takes historical and current behavior data as input and outputs the corrected attention state. Specifically, a labeled dataset is collected, containing historical behavior samples (e.g., dwell time and interaction frequency from the past 30 days), current behavior samples, and their corresponding attention state labels (e.g., from user feedback or experimental annotations). A classification model, such as a support vector machine, random forest, or deep neural network, is used as input, including historical and current dwell time, interaction frequency, and other features, and the output is a "high / low / distracted" state. The model is trained and the cross-entropy loss function is optimized to enable it to learn the comparative relationship between historical and current behavior.

[0047] Furthermore, this application can also determine the user's attention status information based on the user's page jump time. As part of user behavior information, jump time can indirectly reflect the user's attention level. Generally speaking, a short jump time may indicate that the user is quickly browsing content, has low attention, or is distracted; while a long jump time may mean that the user stays on a page and reads or operates in depth, with high attention. The specific judgment requires combining the statistical distribution of jump time and context.

[0048] In practice, attention state can be determined based on jump time through the following steps: First, collect the user's real-time jump time series, for example, by using front-end tracking to record the timestamp of each page switch in the app or website, and calculate the interval between adjacent jumps. Second, establish rules for determining attention state. For example, if the average jump time is less than a certain threshold (such as 5 seconds), it can be preliminarily determined as a "low attention state," as rapid jumps suggest the user is not deeply engaged with the content. If the jump time is in the medium range (such as 5-15 seconds), it may indicate a "medium attention state." If the jump time is longer (such as over 15 seconds), it tends to indicate a "high attention state." Furthermore, the volatility of jump time should be considered: irregular jump intervals (such as 2 seconds, 10 seconds, or 3 seconds) may indicate a "distracted state," indicating that the user is not paying attention. These rules can adjust the thresholds based on actual application scenarios.

[0049] The judgment results can be further optimized by combining historical behavior data. Historical behavior data may contain the user's past jump time patterns. For example, a user's average jump time is 20 seconds, and the jumps are relatively stable (low standard deviation), indicating that they are accustomed to browsing pages for a longer time. If the current jump time deviates significantly from the historical pattern (such as suddenly dropping to 3 seconds), it may be caused by temporary distraction or environmental changes. The system can correct the initial judgment through comparison. For example, user A's historical data shows an average jump time of 25 seconds, and the current jump time is 4 seconds. The initial judgment is "low attention state", but considering his historical tendencies, the system may correct it to "medium attention state". This comparison method can be implemented through statistical analysis (such as calculating the deviation percentage) or machine learning models (such as classifiers).

[0050] The above step 204, i.e., "generating a multi-dimensional user portrait based on the device type, network status, and attention status information," is described in detail below in conjunction with an embodiment.

[0051] This invention builds a comprehensive user profile model by integrating device-related hardware characteristics, network environment, and user behavior. This multi-dimensional user profile, fused with multi-source data, reflects the user's overall usage habits, environmental context, and attentional tendencies, providing an accurate basis for subsequent ad length selection and content generation. The following will explain this feature from a technical implementation perspective and illustrate it with specific examples.

[0052] The process of generating a multi-dimensional user profile involves integrating the three dimensions mentioned above (device type, network status, and attention state) into a comprehensive feature vector or set of labels. This can be achieved through data concatenation or feature fusion techniques. For example, the system quantifies device type, network status, and attention state into feature values ​​(e.g., device type: high-end = 1, mid-range = 0.5, low-end = 0; network status: high speed = 1, mid-range = 0.5, low speed = 0; attention state: high = 1, low = 0.5, distracted = 0). These features are then concatenated into a three-dimensional vector (e.g., [1, 1, 1] represents "high-end device, high-speed network, high attention"). Alternatively, machine learning methods (such as clustering or embedding models) can be used to map these dimensions into a higher-level profile label. For example, K-means clustering can categorize users into groups such as "dedicated high-end users" and "fast browsing mid-range users."

[0053] Furthermore, the multi-dimensional user portrait in this application may also include other dimensional information, such as user preference information. The user preference dimension reflects the user's interest in content type, style or theme, such as preference for entertainment videos, technology news or shopping promotions. Preference features can be extracted by analyzing the user's historical behavior data (such as browsing history, search keywords, and clicked advertising types).

[0054] Multi-dimensional user profiles can also include time and scene information. The time and scene dimensions capture the specific time and environmental context of user behavior, such as morning commuting, lunch break, or evening leisure. This dimension can reflect the differences in user habits and needs in different time periods. Time and scene are inferred through the device's clock and positioning data. Specific implementations include: (1) recording the timestamp of the behavior (such as 8:00 am or 8:00 pm); (2) inferring the scene by combining geographic location information (such as subway station or home through GPS positioning); (3) performing pattern analysis on time and scene data. For example, user C uses his phone on the subway from 8:00 to 9:00 every day, which is determined to be a "commuting scene"; (4) quantifying the results into features (such as "morning commuting = 1, evening leisure = 0") and adding them to the profile. For example, the profile of user C can be expanded to "mid-end device, medium-speed network, low attention, commuting scene", which is suitable for generating short-term, high-impact advertisements.

[0055] Multi-dimensional user profiles can also include basic attributes such as the user's age, gender, and occupation. These attributes can supplement personal information not directly reflected in behavioral data, enhancing the comprehensiveness of the profile. This can be obtained through user registration information, third-party data, or behavioral inference. Explicit data can be obtained from the user's account (e.g., gender: male, age: 30). If direct data is unavailable, inference can be made based on behavioral patterns. For example, if user D frequently browses maternity and baby products, we can infer her gender and age range. A machine learning model (e.g., logistic regression) can be used to train an inference engine to predict gender and age range based on historical browsing categories (e.g., fashion, games). The results are encoded into profile dimensions. For example, user D's profile might be "high-end devices, high-speed internet, high attention span, female 25-35 years old," making it suitable for advertising maternity and baby products.

[0056] The above step 205, i.e., "determining the advertisement length type according to the user portrait, wherein the advertisement length types include: ultra-short type, shorter type, and full type" is described in detail below in conjunction with an embodiment.

[0057] This step maps the device type, network status, attention status and other dimensions in the user portrait with the ad length type (ultra-short, short, full) to ensure that the ad length can both meet the user's current acceptance capacity and maximize information transmission efficiency and user experience.

[0058] Ad length types are used to represent ads of different lengths. The ad length types of this application include: ultra-short, shorter, and full, with ad lengths increasing in sequence. Among them, ultra-short ads are usually 3-15 seconds long, suitable for quickly delivering core information; shorter ads are 15-30 seconds long, which can balance the amount of information and user tolerance; full ads are more than 30 seconds long, suitable for detailed narratives or brand displays. The process of determining ad length is based on the analysis of user portraits, and by evaluating the user's attention level, network conditions, and device performance, its acceptance of different lengths is judged. For example, users in a low-attention state may find it difficult to accept long ads, while high-speed networks and high-performance devices support longer loading and playback times. This step can be achieved by using rule mapping or machine learning models to convert portrait features into length type decisions.

[0059] In determining the implementation method of ad type, the system sets clear judgment criteria based on various dimensions of the user portrait. For example, if the user portrait shows "low attention state" or "slow network speed", ultra-short ads are selected because short duration can quickly attract attention and adapt to weak network environments; if it is "medium attention state" or "medium network speed", shorter ads are selected to provide an appropriate amount of information; if it is "high attention state" and "high network speed", full-length ads are selected to use the user's focused time to deliver complete content. With device type as a supplementary dimension, high-performance devices (such as high-end smartphones) support the complex content of full-length ads, while low-end devices tend to prefer ultra-short ads to reduce the loading burden.

[0060] As an implementable method, determining the advertisement length type based on the user portrait includes: for users in low attention state or low-speed network, determining the advertisement length type as ultra-short; for users in medium attention state or medium-speed network, determining the advertisement length type as shorter; for users in high attention state or high-speed network, determining the advertisement length type as full; determining the user's historical advertisement tolerance, real-time feedback and / or the content currently browsed by the user based on the user behavior data; dynamically adjusting the advertisement length type based on the user's historical advertisement tolerance, real-time feedback and / or the content currently browsed by the user.

[0061] Specifically, this embodiment uses a dynamic adjustment mechanism to modify the ad length type initially selected based on static user profile characteristics (such as attention state and network status). The user profile provides an initial judgment; for example, users with low attention levels tend to prefer ultra-short ads. However, the user's actual acceptance may vary depending on historical habits, current reactions, or evolving interests. "Historical ad tolerance" reflects the user's long-term acceptance of ad length, such as whether they are accustomed to watching long ads. "Real-time feedback" captures the user's immediate behavior during ad playback, such as whether they skip or interact. "Currently viewed content" reveals the user's immediate interests, which may influence their tolerance for ad length. Using these three factors, the system can dynamically adjust the ad length type, avoiding the impact of rigid initial selections that can lead to a poor user experience or ineffective advertising. For example, a user with high tolerance may accept longer ads even with low attention. A high real-time skip rate may necessitate a shorter ad length.

[0062] The adjustment method based on historical advertising tolerance is as follows: First, collect historical user data, such as the viewing time of each ad in the past 30 days (average 15 seconds), skip time point (usually skipped after 10 seconds), and completion rate (60%); then calculate the tolerance threshold, for example, by counting the average viewing time or the critical skip time to obtain the tolerance classification (low, medium, high); finally, adjust the length type based on the tolerance. For example, user A's profile is "low attention, low-speed network", and the initial choice is ultra-short type (5 seconds), but his historical data shows an average viewing time of 20 seconds, and the tolerance is medium, so the system adjusts to a shorter type (15 seconds). This adjustment is achieved through rule mapping, such as "if the tolerance is higher than the attention state expectation, extend the length type by one level" to ensure that the ad length is more in line with user habits.

[0063] The adjustment method based on real-time feedback is as follows: First, monitor user interaction behavior in real time, such as whether they skip ads (skip at 5 seconds), mute the video, fast-forward, and / or switch apps. Next, analyze feedback trends, such as an increase in skip rate (from 20% to 50%) or an increase in interaction frequency (2 clicks per minute). Finally, adjust the length based on this feedback. For example, if user B's profile is "medium attention span, medium-speed internet," a shorter ad (15 seconds) would be initially selected. However, during playback, the user skips at 5 seconds. Real-time feedback indicates a decrease in tolerance, so the system shortens the ad to an ultra-short ad (10 seconds). This adjustment logic can be implemented using conditional rules (e.g., "skip rate > 50%, shorten by one level") or machine learning models (e.g., reinforcement learning, which optimizes the ad length by rewarding completion). This approach ensures that ad length responds immediately to changes in user status.

[0064] Adjustments based on the user's current browsing content are made as follows: First, the current page content is collected, for example, through NLP analysis of keywords ("tech news," "promotional page") or page type (video, article); second, the content's impact on attention is assessed, for example, video content may increase tolerance, while quickly scrolling through social media may reduce tolerance; finally, the length type is adjusted. For example, user C's profile is "high attention, high-speed network," and the initial selection is a full-length video (30 seconds). However, the system infers that their tolerance is reduced and adjusts to a shorter video (20 seconds). Adjustments can be made through predefined rules (such as "shorten short content pages by one level") or model predictions (input content features, output length offset).

[0065] In practical applications, the three factors of a user's historical advertising tolerance, real-time feedback, and the content they are currently browsing can be used individually or in combination to form a comprehensive adjustment strategy. Specific implementations include: (1) setting priorities, such as real-time feedback taking precedence over tolerance, which takes precedence over browsing content; (2) weighted fusion, such as a tolerance weight of 0.4, feedback weight of 0.4, and content weight of 0.2, to calculate the adjusted length value; and (3) dynamic iteration, recording the effect (such as completion rate) after each adjustment to optimize the next decision. For example, user D's profile is "low attention, low-speed network", with an initial ultra-short type (5 seconds), a medium historical tolerance (15 seconds), no skipping in real-time feedback, and currently browsing scientific articles. The comprehensive adjustment is to a shorter type (15 seconds). This method ensures that the length type is both consistent with the initial profile and adapts to dynamic changes through the coordination of multiple factors.

[0066] As an achievable method, the present application can dynamically adjust the type of advertisement length based on the user's historical advertisement tolerance and real-time feedback. Specifically: collect the user's past advertisement viewing records, the advertisement viewing records include: the complete viewing time of each advertisement, the skipping time point and / or the skipping rate; calculate the user's advertisement tolerance threshold, wherein the tolerance threshold is determined by counting the user's average viewing time or the critical time for skipping advertisements; according to the tolerance threshold, classify the user into low tolerance type, medium tolerance type, and high tolerance type; match the corresponding advertisement length according to the user classification to obtain a first length value; during the advertisement playback process, monitor the user's interactive behavior in real time, the interactive behavior includes: whether to skip advertisements, perform mute operations, fast forward behaviors and / or switch applications; if the real-time feedback shows that the user's skipping rate increases or the tolerance decreases, shorten the first length value; if the real-time feedback shows that the user watches the advertisement completely and the frequency of interactive behavior is high, extend the first length value; determine the advertisement length type according to the first length value.

[0067] In the implementation method of determining the advertisement type, a machine learning model, such as a pre-trained advertisement length determination model, can also be used. As an implementable method, the advertisement length determination model obtained by pre-training can be used to determine the advertisement length type according to the user portrait; wherein, the pre-training includes: obtaining a training data set, the training data set includes: multi-dimensional user portrait samples and their corresponding advertisement length type labels and advertisement viewing behavior data; using a supervised learning method to train the advertisement length determination model, with the user portrait as input and the advertisement length type as output; in the training process, adding time dimension features, the time dimension features include: current time period and usage scenario information; the advertisement length determination model outputs the probability distribution of each length type, and selects the advertisement length type according to a preset threshold or maximum probability; the training objectives include: minimizing the difference between the advertisement length type and the advertisement length type label.

[0068] Specifically, the pre-training process involves acquiring a training dataset containing user profile samples, ad length labels, and viewing behavior data. Using user profiles as input and ad length types as output, the trained model outputs a probability distribution for each length type and selects the final type based on a preset threshold or maximum probability. The training goal is to minimize the discrepancy between the predicted results and the actual labels to ensure model accuracy. In particular, incorporating time-dimensional features (such as the current time period and usage scenario information) enhances the model's contextual awareness, making ad length selection more tailored to the user's actual state.

[0069] This application incorporates dynamic information about time and context to improve the ad length determination model's ability to understand and predict user behavior. The temporal dimension places user behavior in a specific spatiotemporal context, avoiding the problem of static models ignoring changes in user status over time. For example, the same user's attention level and ad receptivity may differ significantly during their morning commute and their evening leisure time. By incorporating these features into the training process, the model can learn the impact of time and context on ad length preferences, thereby generating ads that better meet user needs. Specifically, the "current time period" refers to the specific time range in which user behavior occurs, typically divided into hours or finer-grained units, such as morning (6:00-9:00), noon (11:00-14:00), and evening (18:00-22:00). This feature can be directly obtained from device timestamp data and input into the model as a numeric value (e.g., one-hot encoding: morning = [1,0,0]). The inclusion of time periods is based on the cyclical nature of user behavior patterns. For example, research shows that users tend to scan quickly during their morning commute, which favors ultra-short ads; whereas, users tend to be more focused and receptive to full-length ads during their evening leisure time. During training, the model analyzes the relationship between time periods and ad length labels in the dataset to learn the optimal lengths for different time periods. For example, if the dataset has a high prevalence of very short ad lengths in the morning, the model will tend to predict short durations in the morning.

[0070] "Usage scenario information" further enriches the semantics of the temporal dimension, referring to the specific environment or activity context in which user behavior occurs, such as commuting, relaxing at home, or working. This feature is typically inferred through timestamps combined with geolocation data (GPS) or behavioral patterns. For example, a user rapidly browsing pages near a subway station at 8:00 AM daily can be inferred as a "commuting scenario," while a prolonged stay on a fixed Wi-Fi network at 8:00 PM could indicate "relaxing at home." Scenario information can be represented by categorical labels (e.g., commuting = 1, relaxing = 0) or embedded vectors and incorporated into training data. The correlation between scenario and ad length stems from the fact that users' attention allocation and tolerance vary across different scenarios. For example, commuting may require short, quick ads, while relaxing scenarios are more suitable for longer, narrative-based ads. By learning these patterns, the model becomes more adaptable to changing scenarios.

[0071] Technically, the specific method for incorporating time dimension features is to use "current time period" and "usage scenario information" as additional input features, concatenating them with other dimensions of the user profile (such as device type, network status, and attention state) to form a complete feature vector. For example, the profile vector for user A might be [Device: 1, Network: 1, Attention: 0.5, Time period: Morning = 1, Scenario: Commute = 1]. During training, this extended vector is input into a supervised learning model (such as a deep neural network (DNN). Backpropagation optimizes weights to enable the model to capture the influence of time on ad length selection. If the dataset shows that morning commuters tend to choose ultra-short ads, the model will increase the predicted probability of ultra-short ads under similar conditions. This approach ensures that the model not only focuses on static user attributes but also adapts to dynamic scenarios.

[0072] The above step 206, i.e., "generating advertisements according to advertisement length types and user portraits by using an advertisement editing model," is described in detail below in conjunction with an embodiment.

[0073] An ad editing model is an AI-based system that typically integrates computer vision, natural language processing (NLP), and video editing technologies. It extracts content from a library of assets and edits it to a specified length and user characteristics. Its workflow involves three main steps: First, it selects assets that match user preferences based on user profiles. For example, it chooses detailed product displays relevant to the interests of users in a "high-attention state" or lightweight content for "low-speed network users." Second, it determines the editing framework based on the ad length type. For example, ultra-short ads (5 seconds) require a highly condensed core message, while full-length ads (30 seconds) retain more narrative elements. Finally, an algorithm adjusts the editing rhythm and content layout to generate the final ad. This process ensures that the ad is not only of the appropriate length but also maximizes user engagement and message delivery.

[0074] Technically, the ad editing model first uses computer vision to analyze the video content in the stock footage library. For example, it uses convolutional neural networks or object detection models to identify key scenes within the video, including brand logos, product close-ups, or emotionally charged moments. Each frame is then scored to assess its importance. The model then adjusts the editing rhythm based on the user's attention state as reflected in the user profile: for "high attention state," the frame transitions are slow (e.g., every 5 seconds) to preserve details; for "low attention state," the transitions are accelerated (e.g., every 2 seconds) to highlight key points; and for "distracted state," dynamic effects (e.g., flashing) are added to attract attention. The model then trims the footage based on length. For example, for ultra-short ads, only 3-5 high-scoring frames are retained, while for full-length ads, more frames are spliced ​​together to optimize narrative coherence. Finally, natural language processing (NLP) technology optimizes the ad copy and voiceover to ensure the message is clearly conveyed within the specified duration.

[0075] Generative technologies may also be incorporated into ad editing models to further enhance personalization. For example, using generative adversarial networks (GANs) or text-to-video (TTV) technology, models can generate entirely new content based on user profiles, rather than relying solely on existing footage. For example, if a user's profile indicates a preference for "tech content," GANs can generate footage of a virtual spokesperson explaining tech products, while TTV can convert a text message about "the latest smartphone launch" into a dynamic video. These technologies, combined with traditional editing, can make ad content more tailored to user needs. For example, for users with "high-end devices and high attention spans," the model might generate a full-length 30-second ad featuring GAN-generated tech scenes; for users with "low-end devices and low attention spans," a shorter 5-second ad might be edited, showcasing only a close-up of the product.

[0076] As an implementable method, the historical behavior data in this application includes the user's historical preferred video clips, image styles and copy content; the advertising editing model generates advertisements based on the advertisement length type and the user portrait, including: based on the historical behavior data, screening materials that match the user portrait from the advertising material library; using computer vision technology to extract key pictures in the material, and adjusting the editing rhythm and picture effects in combination with the attention state, the editing rhythm includes: picture switching frequency and information density; using natural language processing technology to optimize advertising copy and dubbing; through generative adversarial networks or text-to-video technology, generating personalized advertising content that meets user preferences.

[0077] Specifically, this application adaptively controls the presentation rhythm and visual effects of advertisements based on the user's attention level, ensuring that the content of the advertisements matches the user's receptive capacity. The attention state directly affects the user's ability to process information: in a high-attention state, the user can accept a slower pace and a higher information density; in a low-attention state, the user requires fast switching and concise information to maintain interest; in a distracted state, attention must be re-attracted through high-frequency switching and strong visual stimulation. "Screen switching frequency" determines the speed of the editing, for example, switching once every 5 seconds is a slow pace, and switching once every 1 second is a fast pace; "information density" controls the complexity of the content, for example, low density only displays the brand logo, and high density includes multiple information such as product close-ups and usage scenarios. By adjusting these two, the advertising editing model can generate an advertisement version that suits the attention state.

[0078] In implementation, the ad editing model first uses attention state information to set an editing strategy. Specific steps include: (1) Analyzing attention state: Obtaining attention state (e.g., high, low, distracted) from user profiles. This state is determined based on historical and real-time behavioral data; (2) Adjusting screen switching frequency: Presetting a rhythm template based on attention state, for example, a slow rhythm (3-5 seconds / time) for high attention, a fast rhythm (1-2 seconds / time) for low attention, and an extremely fast rhythm (0.5-1 second / time) for distracted state; (3) Adjusting information density: Selecting the complexity of screen content based on attention state, for example, retaining multi-layered information for high attention and streamlining to core elements for low attention; (4) Optimizing screen effects: Adding visual effects (e.g., fade in and out, flashing) based on rhythm and density to enhance user experience. Computer vision technology is used to extract key images (e.g., product close-ups, logos), and splicing them together according to the adjusted rhythm to ultimately generate an ad.

[0079] As another possible implementation, this application can also generate advertisements based on weather information. Weather conditions can affect users' living conditions and behavior patterns, thereby indirectly influencing their acceptance of advertisement length. Specifically, weather information is obtained, and weather conditions are classified based on the weather information to obtain weather types, including those suitable for going out and those not suitable for going out. The advertisement editing model modifies the advertisement length type based on the weather type, and an advertisement is generated based on the user profile.

[0080] In implementation, weather conditions are first acquired and classified. The device's location service (e.g., GPS) is used to obtain the user's location's latitude and longitude in real time. Weather data, such as temperature (25°C), rainfall (10mm / h), and wind speed (5m / s), are then obtained through a weather API. Then, predefined rules are used to classify the weather type: If there's no rain, a suitable temperature (e.g., 15-30°C), and low wind speed, the weather is considered "suitable for going out." Rain, extreme temperatures (e.g., <0°C or >35°C), and strong winds are considered "unsuitable for going out." Weather parameter thresholds can be preset (e.g., rainfall >5mm / h is unsuitable for going out) or a classifier trained using machine learning (e.g., decision trees). Second, the length of the ad is modified based on the weather type using rules or machine learning models. For example, "unsuitable for going out" ads tend to be lengthened (e.g., from a shorter ad to a full one) because users are likely to stay indoors while "suitable for going out" ads tend to be shortened (e.g., from a full one to a shorter one) to accommodate users' potential outdoor activities. Finally, the advertising editing model edits the material according to the adjusted length and user portrait to generate an advertisement.

[0081] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0082] According to an embodiment of another aspect, an artificial intelligence advertising generation device based on user portraits is provided. Figure 3 A schematic block diagram of an artificial intelligence advertisement generating device based on user portrait according to an embodiment is shown as follows: Figure 3 As shown, the device 300 includes: The information acquisition unit 301 is configured to acquire user behavior information and device status information, where the user behavior information includes historical behavior data and real-time behavior data.

[0083] The information analysis unit 302 is configured to determine the device type and network status according to the device status information.

[0084] The attention state determination unit 303 is configured to determine the user's attention state information based on the user behavior information.

[0085] The user portrait generation unit 304 is configured to generate a multi-dimensional user portrait based on the device type, network status and attention status information.

[0086] The length type determining unit 305 is configured to determine an advertisement length type according to the user portrait, where the advertisement length types include: ultra-short, shorter, and full.

[0087] The advertisement generating unit 306 is configured to generate an advertisement based on the advertisement length type and the user portrait by using an advertisement editing model.

[0088] As one of the possible implementation methods, the real-time behavior data includes: the current page dwell time and the interaction frequency; the attention state determination unit 303 can be configured to determine the user's attention state information based on the user behavior information: based on the current page dwell time and the interaction frequency, judge the attention state as high attention, low attention or distraction state; in combination with the historical behavior data, correct the attention state determination result by comparing the current behavior with the user's historical behavior pattern.

[0089] As one of the feasible ways, the length type determination unit 305 can be configured to determine the advertisement length type according to the user portrait by: using the advertisement length determination model obtained by pre-training to determine the advertisement length type according to the user portrait; wherein, the pre-training includes: obtaining a training data set, the training data set including: multi-dimensional user portrait samples and their corresponding advertisement length type labels and advertisement viewing behavior data; using a supervised learning method to train the advertisement length determination model, with the user portrait as input; the advertisement length determination model outputs the probability distribution of each length type, and determines the advertisement length type according to a preset threshold or maximum probability; in the training process, adding time dimension features, the time dimension features including: current time period and usage scenario information; the training objectives include: minimizing the difference between the advertisement length type and the advertisement length type label.

[0090] As one of the feasible ways, the length type determination unit 305 can be configured as follows when determining the advertisement length type according to the user portrait: for users whose attention status information is low attention state or low-speed network user, the advertisement length type is determined to be ultra-short; for users whose attention status information is medium attention state or medium-speed network user, the advertisement length type is determined to be shorter; for users whose attention status information is high attention state or high-speed network user, the advertisement length type is determined to be complete; determine the user's historical advertisement tolerance, real-time feedback and / or the content currently browsed by the user according to the user behavior data; and dynamically adjust the advertisement length type according to the user's historical advertisement tolerance, real-time feedback and / or the content currently browsed by the user.

[0091] As one of the possible implementation methods, the length type determination unit 305 can be configured to: collect the user's past advertising viewing records, the advertising viewing records including: the complete viewing time, skip time point and / or skip rate of each advertisement; calculate the user's advertising tolerance threshold, wherein the tolerance threshold is determined by counting the user's average viewing time or the critical time for skipping advertisements; classify the user into low tolerance type, medium tolerance type, and high tolerance type according to the tolerance threshold; match the corresponding advertisement length according to the user classification to obtain a first length value; during the advertisement playback process, monitor the user's interactive behavior in real time to obtain real-time feedback, the interactive behavior including: whether to skip advertisements, perform mute operations, fast forward and / or switch applications; if the real-time feedback shows that the user's skip rate has increased or the tolerance has decreased, shorten the first length value; if the real-time feedback shows that the user has watched the advertisement completely and the frequency of interactive behavior is high, extend the first length value; determine the advertisement length type according to the first length value.

[0092] As one of the possible ways to implement it, historical behavior data includes the user's historical preferred video clips, image styles and copy content; when the advertisement generation unit 306 uses the advertisement editing model to generate advertisements according to the advertisement length type and the user portrait, it can be configured to: based on the historical behavior data, filter the materials that match the user portrait from the advertisement material library; use computer vision technology to extract key pictures in the materials, and adjust the editing rhythm and picture effects in combination with the attention state, the editing rhythm includes: picture switching frequency and information density; use natural language processing technology to optimize advertising copy and dubbing; generate personalized advertising content that meets user preferences through generative adversarial networks or text-to-video technology.

[0093] As one of the feasible ways, when the advertisement generation unit 306 generates advertisements according to the advertisement length type and the user portrait by using the advertisement editing model, it can be configured to: obtain weather condition information, classify the weather conditions according to the weather condition information, and obtain weather types, wherein the weather types include: suitable for going out and not suitable for going out; the advertisement editing model modifies the advertisement length type according to the weather type, and generates advertisements according to the advertisement length type and the user portrait.

[0094] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple. For the relevant parts, refer to the partial description of the method embodiment. The device embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0095] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0096] In addition, an embodiment of the present application further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of any one of the methods in the aforementioned method embodiments are implemented.

[0097] And an electronic device comprising: one or more processors; and A memory associated with the one or more processors, the memory being used to store program instructions, wherein the program instructions, when read and executed by the one or more processors, execute the steps of the method described in any one of the aforementioned method embodiments.

[0098] The present application also provides a computer program product, comprising a computer program, which implements the steps of any one of the methods described in the aforementioned method embodiments when executed by a processor.

[0099] in, Figure 4The electronic device architecture is shown as an example, and may include a processor 410, a video display adapter 411, a disk drive 412, an input / output interface 413, a network interface 414, and a memory 420. The processor 410, the video display adapter 411, the disk drive 412, the input / output interface 413, the network interface 414, and the memory 420 may be communicatively connected via a communication bus 430.

[0100] The processor 410 may be implemented as a general-purpose CPU, a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and may be used to execute relevant programs to implement the technical solutions provided in this application.

[0101] The memory 420 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 420 can store an operating system 421 for controlling the operation of the electronic device 400, and a basic input and output system (BIOS) 422 for controlling the low-level operations of the electronic device 400. In addition, a web browser 423, a data storage management system 424, and an artificial intelligence advertising generation device 425 based on user portraits, etc. can also be stored. The above-mentioned artificial intelligence advertising generation device 425 based on user portraits can be an application program that specifically implements the operations of the aforementioned steps in the embodiment of the present application. In short, when the technical solution provided in this application is implemented by software or firmware, the relevant program code is stored in the memory 420 and is called and executed by the processor 410.

[0102] The input / output interface 413 is used to connect to input / output modules to enable information input and output. The input / output modules can be configured as components within the device (not shown) or externally connected to the device to provide corresponding functions. Input devices may include a keyboard, mouse, touch screen, microphone, and various sensors, while output devices may include a display, speaker, vibrator, indicator light, and the like.

[0103] The network interface 414 is used to connect to a communication module (not shown) to enable communication between the device and other devices. The communication module can communicate via wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.).

[0104] The bus 430 comprises a pathway for transmitting information between the various components of the device, such as the processor 410 , the video display adapter 411 , the disk drive 412 , the input / output interface 413 , the network interface 414 , and the memory 420 .

[0105] It should be noted that although the above device only shows a processor 410, a video display adapter 411, a disk drive 412, an input / output interface 413, a network interface 414, a memory 420, a bus 430, etc., in a specific implementation, the device may also include other components necessary for normal operation. In addition, it will be understood by those skilled in the art that the above device may also include only the components necessary to implement the solution of the present application, and does not necessarily include all the components shown in the figure.

[0106] Through the description of the above embodiments, it can be seen that those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer program product. The computer program product can be stored in a storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present application or certain parts of the embodiments.

[0107] The above is a detailed introduction to the technical solutions provided by this application. Specific examples are used herein to illustrate the principles and implementation methods of this application. The description of the above embodiments is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the contents of this specification should not be understood as limiting this application.

Claims

1. A method for generating artificial intelligence advertisements based on user portraits, characterized in that: The method comprises: Obtain user behavior information and device status information, wherein the user behavior information includes historical behavior data and real-time behavior data; determining a device type and a network status according to the device status information; determining the user's attention state information based on the user behavior information; Generate a multi-dimensional user profile based on the device type, network status, and attention status information; Determining an advertisement length type according to the user profile, the advertisement length types including: ultra-short, short, and full; An advertisement is generated according to the advertisement length type and the user portrait by using an advertisement editing model.

2. The method according to claim 1, characterized in that The real-time behavior data includes: current page dwell time and interaction frequency; Determining the user's attention state information based on the user behavior information includes: Determining the attention state as high attention, low attention, or distracted state based on the current page dwell time and the interaction frequency; In combination with the historical behavior data, the attention state determination result is corrected by comparing the current behavior with the user's historical behavior pattern.

3. The method according to claim 1, characterized in that Determining the advertisement length type according to the user portrait includes: Determine the advertisement length type according to the user profile using the pre-trained advertisement length determination model; Wherein, the pre-training includes: Obtaining a training data set, the training data set including: multi-dimensional user portrait samples and their corresponding advertisement length type labels and advertisement viewing behavior data; The advertisement length determination model is trained using a supervised learning method, with user profiles as input; The advertisement length determination model outputs a probability distribution of each length type and determines the advertisement length type according to a preset threshold or a maximum probability; During the training process, time dimension features are added, and the time dimension features include: current time period and usage scenario information; The training objective of the pre-training includes minimizing the difference between the advertisement length type and the advertisement length type label.

4. The method according to claim 1, wherein Determining the advertisement length type according to the user portrait includes: For users whose attention status information indicates a low attention state or a low-speed network user, determining the advertisement length type to be an ultra-short type; For users whose attention state information indicates a medium attention state or a medium speed network user, determining the advertisement length type to be a shorter type; For users whose attention state information indicates a high attention state or a high-speed network user, determining that the advertisement length type is a complete type; determining the user's historical advertising tolerance, real-time feedback, and / or the content currently being browsed by the user based on the user behavior data; The advertisement length type is dynamically adjusted based on the user's historical advertisement tolerance, real-time feedback, and / or the content currently browsed by the user.

5. The method according to claim 4, wherein dynamically adjusting the advertisement length type based on the user's historical advertisement tolerance and the real-time feedback comprises: Collecting the user's past ad viewing history, wherein the ad viewing history includes: the complete viewing duration, skipping time point, and / or skipping rate of each ad; Calculating a user's advertising tolerance threshold, wherein the tolerance threshold is determined by counting the user's average viewing time or the critical time for skipping advertisements; Classifying users into low tolerance type, medium tolerance type, and high tolerance type according to the tolerance threshold; Match the corresponding advertisement length according to the user classification to obtain a first length value; During the ad playback process, real-time monitoring of user interaction behavior to obtain real-time feedback, such as whether to skip the ad, perform a muting operation, fast forward, and / or switch applications; If the real-time feedback shows that the user's skipping rate increases or the tolerance decreases, shortening the first length value; If the real-time feedback shows that the user has watched the entire ad and the frequency of interactive behavior is high, then the first length value is extended; The advertisement length type is determined according to the first length value.

6. The method according to claim 1, characterized in that The historical behavior data includes the user's historical preferred video clips, image styles and text content; The step of generating an advertisement by using the advertisement editing model according to the advertisement length type and the user portrait includes: Based on the historical behavior data, selecting materials that match the user profile from an advertising material library; Using computer vision technology to extract key images from the material, and adjusting the editing rhythm and image effects based on the attention state, the editing rhythm includes: image switching frequency and information density; Leverage natural language processing technology to optimize advertising copy and voiceover; Generate personalized advertising content that meets user preferences through generative adversarial networks or text-to-video technology.

7. The method according to claim 1, characterized in that The step of generating an advertisement by using the advertisement editing model according to the advertisement length type and the user portrait includes: Acquire weather condition information, classify the weather conditions according to the weather condition information, and obtain weather types, wherein the weather types include: suitable for going out and not suitable for going out; The advertisement editing model modifies the advertisement length type according to the weather type, and generates an advertisement according to the advertisement length type and the user portrait.

8. An artificial intelligence advertisement generation device based on user portrait, characterized in that: The device comprises: an information acquisition unit configured to acquire user behavior information and device status information, wherein the user behavior information includes historical behavior data and real-time behavior data; an information analysis unit, configured to determine a device type and a network status based on the device status information; an attention state determining unit, configured to determine the user's attention state information based on the user behavior information; A user portrait generating unit, configured to generate a multi-dimensional user portrait based on the device type, network status, and attention status information; a length type determining unit, configured to determine an advertisement length type according to the user portrait, the advertisement length types including: ultra-short, shorter, and full; The advertisement generating unit is configured to generate an advertisement according to the advertisement length type and the user portrait by using an advertisement editing model.

9. An electronic device, characterized in that: include: one or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions, wherein when the program instructions are read and executed by the one or more processors, the steps of the method according to any one of claims 1 to 7 are performed.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images