Precise advertisement putting method and system based on artificial intelligence

Abstract

The invention discloses an accurate advertisement putting method and system based on artificial intelligence, and relates to the technical field of information, and the method comprises the steps: collecting user behavior data and real-time situation data, and carrying out the preprocessing of the user behavior data; analyzing the preprocessed user behavior data to generate a user interest label; combining the real-time situation data with the user interest label, constructing a situation interest model by adopting a deep learning technology, and predicting a user interest change mode; using an interest-driven recommendation method to obtain an initialized advertisement putting strategy; generating feedback information based on the analysis of the initialized advertisement putting strategy; and optimizing the feedback information through a reinforcement learning algorithm to generate an optimized advertisement display strategy. By vectorizing the user interest and combining the advertisement strategy characteristics, the generator model can meet the user preference and follow the advertisement materials of the optimal display strategy, so that the visual experience and the emotion resonance of the user are improved, and finally the click rate and the conversion rate of the advertisement are improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an artificial intelligence-based precise advertising delivery method and system. Background Art

[0002] With the rapid development of Internet technology and the popularization of smart devices, the advertising industry has undergone a transformation from traditional media to digitalization and personalization. In recent years, online advertising, especially precise advertising delivery methods based on big data and artificial intelligence, has been widely used. Traditional advertising delivery methods mainly rely on preset audience group segmentation, and the effectiveness of advertising delivery is often limited by time, region and audience group.

[0003] Many advertising delivery platforms rely on static interest tags and behavioral data analysis methods, and are unable to capture and respond to dynamic changes in user interests in real time. Many recommendation systems use static models based on historical data and react slowly to changes in user interests, resulting in delivery strategies that cannot be optimized in a timely manner. They lack the ability to update in real time through feedback information, which greatly reduces the accuracy and real-time nature of advertising delivery. Its application in the field of advertising delivery is still in the exploratory stage. How to effectively apply generative adversarial network technology to the personalized generation of advertising materials so that they are highly matched with user interests is a major technical challenge. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an artificial intelligence-based precise advertising delivery method to solve the problem in the prior art that it is impossible to capture user interest changes in real time and optimize advertising delivery strategies.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an artificial intelligence-based precise advertising delivery method, which includes collecting user behavior data and real-time context data, and preprocessing the user behavior data;

[0008] Analyze the preprocessed user behavior data and generate user interest tags;

[0009] Combining real-time contextual data with user interest tags, using deep learning technology to build a contextual interest model to predict user interest change patterns;

[0010] Use interest-driven recommendation method to get initial advertising delivery strategy;

[0011] Generate feedback information based on the analysis of the initial advertising delivery strategy;

[0012] Feedback information is optimized through reinforcement learning algorithms to generate optimized advertising display strategies.

[0013] As a preferred solution of the artificial intelligence-based precise advertising delivery method of the present invention, wherein: the user behavior data includes user browsing history, search history, purchase history and interaction data;

[0014] The preprocessing includes data cleaning and data standardization.

[0015] As a preferred solution of the method for accurate advertising delivery based on artificial intelligence described in the present invention, the following steps are included in the analysis of the pre-processed user behavior data and the generation of user interest tags:

[0016] Based on the preprocessed user behavior data, the user's interest features are extracted through collaborative filtering;

[0017] Using LDA topic modeling, the user's interest features are modeled into interest topic distribution, the distribution of each user on different interest topics is obtained, and the interest topic of each user is identified;

[0018] Based on each interest topic in the LDA model, the high-frequency words of each topic are extracted through collaborative filtering, and the mapping relationship between the topic and the advertising category is established according to the similarity between the high-frequency words of each topic and the advertising category;

[0019] According to the mapping relationship between topics and advertising categories, the topic distribution of each user in the LDA model is associated with the corresponding advertising category to generate user interest tags.

[0020] As a preferred solution of the method for precise advertising delivery based on artificial intelligence described in the present invention, the following steps are included: combining real-time context data with user interest tags, using deep learning technology to build a context interest model, and predicting the user interest change pattern:

[0021] Use graph neural networks to take users’ interest tags and context data as nodes in the graph neural network structure, model the relationship between nodes through neural networks, and build a multi-layer graph model of user-context-interest;

[0022] Input real-time context data and user interest tags into the multi-layer graph model of user-context-interest, adjust the changes of different context factors on user interests, and obtain the deep correlation between context changes and interest changes;

[0023] Use data processing methods with different time granularities to capture the changing trend of the deep correlation between situational changes and interest changes, and predict interest changes at different time scales;

[0024] A variational autoencoder combined with an adversarial network is used to construct a situational interest model. The predicted interest changes at different time scales are input into the situational interest model to predict the user's interest change patterns in different situations and obtain the user preference changes in different situations.

[0025] As a preferred solution of the method for accurate advertising delivery based on artificial intelligence described in the present invention, wherein: according to user interest tags, using an interest-driven recommendation method to obtain an initialization advertising delivery strategy includes the following steps:

[0026] The user's preference for different advertising categories is obtained from the personalized user interest tags. Through the user interest tags, the interest-driven recommendation method is used to combine the user's preference for different advertising types to calculate the initial score of the advertisement. The expression is:

[0027] ;

[0028] in, is the initial score of the advertisement, is the number of interest tags, For users For The interaction value between interest-based tags and recommended ads, is the similarity between the target ad and the recommended ad, For the The weight parameter of the interest-like tag, For users, For targeted advertising, To recommend an ad, is the index of the target ad, An index of recommended ads;

[0029] Sort the initial scores of the advertisements from the highest score to the lowest score using a descending sorting algorithm;

[0030] A score threshold is set. When the score threshold is greater than the initial score of the advertisement, the advertisement is a high-scoring advertisement. The high-scoring advertisement is displayed first, and the initialization advertisement delivery strategy is obtained.

[0031] As a preferred solution of the artificial intelligence-based precise advertising delivery method of the present invention, wherein: based on the analysis of the initialization advertising delivery strategy, generating feedback information includes the following steps:

[0032] By initializing the advertising delivery strategy, high-scoring ads are arranged to be displayed during the time period when users are most active, and the interaction data between users and ads after the ads are displayed is collected to obtain an effective advertising collection;

[0033] Analyze the emotional expressions of users in effective advertising collections through NLP technology to obtain feedback at the emotional level;

[0034] Combine emotional feedback, contextual data, and user interest tags, use a weighted approach to evaluate the suitability of ads, and get an ad suitability score.

[0035] Based on the suitability score of the advertisement, the actual performance of each advertisement is analyzed using the advertisement effectiveness evaluation method to obtain specific feedback information.

[0036] As a preferred solution of the artificial intelligence-based advertising precision delivery method of the present invention, wherein: optimizing the feedback information by a reinforcement learning algorithm and generating an optimized advertising display strategy include the following steps:

[0037] Based on the interaction data between users and ads and user interest tags, deep The network algorithm constructs the state space;

[0038] The reward function is used to define the feedback information after the advertisement is released, with the goal of maximizing the conversion rate of the advertisement and the user's satisfaction. The expression is:

[0039] ;

[0040] in, is the reward value, is the weight parameter of click rate, is the click rate, is the conversion rate, is the penalty item for ad display frequency. is the weight parameter of the ad display frequency penalty item, The weight of conversion rate

[0041] Heavy parameters;

[0042] According to the initial advertising delivery strategy, the Q-value function is estimated through a neural network, the optimal action is selected in a given state space, and the Q-value function is updated using Q-learning;

[0043] Optimize the feedback information each time an ad is displayed. The value function is continuously updated according to the new reward value, and the ad display strategy is also optimized accordingly to obtain the optimized ad display strategy.

[0044] According to the initial advertising delivery strategy, the Q-value function is estimated through the neural network, the optimal action is selected in the given state space, and the Q-value function is updated using Q-learning. The expression is:

[0045] ;

[0046] in, For the current state Next select action The current value of is the learning rate, is the reward value, is the discount factor, For the status All Actions maximum value.

[0047] In a second aspect, the present invention provides an artificial intelligence-based precise advertising delivery system, including an artificial intelligence-based precise advertising delivery method, characterized in that: including:

[0048] Data collection module, collects user behavior data and performs preprocessing;

[0049] User interest tag module, which analyzes the preprocessed behavior data and generates user interest tags;

[0050] The user interest change prediction module combines real-time context data and user interest tags, uses deep learning technology to build a context interest model, and predicts the user interest change pattern;

[0051] Initialize the advertising delivery strategy module and use the interest-driven recommendation method to obtain the initial advertising delivery strategy;

[0052] A feedback information module generates feedback information based on the analysis of the initial advertisement delivery strategy;

[0053] The module for optimizing advertising display strategy optimizes feedback information through reinforcement learning algorithm and generates optimized advertising display strategy.

[0054] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the artificial intelligence-based precise advertising delivery method as described in the first aspect of the present invention is implemented.

[0055] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the artificial intelligence-based precise advertising delivery method as described in the first aspect of the present invention.

[0056] The beneficial effects of the present invention are as follows: the behavioral data is divided into different data types through the data source allocation method, and is cleaned and standardized, thereby improving the accuracy of data analysis and making the advertising delivery strategy more accurate; through the weighted sum method and LDA topic modeling, the interest topic distribution of each user can be identified and associated with a specific advertising category, thereby generating personalized interest tags, making the advertisement more in line with the user's actual interests, increasing the user's engagement and conversion rate; through continuous learning and optimization, it can automatically adapt to market changes and the evolution of user preferences, maximize the conversion rate of advertisements and user satisfaction, and significantly improve the effect of advertising delivery; by quantizing user interests and combining them with advertising strategy features, the generator model can generate advertising materials that meet user preferences and follow the best display strategy, thereby enhancing the user's visual experience and emotional resonance, and ultimately improving the click-through rate and conversion rate of advertisements. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0058] Figure 1 This is a flow chart of the method for precise advertising delivery based on artificial intelligence in Example 1.

[0059] Figure 2 This is a system diagram of the artificial intelligence-based precise advertising delivery system in Example 1. DETAILED DESCRIPTION

[0060] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0061] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0062] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0063] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides an artificial intelligence-based advertising precision delivery method, comprising the following steps:

[0064] S1. Collect user behavior data and real-time context data, and pre-process the user behavior data.

[0065] Furthermore, user behavior data includes user browsing history, search history, purchase history and interaction data. During the data collection stage, the data source allocation method is used to classify user behavior data from different channels, specifically identifying and separating four main data types: user browsing history (such as visited web pages and dwell time), search history (such as search keywords and frequency), purchase history (purchased goods and transaction amount) and interaction data (such as likes, comments and shares). Real-time contextual data includes geographic location and weather conditions.

[0066] S1.1. Preprocessing includes data cleaning and data standardization.

[0067] Furthermore, after collecting user behavior data, we first clean the data to remove duplicates, correct erroneous data, fill in missing values, and exclude abnormal or noisy data points. Then we implement data standardization to convert all data into a unified format and scale, such as unifying timestamps into a standard format, normalizing numerical data, and encoding text data, which improves the availability and reliability of the data.

[0068] S2. Analyze the preprocessed user behavior data and generate user interest tags.

[0069] S2.1. Based on the preprocessed user behavior data, the user's interest features are extracted through collaborative filtering.

[0070] Furthermore, by using user behavior data (such as browsing history, search history, and purchase history), the similarity between users is calculated through collaborative filtering to find user groups with similar interests. By calculating each user's potential interest value in the item, the user's interest feature vector is generated to represent the user's preference for different item categories. This interest feature vector will serve as the basis for subsequent steps and can capture the user's implicit interest characteristics and behavior patterns.

[0071] S2.2. Using LDA topic modeling, the user's interest characteristics are modeled into interest topic distribution, the distribution of each user on different interest topics is obtained, and the interest topic of each user is identified.

[0072] Furthermore, the interest feature vector extracted from collaborative filtering is used as the input of LDA topic modeling. Multiple interest topics are identified through the LDA model. Each interest topic consists of a set of high-frequency words, indicating the interest direction of a certain category. The LDA model generates an interest topic distribution for each user based on the user's behavior pattern (such as browsing and purchasing items), indicating the user's interest intensity in different topics. Each user's interest topic distribution vector helps identify the user's core interest area.

[0073] S2.3. Based on each interest topic in the LDA model, the high-frequency words of each topic are extracted through collaborative filtering, and the mapping relationship between the topic and the advertising category is established according to the similarity between the high-frequency words of each topic and the advertising category.

[0074] Furthermore, by analyzing each interest topic generated by the LDA model, high-frequency words related to the topic are extracted. These high-frequency words can be certain item categories, brands or other keywords. The collaborative filtering method is used to calculate the similarity between these high-frequency words and each advertising category, and a mapping relationship between interest topics and advertising categories is established. If the high-frequency words of the topic are "smartphones", this topic can be associated with the "electronic products" category. Through this mapping, the interest topics identified by LDA can be matched with specific advertising categories, thereby improving the accuracy of advertising delivery.

[0075] S2.4. According to the mapping relationship between topics and advertising categories, the topic distribution of each user in the LDA model is associated with the corresponding advertising category to generate user interest tags.

[0076] Furthermore, according to the distribution of interest topics of each user in the LDA model and the mapping relationship between topics and advertising categories constructed in the previous step, the user's interest topics are matched with the corresponding advertising categories, and each user's interest topic is mapped to multiple advertising categories to generate personalized user interest tags. When user A has a high interest in the topic of "electronic products", the tag "electronic products" is generated. At the same time, corresponding tags are generated based on other topics such as "fashion clothing" or "fitness". The user interest tag will contain multiple advertising categories to reflect the user's diverse interests.

[0077] S3. Combine real-time contextual data with user interest tags, use deep learning technology to build a contextual interest model, and predict user interest change patterns.

[0078] S3.1. Use graph neural networks to take users’ interest tags and context data as nodes in the graph neural network structure, model the relationship between nodes through neural networks, and build a multi-layer graph model of user-context-interest.

[0079] Furthermore, to encode the user's interest tags, word embedding technology can be used to convert the textual interest tags into vector representations, and the geographic location can be represented by longitude and latitude coordinates, and the weather conditions can be classified and converted into category codes;

[0080] In the graph neural network, user interest tags and context data are mapped to graph nodes. Users can be regarded as one node in the graph, interest tags as another node, and context data (such as geographic location and weather) as another type of node. A multi-layer graph structure is designed based on historical interest tags and historical context data. Each layer contains users, interest tags, and context data nodes. Different layers are connected by specific rules to simulate complex interactions in reality.

[0081] By using graph neural networks to train multi-layer graph structures, GNN can automatically learn the dependencies between nodes and reveal the changing patterns of user interests in different situations.

[0082] S3.2. Input the real-time context data and user interest tags into the multi-layer graph model of user-context-interest, adjust the changes of different context factors on user interests, and obtain the deep correlation between context changes and interest changes.

[0083] Furthermore, real-time contextual data and user interest tags are used as input to the constructed multi-layer graph model, and the message passing mechanism of the graph neural network is used to propagate information in the graph neural network. Each round of propagation will update the representation of the current node according to the node information, thereby reflecting the impact of contextual factors on user interests.

[0084] In the process of information dissemination, we use multi-layer graph structures to analyze the impact of contextual changes on interest changes from the short-term (such as immediate weather changes), medium-term (such as seasonal changes), and long-term (such as evolution of user habits) time scales. We use deep neural networks to identify deep and multi-variable interactions in contextual data, capture the complex interactions between contextual variables and their impact on user interests, and mine complex patterns and deep associations in contextual data.

[0085] S3.3. Use data processing methods with different time granularities to capture the changing trends of the deep correlation between situational changes and interest changes, and predict interest changes at different time scales.

[0086] Furthermore, data processing methods with different time granularities are used to capture the changing trends of the deep correlation between situational changes and interest changes. Real-time situational data and user interest tags are divided and aggregated according to different time scales. For short-term changes (such as immediate changes in the weather), high-frequency time granularity (such as hourly or daily updates) is used to quickly respond to fluctuations in user interests. For long-term changes (such as the impact of seasonal preferences or annual events), lower-frequency time granularity (such as weekly or monthly aggregation) is used to identify and predict trends. For short-term changes, sliding windows (such as hourly) are used and LSTM processing is used to capture instantaneous interest fluctuations. For long-term changes, time series analysis (such as weekly or monthly) is used to identify long-term change trends, ensuring that various interest change patterns from short-term to long-term can be effectively captured, accurately predicting users' interest change trends in future real-time situations, and providing a solid basis for personalized recommendations and services.

[0087] S3.4. Use variational autoencoder combined with adversarial network to build a situational interest model, input the predicted interest changes at different time scales into the situational interest model, predict the user's interest change pattern in different situations, and obtain the user preference changes in different situations.

[0088] Furthermore, VAE is used to encode the user's contextual data and interest tags. VAE is first used to map the user's contextual data (such as geographic location, weather conditions) and interest tags to a low-dimensional latent space. The user's interest pattern is constructed in the decoding stage. The generator in GAN is used to discriminate the similarity between the generated interest pattern and the real interest pattern. The adversarial training between the two is used to improve the authenticity and accuracy of the generated interest pattern. The entire model contains an encoder (for feature extraction), a decoder (for interest reconstruction), a generator (for generating new interest patterns) and a discriminator (for identifying authenticity). The variational autoencoder (VAE) is combined with the generative adversarial network (GAN) to construct a contextual interest model. The predicted interest changes at different time scales are used as input data. VAE is used to learn the latent space of user interest distribution, accurately predict the user's interest change trend, and provide personalized recommendations, thereby realizing accurate simulation and prediction of user preference changes, predicting user interest changes, and then obtaining user preference change trends in different contexts, thereby improving the accuracy of advertising and recommendations.

[0089] S4. Use interest-driven recommendation method to obtain the initial advertising delivery strategy.

[0090] S4.1. Obtain the user's preference for different advertising categories in the personalized user interest tags. Through the user interest tags, use the interest-driven recommendation method and combine the user's preference for different advertising types to calculate the initial score of the advertisement. The expression is:

[0091] ;

[0092] in, is the initial score of the advertisement, is the number of interest tags, For users For The interaction value between interest-based tags and recommended ads, is the similarity between the target ad and the recommended ad, For the The weight parameter of the interest-like tag, For users, For targeted advertising, To recommend an ad, is the index of the target ad, Index of recommended ads

[0093] Furthermore, by embedding the user's interest tags and the features of the ads into the same latent space, matrix decomposition is used to generate the embedding representation of the user's interest tags and the ad features, and the interaction value between the interest tags and the recommended ads is determined by calculating the vector similarity between the user's interest tags and the ad categories;

[0094] Based on the personalized interest tags generated by users, we first determine the user's preference for different advertising categories, and use the interest-driven recommendation method to combine the user's preference for various advertising types to calculate the initial score of each advertisement. The formula comprehensively considers the similarity between advertisements and the user's preference for different interest tags to ensure that the recommended advertisements not only meet the user's interests, but also stand out from similar advertisements, thereby improving the relevance and accuracy of the recommendations.

[0095] S4.2. Use a descending sorting algorithm to sort the initial scores of the advertisements from the highest score to the lowest score.

[0096] Furthermore, a descending sorting algorithm is used to sort all ads from high to low according to their initial scores. Then, a score threshold is set. Only ads with scores higher than the score threshold are considered high-scoring ads and are displayed to users first, ensuring that the ads displayed are high-quality candidate ads that have been strictly screened, thereby improving user experience and ad click-through rate.

[0097] S4.3. Set a score threshold. When the score threshold is greater than the initial score of the advertisement, it is a high-scoring advertisement. The high-scoring advertisement is displayed first, and the initialization advertisement delivery strategy is obtained.

[0098] Furthermore, for the ranked ad list, when the initial score of an ad is greater than the initial score of the ad, the ad is considered to be a high-scoring ad and will be displayed first; otherwise, the ad will be delayed or not displayed. This ensures that ads that are more in line with user interests and more likely to attract user attention are displayed first, thereby optimizing the ad delivery strategy.

[0099] S5. Generate feedback information based on the analysis of the initialization advertisement delivery strategy.

[0100] S5.1. By initializing the advertising delivery strategy, high-scoring ads are arranged to be displayed during the time period when users are most active, and the interaction data between users and ads after the ads are displayed is collected to obtain an effective advertising set.

[0101] Furthermore, according to the initialization advertising delivery strategy, high-scoring ads are prioritized for display during the time periods when users are most active. Specific operations include identifying the active time periods of users and displaying screened ads with high initial scores during this period. User interaction data after ad display, such as clicks, dwell time, and conversions, is collected in real time to form an effective ad collection.

[0102] S5.2. Analyze the emotional expressions of users in effective advertising collections through NLP technology to obtain feedback at the emotional level.

[0103] Furthermore, when interacting with advertisements, users may express their emotional attitudes in comments, social media, or other text data. Sentiment analysis technology can automatically identify and classify these emotions, such as positive, negative, or neutral emotions. The vocabulary, tone, and grammatical structure that appear in user comments can reflect their likes or dissatisfaction with the advertising content, and obtain emotional feedback.

[0104] S5.3. Combine emotional feedback, contextual data, and user interest tags, use a weighted method to evaluate the suitability of the advertisement, and obtain an advertisement suitability score.

[0105] Furthermore, emotional feedback provides users' subjective reactions to advertising content, contextual data reflects the external environment when the advertisement is displayed, and user interest tags reveal users' potential preferences. Each advertisement in the effective advertisement set is compared with the user's interest tag, and a weighted method is used to calculate the adaptation score of each advertisement, taking into account the user's preference for different interest tags and the matching degree between the advertisement and these tags. This method not only measures the relevance of the advertisement content to the user's interests, but also takes into account the performance of the advertisement in a specific time period, thereby obtaining a more accurate adaptation score.

[0106] S5.4. Based on the suitability score of the advertisement, use the advertisement effect evaluation method to analyze the actual performance of each advertisement and obtain specific feedback information.

[0107] Furthermore, based on the adaptability score of the advertisement, a variety of advertising effectiveness evaluation methods can be used to analyze the actual performance of the advertisement. The effectiveness of the advertisement can be evaluated through click-through rate, conversion rate and exposure indicators. For each advertisement, the actual performance is calculated based on its display times and user interactions (such as clicks, purchases and comments). Through these feedback information, it is evaluated whether the advertisement has achieved the expected effect, and the advertisement display strategy is adjusted according to user behavior. The actual performance analysis can not only help optimize advertising delivery, but also provide data support for subsequent advertising strategy adjustments, thereby continuously improving advertising delivery efficiency and user satisfaction.

[0108] S6. Optimize feedback information through reinforcement learning algorithm to generate optimized advertising display strategy.

[0109] S6.1. Based on the interaction data between users and ads and user interest tags, use deep The network algorithm builds the state space.

[0110] Furthermore, we collect interaction data between users and ads, including clicks, browsing and purchasing behaviors, as well as users’ interest tags. We build a state space through the interaction data. Using the deep Q network (DQN) algorithm, we can convert these interaction data into a state representation and use it as input to predict the optimal advertising delivery strategy. The construction of the state space ensures that the advertising display strategy is highly correlated with users’ interests and behaviors, thereby improving the effectiveness of advertising delivery.

[0111] S6.2. Use the reward function to define the feedback information after the advertisement is placed, with the goal of maximizing the conversion rate of the advertisement and the user's satisfaction. The expression is:

[0112] ;

[0113] in, is the reward value, is the weight parameter of click rate, is the click rate, is the conversion rate, is the penalty item for ad display frequency. is the weight parameter of the ad display frequency penalty item, is the weight parameter of conversion rate.

[0114] Furthermore, after the advertisement is released, the reward function is defined by the user's behavioral feedback (such as clicks, conversions, etc.). The reward value consists of three parts: click-through rate, conversion rate and display frequency penalty. The click-through rate measures the attractiveness of the advertisement, the conversion rate measures the actual effect of the advertisement, and the display frequency is used to avoid user fatigue caused by excessive display of advertisements. By setting weight parameters, the importance of these three factors in the reward can be controlled, thereby ensuring that the advertising delivery strategy can improve the user conversion rate while avoiding the negative effects caused by excessive display. This reward function will provide effective feedback for the reinforcement learning algorithm and help optimize the advertising display strategy.

[0115] S6.3. According to the initialization advertising delivery strategy, estimate the Q-value function through the neural network, select the optimal action in the given state space, and use Q-learning to update the Q-value function.

[0116] Furthermore, according to the initialized advertising delivery strategy, a neural network is first used to estimate the Q-value function. The Q-value function represents the expected return of taking an action (such as displaying an advertisement) under a given state. Under the DQN framework, the neural network predicts the Q-value based on the state space and action selection. Through training, the network gradually approaches the optimal Q-value. After each advertisement is displayed, the Q-value function is updated according to the actual reward information. The strategy is adjusted through the Q-learning algorithm, and the advertisement display action that can maximize the long-term return is selected, so that the advertisement display strategy is gradually optimized.

[0117] S6.4. Optimize the feedback information of each advertisement when it is displayed. The value function is continuously updated according to the new reward value, and the ad display strategy is also optimized accordingly to obtain the optimized ad display strategy.

[0118] Furthermore, after each ad display, a new reward value is calculated based on user feedback (such as clicks and conversions), and the reward value and the current state and action are input into the Q-value function for update. In this way, the Q-value function will continuously receive the latest feedback information, so that the ad display strategy can be adjusted and optimized in each delivery. With continuous training, the Q-value function gradually tends to the optimal level, and the ad display strategy becomes more and more precise, so as to improve the conversion rate of ads, user satisfaction and the overall effect of advertising delivery.

[0119] S6.5. According to the initial advertising delivery strategy, estimate the Q value function through the neural network, select the optimal action in the given state space, and use Q-learning to update the Q value function. The expression is:

[0120] ;

[0121] in, For the current state Next select action The current value of is the learning rate, is the reward value, is the discount factor, For the status All Actions maximum value.

[0122] Furthermore, based on the initialization of the advertising delivery strategy, a neural network is used to estimate the Q-value function of different actions in each state. The Q-value function represents the Q-value function of different actions in the current state. Next select action The expected return is continuously updated through the Q-learning algorithm to find the optimal action sequence. After each ad is displayed, the actual reward value is calculated. and the next state The maximum Q value, adjust the Q value in the current state, and the learning rate Control the update speed, discount factor Weighing the importance of immediate rewards and future rewards can gradually optimize the ad display strategy in practice.

[0123] This embodiment also provides an artificial intelligence-based precise advertising delivery system, including: a data collection module, which collects user behavior data and performs pre-processing;

[0124] User interest tag module, which analyzes the preprocessed behavior data and generates user interest tags;

[0125] The user interest change prediction module combines real-time context data and user interest tags, uses deep learning technology to build a context interest model, and predicts the user interest change pattern;

[0126] Initialize the advertising delivery strategy module and use the interest-driven recommendation method to obtain the initial advertising delivery strategy;

[0127] A feedback information module generates feedback information based on the analysis of the initial advertisement delivery strategy;

[0128] The module for optimizing the advertising display strategy optimizes the feedback information through the reinforcement learning algorithm and generates an optimized advertising display strategy.

[0129] This embodiment also provides a computer device, which is suitable for the case of an artificial intelligence-based precise advertising delivery method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the artificial intelligence-based precise advertising delivery method proposed in the above embodiment.

[0130] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0131] The present embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for implementing the precise advertising delivery based on artificial intelligence as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.

[0132] In summary, the present invention improves the accuracy of data analysis and makes the advertising delivery strategy more accurate by: dividing the behavioral data into different data types through the data source allocation method, and cleaning and standardizing them; through the weighted sum method and LDA topic modeling, it can identify the interest topic distribution of each user and associate it with a specific advertising category, thereby generating personalized interest tags, making the advertisement more in line with the user's actual interests, increasing the user's engagement and conversion rate; through continuous learning and optimization, it can automatically adapt to market changes and the evolution of user preferences, maximize the conversion rate of advertisements and user satisfaction, and significantly improve the effect of advertising delivery; by quantizing user interests and combining them with advertising strategy features, the generator model can generate advertising materials that meet user preferences and follow the best display strategy, thereby enhancing the user's visual experience and emotional resonance, and ultimately improving the click-through rate and conversion rate of advertisements.

[0133] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based precise advertising delivery method, characterized by: include, Collect user behavior data and real-time context data, and pre-process user behavior data; Analyze the preprocessed user behavior data and generate user interest tags; Combining real-time contextual data with user interest tags, using deep learning technology to build a contextual interest model to predict user interest change patterns; Use interest-driven recommendation method to get initial advertising delivery strategy; Generate feedback information based on the analysis of the initial advertising delivery strategy; Feedback information is optimized through reinforcement learning algorithms to generate optimized advertising display strategies.

2. The method for accurate advertising delivery based on artificial intelligence according to claim 1, characterized in that: The user behavior data includes user browsing history, search history, purchase history and interaction data, and the real-time situational data includes geographic location and weather conditions; The preprocessing includes data cleaning and data standardization.

3. The method for accurate advertising delivery based on artificial intelligence as claimed in claim 2, characterized in that: Analyzing the preprocessed user behavior data and generating user interest tags includes the following steps: Based on the preprocessed user behavior data, the user's interest features are extracted through collaborative filtering; Using LDA topic modeling, the user's interest features are modeled into interest topic distribution, the distribution of each user on different interest topics is obtained, and the interest topic of each user is identified; Based on each interest topic in the LDA model, the high-frequency words of each topic are extracted through collaborative filtering, and the mapping relationship between the topic and the advertising category is established according to the similarity between the high-frequency words of each topic and the advertising category; According to the mapping relationship between topics and advertising categories, the topic distribution of each user in the LDA model is associated with the corresponding advertising category to generate user interest tags.

4. The method for accurate advertising delivery based on artificial intelligence as claimed in claim 3, characterized in that: Combining real-time contextual data with user interest tags, using deep learning technology to build a contextual interest model, and predicting the user interest change pattern includes the following steps: Use graph neural networks to take users’ interest tags and context data as nodes in the graph neural network structure, model the relationship between nodes through neural networks, and build a multi-layer graph model of user-context-interest; Input real-time context data and user interest tags into the multi-layer graph model of user-context-interest, adjust the changes of different context factors on user interests, and obtain the deep correlation between context changes and interest changes; Use data processing methods with different time granularities to capture the changing trend of the deep correlation between situational changes and interest changes, and predict interest changes at different time scales; A variational autoencoder combined with an adversarial network is used to construct a situational interest model. The predicted interest changes at different time scales are input into the situational interest model to predict the user's interest change patterns in different situations and obtain the user preference changes in different situations.

5. The method for accurate advertising delivery based on artificial intelligence according to claim 4, characterized in that: Using the interest-driven recommendation method to obtain the initial advertising delivery strategy includes the following steps: According to the user preference changes in different situations, we get the user's preference for different advertising categories. We use the interest-driven recommendation method and combine the user's preference for different advertising types to calculate the initial score of the advertisement. The expression is: ； in, is the initial score of the advertisement, is the number of interest tags, For users For The interaction value between interest-based tags and recommended ads, is the similarity between the target ad and the recommended ad, For the The weight parameter of the interest-like tag, For users, For targeted advertising, To recommend an ad, is the index of the target ad, An index of recommended ads; Sort the initial scores of the advertisements from the highest score to the lowest score using a descending sorting algorithm; A score threshold is set. When the score threshold is greater than the initial score of the advertisement, the advertisement is a high-scoring advertisement. The high-scoring advertisement is displayed first, and the initialization advertisement delivery strategy is obtained.

6. The method for accurate advertising delivery based on artificial intelligence according to claim 5, characterized in that: Based on the analysis of the initial advertising delivery strategy, generating feedback information includes the following steps: By initializing the advertising delivery strategy, high-scoring ads are arranged to be displayed during the time period when users are most active, and the interaction data between users and ads after the ads are displayed is collected to obtain an effective advertising collection; Analyze the emotional expressions of users in effective advertising collections through NLP technology to obtain feedback at the emotional level; Combine emotional feedback, contextual data, and user interest tags, use a weighted approach to evaluate the suitability of ads, and get an ad suitability score. Based on the suitability score of the advertisement, the actual performance of each advertisement is analyzed using the advertisement effectiveness evaluation method to obtain specific feedback information.

7. The method for accurate advertising delivery based on artificial intelligence according to claim 6, characterized in that: The feedback information is optimized through the reinforcement learning algorithm, and the optimized advertising display strategy is generated, which includes the following steps: Based on the interaction data between users and ads and user interest tags, deep The network algorithm constructs the state space; The reward function is used to define the feedback information after the advertisement is released, with the goal of maximizing the conversion rate of the advertisement and the user's satisfaction. The expression is: ； in, is the reward value, is the weight parameter of click rate, is the click rate, is the conversion rate, is the penalty item for ad display frequency. is the weight parameter of the ad display frequency penalty item, is the weight parameter of conversion rate; According to the initial advertising delivery strategy, the Q-value function is estimated through a neural network, the optimal action is selected in a given state space, and the Q-value function is updated using Q-learning; Optimize the feedback information each time an ad is displayed. The value function is continuously updated according to the new reward value, and the ad display strategy is also optimized accordingly to obtain the optimized ad display strategy; According to the initialization advertising delivery strategy, the Q-value function is estimated through a neural network, the optimal action is selected in a given state space, and the Q-value function is updated using Q-learning.

8. An artificial intelligence-based precise advertising delivery system, based on the artificial intelligence-based precise advertising delivery method according to any one of claims 1 to 7, characterized in that: include, Data collection module, collects user behavior data and performs preprocessing; User interest tag module, which analyzes the preprocessed behavior data and generates user interest tags; The user interest change prediction module combines real-time context data and user interest tags, uses deep learning technology to build a context interest model, and predicts the user interest change pattern; Initialize the advertising delivery strategy module and use the interest-driven recommendation method to obtain the initial advertising delivery strategy; A feedback information module generates feedback information based on the analysis of the initial advertisement delivery strategy; The module for optimizing the advertising display strategy optimizes the feedback information through the reinforcement learning algorithm and generates an optimized advertising display strategy.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the artificial intelligence-based precise advertising delivery method described in any one of claims 1 to 7 are implemented.

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

Images