Internet advertisement marketing method and system based on artificial intelligence

Abstract

The invention discloses an Internet advertisement marketing method and system based on artificial intelligence, and relates to the technical field of Internet advertisement, and the method comprises the steps: collecting behavior data of a user in real time through the Internet, carrying out the preprocessing, extracting user behavior features, and fusing the user behavior features into a comprehensive feature vector; constructing a purchase intention dynamic prediction model, and predicting a commodity purchase probability of the user based on the comprehensive feature vector; generating a commodity recommendation list according to the commodity purchase probability of the user; generating personalized advertisement content based on the commodity recommendation list, and pushing the personalized advertisement content to the user according to the network usage scene of the user; in the advertisement pushing process, interaction data between the user and personalized advertisement content are continuously collected, and the purchase intention dynamic prediction model is optimized. By constructing the purchase intention dynamic prediction model, the purchase probability of the user is accurately predicted, and the click rate and the conversion rate of the advertisement are improved.

Description

Technical Field

[0001] The present invention relates to the field of Internet advertising technology, and in particular to an Internet advertising marketing method and system based on artificial intelligence. Background Art

[0002] With the rapid development of Internet technology, advertising marketing has gradually shifted from the traditional offline model to the online model. In particular, advertising push based on personalized user needs has become a key means in digital marketing. At present, intelligent advertising push technology mainly relies on big data analysis and artificial intelligence algorithms. It can predict users' potential needs and push targeted advertisements through users' historical behavior data, such as browsing records, purchase behaviors, search keywords, etc. This advertising push method based on user behavior analysis not only improves the accuracy of advertisements, but also improves the user experience and reduces the interference of irrelevant information. However, with the increasing complexity and diversity of user behavior data, how to more efficiently and accurately tap into users' real needs and form more personalized advertising content has become the main research direction of advertising marketing technology at this stage.

[0003] Existing intelligent advertising push technology still has certain limitations in the processing of user behavior data and the personalized generation of advertising content. First, the existing methods for analyzing user behavior data are mostly based on static models, lacking the ability to dynamically respond to users' real-time behavior, resulting in the possibility that the pushed advertising content may not match the user's immediate needs, affecting the click-through rate and conversion rate of the advertisement. Secondly, existing advertising push systems usually use a single recommendation algorithm, which fails to fully combine the short-term and long-term behavioral characteristics of users, making it difficult to accurately predict the user's purchase probability, which in turn affects the personalized recommendation effect of advertisements. Therefore, how to dynamically update user behavior characteristics in real time during the advertising push process, and conduct a comprehensive analysis based on the user's short-term and long-term behaviors, has become a key technical challenge to improve the accuracy of advertising push. 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 Internet advertising marketing method and system to solve the problems in the prior art of lack of dynamic response to user behavior data and inability to combine short-term and long-term behavior characteristics for accurate advertising push.

[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 Internet advertising marketing method based on artificial intelligence, which includes collecting user behavior data in real time through the Internet and preprocessing it, and extracting user behavior features and fusing them into a comprehensive feature vector; constructing a purchase intention dynamic prediction model, and predicting the user's purchase probability of a product based on the comprehensive feature vector; generating a product recommendation list based on the user's purchase probability of the product; generating personalized advertising content based on the product recommendation list, and pushing personalized advertising content to the user based on the user's network usage scenario; during the advertising push process, continuously collecting the user's interaction data with the personalized advertising content, and optimizing the purchase intention dynamic prediction model.

[0008] As a preferred solution of the artificial intelligence-based Internet advertising marketing method of the present invention, wherein: the user's behavior data includes the user's browsing history, search keywords, purchase history, click behavior and social media interaction;

[0009] The preprocessing of user behavior data is as follows:

[0010] Clean user behavior data by removing invalid data, correcting erroneous data, and filling missing values;

[0011] Use hashing algorithms to identify duplicate user behavior records and delete multiple occurrences of the same data to avoid data redundancy.

[0012] Filter outliers and noise data through denoising;

[0013] Through standardization, the timestamps in different user behavior data are converted into a unified format.

[0014] As a preferred solution of the artificial intelligence-based Internet advertising marketing method of the present invention, the specific steps of extracting user behavior characteristics are as follows:

[0015] Group user behavior data by time and divide it into different categories according to behavior type;

[0016] For each type of behavior data, extract the user's behavior frequency within a specific time range;

[0017] Through page event tracking, the duration and interaction depth of each user behavior can be extracted;

[0018] Extract behavior sequence features by analyzing the order in which users' behaviors occur;

[0019] Based on nonlinear transformation and feature product, various user behavior features are processed through different linear transformations, and a comprehensive feature vector is generated through the product interaction between features. The expression is:

[0020]

[0021] Among them, F is the comprehensive feature vector, f i represents the behavior characteristics of the i-th category, W i represents the weight matrix for linear transformation of the i-th type of behavior feature vector, n represents the total number of user behavior features, σ is the nonlinear activation function, and b i It is the bias term when linear transformation is performed on the behavior characteristics of the i-th category.

[0022] As a preferred solution of the artificial intelligence-based Internet advertising marketing method of the present invention, the following specific steps are used to construct a dynamic prediction model of purchase intention and predict the user's purchase probability of the product based on a comprehensive feature vector:

[0023] Through hierarchical behavior modeling, the purchase intention dynamic prediction model is divided into a short-term behavior evaluation layer and a long-term behavior evaluation layer;

[0024] The short-term behavior evaluation layer adopts a similarity weighting mechanism based on the Gaussian kernel. By comparing the difference between the current feature and the neighborhood feature, the weight is calculated and the neighborhood feature is weighted and summed through the ReLU activation function to extract the features of the short-term behavior. The expression is:

[0025]

[0026] Among them, Z s represents the short-term behavior characteristics of the user, F is the comprehensive feature vector, and F x represents the xth neighborhood feature in the comprehensive feature vector F, μ is the standard deviation of the Gaussian kernel, b is the bias term, k is the total number of short-term behavior feature vectors, and x is the index of the neighborhood feature;

[0027] The long-term behavior evaluation layer uses Memory Network to embed the user's long-term behavior into the memory unit, and selectively extracts and aggregates the long-term behavior features related to the user's purchase intention through the query mechanism. The expression is:

[0028]

[0029] Among them, Z l represents the long-term behavior characteristics of users, k j represents the jth key vector, M j represents the jth key vector h j The corresponding memory unit, W q Represents the weight matrix that the user uses to map the comprehensive feature vector F to the query vector;

[0030] The gating mechanism is used to dynamically adjust the purchase intention dynamic prediction model's attention to short-term and long-term behaviors. When the user shows an impulse to buy, the gating mechanism automatically increases the attention to short-term behaviors. When the user's behavior is stable, the gating mechanism pays more attention to the impact of long-term behaviors. The expression is:

[0031]

[0032] Among them, g represents the gating signal, σ is the nonlinear activation function, and ∈ represents the numerical offset;

[0033] The short-term behavior evaluation layer, the long-term behavior evaluation layer and the gating mechanism are combined through the gating mechanism and nonlinear fusion to form a dynamic prediction model of purchase intention;

[0034] The comprehensive feature vector is input into the purchase intention dynamic prediction model to predict the user's purchase probability of the product. The expression is:

[0035] y k =Sigmoid(W o ·(g·tanh(Z s,k )+(1-g)·tanh(Z l,k ))+b o );

[0036] Among them, y k represents the user's purchase probability of the kth item, W o The weight matrix used to linearly transform the gated feature combination, Z s,k represents the short-term behavior characteristics associated with the kth commodity, Z l,k Represents the long-term behavioral characteristics of the k-th commodity.

[0037] As a preferred solution of the artificial intelligence-based Internet advertising marketing method of the present invention, wherein: the product recommendation list is generated according to the user's purchase probability of the product, and the specific steps are as follows:

[0038] According to the user's purchase probability y for each product k , sort all products from high to low, select the first N products after sorting, and generate a preliminary product recommendation list R;

[0039] Based on the user's historical behavior data, calculate the behavior similarity between the current user and other users;

[0040] Based on the similarity between the current user and other users, select the M users with the highest similarity to the current user as the similar user set U;

[0041] Based on the similar user set U, calculate the collaborative filtering score of similar users for product k;

[0042] The probability y of the user purchasing the kth item k and collaborative filtering score r CF,k Combined, a new sorting index is generated, the expression is:

[0043] S k =λ·y k +(1-λ)·r CF,k ;

[0044] Among them, r CF,k represents the collaborative filtering score of similar users for product k, and λ is the purchase probability y of the balanced user for the kth product k and collaborative filtering score r CF,k The weight coefficient, S k is a new ranking metric, CF stands for collaborative filtering;

[0045] According to the new sorting index S k The preliminary product recommendation list R is optimized to generate a final product recommendation list R′.

[0046] As a preferred solution of the artificial intelligence-based Internet advertising marketing method of the present invention, wherein: based on the product recommendation list, personalized advertising content is generated, and personalized advertising content is pushed to the user according to the user's network usage scenario. The specific steps are as follows:

[0047] According to the user's current network usage and scenario, the ad display format is selected from the device type, access time period and visited website type, and personalized ad content is generated for each product based on the recommended product list;

[0048] Based on the user's historical interests and behavior data, analyze the product categories that the user is most interested in, and give priority to pushing advertisements related to the user's historical interests and behavior data to increase the click-through rate and conversion rate of advertisements.

[0049] As a preferred solution of the artificial intelligence-based Internet advertising marketing method of the present invention, in which: during the advertising push process, the interaction data between users and personalized advertising content is continuously collected to optimize the purchase intention dynamic prediction model. The specific steps are as follows:

[0050] During the ad push process, all interaction data between users and ads is monitored and collected in real time;

[0051] Use the collected user interaction data to update the user's behavior characteristics in real time, integrate the updated behavior characteristics with the historical behavior characteristics, and optimize the comprehensive feature vector F;

[0052] Based on the optimized comprehensive feature vector F′, the purchase intention dynamic prediction model is optimized.

[0053] In the second aspect, the present invention provides an Internet advertising marketing system based on artificial intelligence, including a comprehensive feature vector generation module, a purchase probability prediction module, a recommendation list generation module, an advertising push module and a model optimization module; the comprehensive feature vector generation module is used to collect user behavior data in real time through the Internet and preprocess it, and extract user behavior characteristics and fuse them into a comprehensive feature vector; the purchase probability prediction module is used to build a purchase intention dynamic prediction model, and predict the user's purchase probability of the product based on the comprehensive feature vector; the recommendation list generation module is used to generate a product recommendation list based on the user's purchase probability of the product; the advertising push module is used to generate personalized advertising content based on the product recommendation list, and push personalized advertising content to the user according to the user's network usage scenario; the model optimization module is used to continuously collect the interaction data between the user and the personalized advertising content during the advertising push process, and optimize the purchase intention dynamic prediction model.

[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 Internet advertising marketing 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 Internet advertising marketing method as described in the first aspect of the present invention.

[0056] The beneficial effects of the present invention are as follows: by collecting and preprocessing user behavior data in real time, extracting comprehensive feature vectors, and combining users' short-term and long-term behaviors, a dynamic prediction model for purchase intention is constructed to accurately predict users' purchase probability, thereby generating a personalized product recommendation list. At the same time, during the advertising push process, user interaction data is continuously collected and the model is dynamically optimized, thereby achieving accurate push of advertising content and improving the click-through rate and conversion rate of advertising. 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 artificial intelligence-based Internet advertising and marketing method in Example 1.

[0059] Figure 2This is a flow chart of the artificial intelligence-based Internet advertising and marketing 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 Internet advertising marketing method based on artificial intelligence, comprising the following steps:

[0064] S1: Collect user behavior data in real time through the Internet and pre-process it, extract user behavior features, and fuse them into a comprehensive feature vector.

[0065] The above information is obtained with the user's consent and is used for legal purposes.

[0066] S1.1: User behavior data includes user browsing history, search keywords, purchase history, click behavior, and social media interactions;

[0067] It should be noted that browsing history refers to the page information visited by users on a website or application, including records of browsed products, articles or other content.

[0068] Search keywords: refers to the query words that users enter into search engines, websites or applications. Search keywords reflect users' active needs and immediate interests.

[0069] Purchase record: refers to the purchase behavior completed by users on e-commerce platforms or other trading platforms.

[0070] Click behavior: refers to the specific content that users click on on a web page or in an advertisement, including ad links, product detail pages, etc.

[0071] Social media interaction: refers to users’ activities on social platforms, including likes, comments, shares, and follows.

[0072] S1.2: Preprocessing of user behavior data is as follows:

[0073] S1.2.1: Clean user behavior data by removing invalid data, correcting erroneous data, and filling missing values;

[0074] Furthermore, ensure that user behavior data is complete and accurate, such as deleting records that are missing key fields or have incorrect formats;

[0075] S1.2.2: Use hashing algorithms to identify duplicate user behavior records and delete multiple occurrences of the same data to avoid data redundancy and affect subsequent analysis;

[0076] S1.2.3: Filter outliers and noise data through denoising; for example, exclude users who stay for abnormally long periods of time;

[0077] The specific process is as follows: First, convert each user behavior record (such as browsing history, click behavior, etc.) into a unique hash value to ensure that the same behavior record generates the same hash value. Then, the system compares all generated hash values ​​to identify duplicate behavior records. For multiple occurrences of the same record, only the first occurrence is retained and subsequent duplicate data is deleted. This can effectively avoid data redundancy, ensure that each valid data appears only once, thereby improving data accuracy and avoiding the negative impact of duplicate data on subsequent behavior analysis and model training.

[0078] S1.2.4: Convert timestamps in different user behavior data into a unified format through standardization.

[0079] Furthermore, standardization was performed to ensure that all behavioral data could be correctly aligned and analyzed in chronological order.

[0080] S1.3: Group user behavior data by time and divide them into different categories according to behavior type;

[0081] For example, the user's behavior data in the past 7 days can be divided into categories such as browsing behavior (such as browsing a product page 5 times), search behavior (such as searching for "smartphone" 2 times), and click behavior (such as clicking on an ad 3 times) for analysis.

[0082] S1.4: For each type of behavior data, extract the user's behavior frequency within a specific time range;

[0083] For example, if a user browses a certain product page 25 times in a week, the frequency characteristic of this behavior is recorded as 25 times;

[0084] S1.5: Through page event tracking, extract the duration and interaction depth of each user behavior;

[0085] For example, the average length of time a user spends browsing a page, the length of time a video ad is viewed, etc. can all be used as behavioral deep features. These features can reflect the user's level of attention to a product and can be converted into feature vectors;

[0086] S1.6: Extract behavior sequence features by analyzing the order in which users' behaviors occur;

[0087] For example, a user's behavioral sequence of searching for a product and adding it to the shopping cart within a short period of time may indicate that the user has a high purchase intention;

[0088] S1.7: Based on nonlinear transformation and feature product, various user behavior features are processed through different linear transformations, and a comprehensive feature vector is generated through the product interaction between features. The expression is:

[0089]

[0090] Among them, F is the comprehensive feature vector, f i represents the behavior characteristics of the i-th category, W i represents the weight matrix for linear transformation of the i-th type of behavior feature vector, n represents the total number of user behavior features, σ is the nonlinear activation function, and b i It is the bias term when linear transformation is performed on the behavior characteristics of the i-th category.

[0091] It should be noted that based on the nonlinear transformation and feature interaction of multiple types of user behavior features, different types of user behaviors (such as browsing, clicking, searching, etc.) are integrated in the form of feature products. Each type of behavior feature f i By linearly transforming the weight matrix W i and the bias term b i After processing, the results of various features interact with each other through product to capture the complex relationship between behaviors. Finally, the feature interaction results are further processed through the nonlinear activation function σ to generate a comprehensive feature vector F, thereby improving the accuracy of modeling user behaviors and capturing potential nonlinear associations.

[0092] S2: Build a dynamic prediction model for purchase intention and predict the user's purchase probability of the product based on the comprehensive feature vector.

[0093] S2.1: Through hierarchical behavior modeling, the purchase intention dynamic prediction model is divided into a short-term behavior evaluation layer and a long-term behavior evaluation layer;

[0094] It should be noted that the benefit of dividing the purchase intention dynamic prediction model into a short-term behavior evaluation layer and a long-term behavior evaluation layer is that this hierarchical modeling can capture users' immediate needs and long-term preferences respectively. The short-term behavior evaluation layer focuses on users' recent dynamic behaviors, reflecting their current interests and purchase intentions, while the long-term behavior evaluation layer reveals users' stable preferences and consumption habits by analyzing their historical behaviors. By combining these two layers of models, the system can predict users' purchase intentions more comprehensively and accurately, thereby achieving personalized advertising push and recommendations.

[0095] S2.1.1: The short-term behavior evaluation layer adopts a similarity weighting mechanism based on a Gaussian kernel. By comparing the difference between the current feature and the neighborhood feature, the weight is calculated and the neighborhood feature is weighted and summed through the ReLU activation function to extract the features of the short-term behavior. The expression is:

[0096]

[0097] Among them, Z s represents the short-term behavior characteristics of the user, F is the comprehensive feature vector, and F x represents the xth neighborhood feature in the comprehensive feature vector F, μ is the standard deviation of the Gaussian kernel, b is the bias term, k is the total number of short-term behavior feature vectors, and x is the index of the neighborhood feature;

[0098] It should be noted that the advantage of extracting short-term features based on Gaussian kernel is that it can effectively smooth noise and capture local patterns in short-term behavior by weighting the similarity between the current feature and the neighboring features, giving higher weights to more similar features. At the same time, the smoothness of the Gaussian kernel ensures the progressive processing between features, which helps to retain key information when extracting short-term behavior features, reduce the risk of overfitting, and finally extract more representative short-term features through the ReLU activation function.

[0099] S2.1.2: The long-term behavior evaluation layer uses Memory Network to embed the user's long-term behavior into the memory unit, and selectively extracts and aggregates the long-term behavior features related to the user's purchase intention through the query mechanism. The expression is:

[0100]

[0101] Among them, Z l represents the long-term behavioral characteristics of users, h j represents the jth key vector, M j represents the jth key vector h j The corresponding memory unit is the j-th long-term behavior embedding of the user, W q Represents the weight matrix that the user uses to map the comprehensive feature vector F to the query vector;

[0102] It should be noted that the reason why the long-term behavior evaluation layer adopts Memory Network is that it can effectively store and manage users' long-term behavior data through the memory mechanism, and selectively extract the long-term behavior features most relevant to the current purchase intention through the query mechanism. Specifically, W in the expression q The user's comprehensive feature vector F is mapped to a query vector by comparing it with the key vector h j , and selectively aggregates the stored long-term behavior embedding M j This mechanism ensures that the model can flexibly extract the most relevant information from a large amount of long-term behaviors, thereby improving the ability to capture users' long-term preferences and behavior patterns, and combining short-term behaviors to make more accurate purchase intention predictions.

[0103] S2.1.3: Use the gating mechanism to dynamically adjust the purchase intention dynamic prediction model's attention to short-term and long-term behaviors. When the user shows an impulse to buy, the gating mechanism automatically increases the attention to short-term behaviors. When the user's behavior is stable, the gating mechanism pays more attention to the impact of long-term behaviors. The expression is:

[0104]

[0105] Among them, g represents the gating signal, σ is the nonlinear activation function, and ∈ represents the numerical offset, which is used to prevent instability caused by zero or very small values ​​in the calculation process;

[0106] It should be noted that the benefit of using the gating mechanism is that the system can dynamically adjust the degree of attention paid to short-term and long-term behaviors according to the user's current behavior. When the user shows an impulse to buy, the gating mechanism automatically increases the attention paid to short-term behavior through the expression g, and when the user's behavior tends to be stable, it focuses more on the impact of long-term behavior. Through this dynamic adjustment, the model can more flexibly balance the weights of short-term and long-term behaviors, thereby improving the accuracy of purchase intention prediction and preventing instability in calculations in extreme cases (avoided by ∈).

[0107] It should also be noted that the specific method for determining impulse buying is to monitor the user's short-term behavior data (such as browsing, clicking, searching, etc. in the past week). If the frequency of a certain type of behavior increases significantly in a short period of time, the user can be determined to be in an "impulsive buying" state. For example, a user views the detail page of the same product or related products multiple times in a week, or has continuous clicking behavior in a short period of time.

[0108] The specific method for determining whether user behavior is stable is to detect the user's long-term behavior data (such as purchase records and search history in the past month). If the user's behavior frequency is low in the short term and there is no obvious high-frequency click or search behavior, it can be determined that "behavior is stable." For example, a user does not frequently browse or click on products for a period of time, and mainly relies on historical purchase records and long-term interests.

[0109] S2.2: The short-term behavior evaluation layer, the long-term behavior evaluation layer and the gating mechanism are combined through the gating mechanism and nonlinear fusion to form a dynamic prediction model of purchase intention;

[0110] S2.3: Input the comprehensive feature vector into the purchase intention dynamic prediction model to predict the user's purchase probability of the product. The expression is:

[0111] y k =Sigmoid(W o ·(g·tanh(Z s,k )+(1-g)·tanh(Z l,k ))+b);

[0112] Among them, y k represents the user's purchase probability of the kth item, W o The weight matrix used to linearly transform the gated feature combination, Z s,k represents the short-term behavior characteristics associated with the kth commodity, Z l,k Represents the long-term behavioral characteristics of the k-th commodity.

[0113] It should be noted that this expression dynamically adjusts the degree of attention paid to short-term and long-term behaviors by inputting the comprehensive feature vector into the purchase intention dynamic prediction model to predict the user's purchase probability y for the kth product. k Specifically, the gating signal g controls the model’s response to the short-term feature Z s,k and long-term characteristics Z l,k The weighted combination of , respectively, uses the tanh activation function to process short-term and long-term behavior features. Then, the gated feature combination is linearly transformed (by the weight matrix W o and bias term b), and outputs the final purchase probability y through the Sigmoid function k This mechanism can dynamically adjust the weights of short-term and long-term features based on the user’s current behavior, thereby more accurately predicting the user’s purchase intention.

[0114] S3: Generate a product recommendation list based on the user's purchase probability of the product.

[0115] S3.1: Based on the user's purchase probability y for each product k, sort all products from high to low, select the first N products after sorting, and generate a preliminary product recommendation list R;

[0116] S3.2: Based on the user's historical behavior data, calculate the behavior similarity between the current user and other users. The expression is:

[0117]

[0118] Among them, u represents the behavior vector of the current user, v represents the behavior vector of other users, and u k represents the number of actions that the current user has taken on product k, v K represents the number of actions taken by other users on product i, and S(u,v) represents the similarity between the current user and other users;

[0119] S3.3: Based on the similarity between the current user and other users, select M users with the highest similarity to the current user as the similar user set U;

[0120] S3.4: Based on the similar user set U, calculate the collaborative filtering score of similar users for product k, expressed as:

[0121]

[0122] Among them, r CF,k represents the collaborative filtering score of similar users for product k, b u,k is the behavior value of user u towards product k;

[0123] Furthermore, the collaborative filtering score represents the collective purchase probability of other users;

[0124] S3.5: The user's purchase probability y for the kth product k and collaborative filtering score r CF,k Combined, a new sorting index is generated, the expression is:

[0125] S k =λ·y k +(1-λ)·r CF,k ;

[0126] Among them, r CF,k represents the collaborative filtering score of similar users for product k, and λ is the purchase probability y of the balanced user for the kth product k and collaborative filtering score r CF,k The weight coefficient, S k is a new ranking metric, CF stands for collaborative filtering;

[0127] It should be noted that this expression combines the user's purchase probability y k and collaborative filtering score rCF,k Generate a new sorting index S k , where λ controls the balance between the two. This can not only reflect the user's personalized purchase intention, but also improve the accuracy and personalization effect of recommendations with the help of collaborative filtering feedback from similar users.

[0128] It should also be noted that λ is set based on the relative importance of personalized recommendation and collaborative filtering. For example, if the user's historical behavior data is relatively rich and the personalized demand is strong, λ can be set to a larger value (such as 0.8) to focus more on the user's purchase probability y of the product. k If the user's behavior data is less or unclear, λ can be set to a smaller value (such as 0.3), and more reliance is placed on the collaborative filtering scores r of similar users. CF,k to generate a list of recommendations.

[0129] S3.6: According to the new ranking index S k The preliminary product recommendation list R is optimized to generate a final product recommendation list R′.

[0130] S4: Generate personalized advertising content based on the product recommendation list, and push personalized advertising content to users based on their network usage scenarios.

[0131] For example, suppose the preliminary product recommendation list R contains products A, B, and C, and their purchase probabilities y are calculated respectively. k and collaborative filtering score r CF,k , and use λ to balance the impact of the two. After calculation, the comprehensive score of product A is 0.66, product B is 0.56, and product C is 0.74. According to these new ranking indicators S k , the system re-sorts the products according to their scores from high to low and generates the final product recommendation list R′, namely: product C, product A, product B.

[0132] S4.1: Select the advertising display format based on the user's current network usage and scenario, device type, access time period, and visited website type, and generate personalized advertising content for each product based on the recommended product list;

[0133] For example, if a user uses a mobile device to browse a social media platform at night, the system will select an appropriate native advertising format based on the device type, and generate personalized advertising content for the user from a list of recommended products based on the characteristics of the access time period and the social platform, such as pushing smart home products suitable for evening use, to ensure the relevance and display effect of the advertisement.

[0134] Advertising display forms include banner ads, pop-up ads, video ads, native ads, and recommendation bar ads, etc., to provide the best display effect based on the user's browsing scenarios and preferences.

[0135] S4.2: Based on the user's historical interests and behavior data, analyze the product categories that the user is most interested in, and give priority to pushing advertisements related to the user's historical interests and behavior data to improve the click-through rate and conversion rate of advertisements.

[0136] For example, suppose a user's historical interest and behavior data shows that he or she frequently browses and purchases electronic products, especially smartphones and headphones. By analyzing this data, the system prioritizes push ads related to electronic products, such as promotional information for the latest smartphones or wireless headphones, thereby increasing the click-through rate and conversion rate of ads, because these products are highly relevant to the user's interests.

[0137] Behavioral data includes browsing history, purchase history, and click behavior, etc.

[0138] S5: During the advertising push process, continuously collect data on the interaction between users and personalized advertising content to optimize the dynamic prediction model of purchase intention.

[0139] S5.1: During the advertising push process, all interaction data between users and advertisements shall be monitored and collected in real time;

[0140] It should be noted that interaction data includes the number of times users click on ads, frequency of ad exposure, length of stay, scrolling behavior, whether they make a purchase, whether they close or skip ads, whether they share or comment on ad content, etc. By collecting these interaction data, the system can evaluate the advertising effect in real time and dynamically adjust the push strategy.

[0141] S5.2: Use the collected user interaction data to update the user's behavior characteristics in real time, integrate the updated behavior characteristics with the historical behavior characteristics, and optimize the comprehensive feature vector F;

[0142] It should be noted that during the advertising push process, the system uses the collected user interaction data (such as number of clicks, length of stay, purchase behavior, etc.) to update the user's behavioral characteristics in real time. First, the system integrates these new data with the user's current behavioral characteristics and dynamically adjusts the short-term behavioral characteristics. Subsequently, the system combines the updated behavioral characteristics with the user's historical behavioral characteristics to ensure that it reflects the user's immediate interests while retaining their long-term preferences. Through this integration, the comprehensive feature vector is optimized, making subsequent advertising push more accurate and personalized, thereby improving the effectiveness of advertising.

[0143] It should also be noted that the updated behavioral features are integrated with the historical behavioral features. Specifically, the gating mechanism gives higher weights to recent behavioral features. When the user shows an impulse to buy, the weight of short-term behavioral features increases, while the weight of long-term behavioral features decreases accordingly. Then, the system fuses the short-term and long-term behavioral features through weighted summation to generate a new behavioral feature vector. Finally, the integrated features are nonlinearly transformed through nonlinear activation functions (such as ReLU or Sigmoid) to generate an optimized comprehensive feature vector F for further purchase probability prediction and personalized advertising push.

[0144] S5.3: Optimize the purchase intention dynamic prediction model based on the optimized comprehensive feature vector F′.

[0145] It should be noted that based on the optimized comprehensive feature vector F′, the system inputs it into the purchase intention dynamic prediction model to further improve the accuracy of the prediction. By combining the user's real-time interaction data and historical behavior characteristics, the optimized F′ more comprehensively reflects the user's current interests and long-term preferences. The model uses the optimized feature vector to dynamically weight short-term and long-term behaviors, readjust the impact of various behaviors on purchase intentions, and thus more accurately predict the user's purchase probability for different products, ultimately achieving a personalized and efficient advertising push strategy.

[0146] This embodiment also provides an artificial intelligence-based Internet advertising marketing system, including: a comprehensive feature vector generation module, a purchase probability prediction module, a recommendation list generation module, an advertising push module and a model optimization module;

[0147] A comprehensive feature vector generation module is used to collect user behavior data in real time through the Internet and perform preprocessing, extract user behavior features, and fuse them into a comprehensive feature vector;

[0148] The purchase probability prediction module is used to build a dynamic prediction model for purchase intention and predict the user's purchase probability of the product based on the comprehensive feature vector;

[0149] A recommendation list generation module is used to generate a product recommendation list based on the user's purchase probability of the product;

[0150] The advertising push module is used to generate personalized advertising content based on the product recommendation list and push personalized advertising content to users according to their network usage scenarios;

[0151] The model optimization module is used to continuously collect data on the interaction between users and personalized advertising content during the advertising push process, and optimize the dynamic prediction model of purchase intention.

[0152] This embodiment also provides a computer device, which is suitable for the case of an artificial intelligence-based Internet advertising marketing 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 Internet advertising marketing method proposed in the above embodiment.

[0153] 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 via a system bus. 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 covering 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.

[0154] The present embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the artificial intelligence-based Internet advertising marketing method 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.

[0155] In summary, the present invention collects and preprocesses user behavior data in real time, extracts comprehensive feature vectors, and combines users' short-term and long-term behaviors to build a dynamic prediction model for purchase intention, accurately predicts users' purchase probability, and then generates a personalized product recommendation list. At the same time, during the advertising push process, users' interactive data is continuously collected and the model is dynamically optimized, thereby achieving accurate push of advertising content and improving the click-through rate and conversion rate of advertisements.

[0156] Example 2, referring to Table 1, is the second example of the present invention. In order to further verify the technical solution of the present invention, experimental simulation data of an Internet advertising marketing method based on artificial intelligence is provided.

[0157] In order to verify the effectiveness of the Internet advertising marketing method based on artificial intelligence, this embodiment selected 100 users and divided them into two groups: the first group used the traditional static advertising push method (traditional push group), and the second group used the intelligent advertising marketing method in the present invention (intelligent push group). The experiment lasted for two weeks, mainly investigating key indicators such as advertising click-through rate, advertising conversion rate, user stay time, purchase success rate, and advertising relevance score. The behavioral data of all users are collected in real time through the Internet and pre-processed to ensure the accuracy and completeness of the data.

[0158] In data preprocessing, we first clean the data such as user browsing behavior, search keywords, purchase records, click behavior and social media interaction to remove invalid and duplicate data. We use hashing algorithms to identify duplicate behaviors, delete redundant data, and standardize timestamps and other formats to ensure that all behavior data can be correctly aligned in chronological order for subsequent analysis.

[0159] Next, user behavior data was divided into short-term and long-term behaviors. Short-term behaviors include browsing and clicking behaviors in the past week, while long-term behaviors include purchase records and search history in the past month. By performing nonlinear transformations on behavior data in different time periods, comprehensive feature vectors of users were generated. For the smart push group, a purchase intention dynamic prediction model was constructed based on these comprehensive feature vectors. Through weighted analysis of short-term and long-term behaviors, the purchase probability of users for different products was predicted.

[0160] Advertisement content is generated based on the product recommendation list, combined with the user's device type, access time, and type of website visited to ensure personalized ad content. During ad push, we continuously collect data on user interaction with ads, and dynamically optimize the purchase intention prediction model to ensure the relevance and real-time nature of push content.

[0161] The details are shown in Table 1 below:

[0162] Table 1 Comparison of advertising push effects

[0163]

[0164]

[0165] Through the analysis of the data in the above table, it can be clearly seen that the intelligent advertising push method is significantly better than the traditional push method in terms of advertising click-through rate, advertising conversion rate, user stay time and purchase success rate. For example, the average advertising click-through rate of the intelligent push group reached 9.77%, which is much higher than the 2.87% of the traditional push group, and the purchase success rate has also increased from the traditional 1.13% to 4.87%. These data show that the intelligent push method significantly improves the effectiveness of advertising and user engagement by accurately predicting user purchase intentions and pushing personalized ads.

[0166] The artificial intelligence-based Internet advertising and marketing method of the present invention can not only effectively improve the click-through rate and conversion rate of advertisements, but also significantly increase the user's stay time on the advertising page and the purchase success rate, thereby achieving more accurate advertising push and higher user engagement. Moreover, by dynamically predicting user purchase intentions and continuously optimizing models, advertising content can be more personalized and meet user needs in real time and efficiently. This reflects the innovation and practical value of the present invention in improving advertising accuracy and user experience.

[0167] 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 Internet advertising marketing method based on artificial intelligence, characterized by: include, Collect user behavior data in real time through the Internet and pre-process it, extract user behavior features, and fuse them into a comprehensive feature vector; Build a dynamic prediction model for purchase intention and predict the user's purchase probability of the product based on the comprehensive feature vector; Generate a product recommendation list based on the user's purchase probability of the product; Generate personalized advertising content based on the product recommendation list, and push personalized advertising content to users based on their network usage scenarios; During the advertising push process, we continuously collect data on the interaction between users and personalized advertising content to optimize the dynamic prediction model of purchase intention.

2. The artificial intelligence-based Internet advertising marketing method according to claim 1, characterized in that: The user's behavior data includes the user's browsing history, search keywords, purchase history, click behavior and social media interactions; The preprocessing of user behavior data is specifically as follows: Clean user behavior data by removing invalid data, correcting erroneous data, and filling missing values; Use hashing algorithms to identify duplicate user behavior records and delete multiple occurrences of the same data to avoid data redundancy. Filter outliers and noise data through denoising; Through standardization, the timestamps in different user behavior data are converted into a unified format.

3. The artificial intelligence-based Internet advertising marketing method according to claim 2, characterized in that: Extract user behavior features. The specific steps are as follows: Group user behavior data by time and divide it into different categories according to behavior type; For each type of behavior data, extract the user's behavior frequency within a specific time range; Through page event tracking, the duration and interaction depth of each user behavior can be extracted; Extract behavior sequence features by analyzing the order in which users' behaviors occur; Based on nonlinear transformation and feature product, various user behavior features are processed through different linear transformations, and a comprehensive feature vector is generated through the product interaction between features. The expression is: Among them, F is the comprehensive feature vector, f i represents the behavior characteristics of the i-th category, W i represents the weight matrix for linear transformation of the i-th type of behavior feature vector, n represents the total number of user behavior features, σ is the nonlinear activation function, and b i It is the bias term when linear transformation is performed on the behavior characteristics of the i-th category.

4. The artificial intelligence-based Internet advertising marketing method according to claim 3, characterized in that: The purchase intention dynamic prediction model is constructed to predict the user's purchase probability of the product based on the comprehensive feature vector. The specific steps are as follows: Through hierarchical behavior modeling, the purchase intention dynamic prediction model is divided into a short-term behavior evaluation layer and a long-term behavior evaluation layer; The short-term behavior evaluation layer adopts a similarity weighting mechanism based on the Gaussian kernel. By comparing the difference between the current feature and the neighborhood feature, the weight is calculated and the neighborhood features are weighted summed. After the ReLU activation function, the characteristics of the short-term behavior are extracted. The expression is: Among them, Z s represents the short-term behavior characteristics of the user, F is the comprehensive feature vector, and F x represents the xth neighborhood feature in the comprehensive feature vector F, μ is the standard deviation of the Gaussian kernel, b is the bias term, k is the total number of short-term behavior feature vectors, and x is the index of the neighborhood feature; The long-term behavior evaluation layer uses Memory Network to embed the user's long-term behavior into the memory unit, and selectively extracts and aggregates the long-term behavior features related to the user's purchase intention through the query mechanism. The expression is: Among them, Z l represents the long-term behavioral characteristics of users, h j represents the jth key vector, M j represents the jth key vector h j The corresponding memory unit, W q Represents the weight matrix that the user uses to map the comprehensive feature vector F to the query vector; The gating mechanism is used to dynamically adjust the degree of attention paid by the purchase intention dynamic prediction model to short-term and long-term behaviors. When the user shows the impulse to buy, the gating mechanism automatically increases the attention paid to short-term behaviors. When the user behavior is stable, the gating mechanism pays more attention to the impact of long-term behaviors. The expression is: Among them, g represents the gating signal, σ is the nonlinear activation function, and ∈ represents the numerical offset; The short-term behavior evaluation layer, the long-term behavior evaluation layer and the gating mechanism are combined through the gating mechanism and nonlinear fusion to form a dynamic prediction model of purchase intention; The comprehensive feature vector is input into the purchase intention dynamic prediction model to predict the user's purchase probability of the product. The expression is: y k =Sigmoid(W o ·(g·tanh(Z s,k )+(1-g)·tanh(Z l,k ))+b o ); Among them, y k represents the probability of a user purchasing the kth item, W o The weight matrix used to linearly transform the gated feature combination, Z s,k represents the short-term behavior characteristics associated with the kth commodity, Z l,k Represents the long-term behavioral characteristics of the k-th commodity.

5. The artificial intelligence-based Internet advertising marketing method according to claim 4, characterized in that: The specific steps of generating a product recommendation list based on the user's purchase probability of the product are as follows: According to the user's purchase probability y for each product k , sort all products from high to low, select the first N products after sorting, and generate a preliminary product recommendation list R; Based on the user's historical behavior data, calculate the behavior similarity between the current user and other users; Based on the similarity between the current user and other users, select the M users with the highest similarity to the current user as the similar user set U; Based on the similar user set U, calculate the collaborative filtering score of similar users for product k; The probability y of the user purchasing the kth item k and collaborative filtering score r CF,k Combined, a new sorting index is generated, the expression is: S k =λ·y k +(1-λ)·r CF,k ; Among them, r CF,k represents the collaborative filtering score of similar users for product k, and λ is the purchase probability y of the balanced user for the kth product k and collaborative filtering score r CF,k The weight coefficient, S k is a new ranking metric, CF stands for collaborative filtering; According to the new sorting index S k The preliminary product recommendation list R is optimized to generate a final product recommendation list R′.

6. The artificial intelligence-based Internet advertising marketing method according to claim 5, characterized in that: The specific steps of generating personalized advertising content based on the product recommendation list and pushing personalized advertising content to users according to the user's network usage scenario are as follows: According to the user's current network usage and scenario, the ad display format is selected from the device type, access time period and visited website type, and personalized ad content is generated for each product based on the recommended product list; Based on the user's historical interests and behavior data, analyze the product categories that the user is most interested in, and give priority to pushing advertisements related to the user's historical interests and behavior data.

7. The artificial intelligence-based Internet advertising marketing method according to claim 6, characterized in that: During the advertising push process, the interaction data between users and personalized advertising content is continuously collected to optimize the purchase intention dynamic prediction model. The specific steps are as follows: During the ad push process, all interaction data between users and ads is monitored and collected in real time; Use the collected user interaction data to update the user's behavior characteristics in real time, integrate the updated behavior characteristics with the historical behavior characteristics, and optimize the comprehensive feature vector F; Based on the optimized comprehensive feature vector F′, the purchase intention dynamic prediction model is optimized.

8. An artificial intelligence-based Internet advertising and marketing system, based on the artificial intelligence-based Internet advertising and marketing method according to any one of claims 1 to 7, characterized in that: Including comprehensive feature vector generation module, purchase probability prediction module, recommendation list generation module, advertisement push module and model optimization module; A comprehensive feature vector generation module is used to collect user behavior data in real time through the Internet and perform preprocessing, extract user behavior features, and fuse them into a comprehensive feature vector; The purchase probability prediction module is used to build a dynamic prediction model for purchase intention and predict the user's purchase probability of the product based on the comprehensive feature vector; A recommendation list generation module is used to generate a product recommendation list based on the user's purchase probability of the product; The advertising push module is used to generate personalized advertising content based on the product recommendation list and push personalized advertising content to users according to their network usage scenarios; The model optimization module is used to continuously collect data on the interaction between users and personalized advertising content during the advertising push process, and optimize the dynamic prediction model of purchase intention.

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 Internet advertising marketing 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 Internet advertising marketing method described in any one of claims 1 to 7 are implemented.

Images