Marketing content recommendation method and system based on AI

Abstract

The invention discloses an AI-based marketing content recommendation method and system, and relates to the technical field of data processing, a candidate marketing content pool is determined through a similar user corresponding to a target user, and other possibly interested marketing contents except for a non-interested label are recommended for the target user; secondly, extracting features of target user data by adopting a first AI model to obtain user feature vectors, and effectively representing the interest direction and trend of a target user; then, a second AI model is adopted to extract features of marketing content data corresponding to candidate marketing content in the candidate marketing content pool, marketing feature vectors are obtained, and the marketing content is expressed more comprehensively; and finally, fusing the user feature vector and the marketing feature vector, identifying the fused feature vector by adopting a third AI model, executing marketing content recommendation of TOPN according to the conversion probability obtained by identification, and performing common recommendation through an interest trend and interests and hobbies of similar users, so that the recommendation accuracy can be effectively improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to an AI-based marketing content recommendation method and system. Background Technology

[0002] With the rapid development of internet technology and e-commerce, the amount of information has exploded, leaving users facing the dilemma of "information overload." In the field of digital marketing, how to accurately and efficiently recommend content that users may be interested in from a massive amount of marketing content such as products, advertisements, and articles has become crucial for improving user experience, increasing marketing conversion rates, and enhancing core competitiveness. Traditional methods of understanding marketing content remain at the level of keyword matching, failing to delve into the semantics, emotion, and creative style of the text. This results in recommended content that may superficially match user interest tags, but whose actual style, tone, or core information does not match user expectations, leading to inaccurate recommendations. Summary of the Invention

[0003] This application aims to provide an AI-based marketing content recommendation method and system, which addresses the problem that traditional methods only understand marketing content at the keyword matching level and cannot deeply understand the semantics, emotion, and creative style of the text content, resulting in inaccurate marketing content recommendations.

[0004] The first aspect of this application provides an AI-based marketing content recommendation method, including: For any target user, obtain the similar users corresponding to the target user, and determine the candidate marketing content pool based on the similar users; Collect target user data corresponding to the target user, and use a first AI model to extract features from the target user data to obtain a user feature vector corresponding to the target user; Obtain marketing content data corresponding to candidate marketing content in the candidate marketing content pool, and use a second AI model to extract features from the marketing content data to obtain a marketing feature vector corresponding to the marketing content data; The user feature vector corresponding to the target user is fused with the marketing feature vector corresponding to the marketing content data to obtain a fused feature vector; A third AI model is used to identify the fused feature vector, determine the conversion probability corresponding to the candidate marketing content, and perform TOPN marketing content recommendation based on the conversion probabilities corresponding to all candidate marketing content.

[0005] In one possible implementation, for any target user, similar users corresponding to the target user are obtained, and a candidate marketing content pool is determined based on the similar users, including: For any given target user, determine the city location corresponding to that target user; Based on the city location corresponding to the target user, multiple candidate users with the same city location are identified; Obtain historical user data corresponding to the target user and historical user data corresponding to the candidate users, and determine similar users corresponding to the target user from multiple candidate users.

[0006] In one possible implementation, historical user data corresponding to the target user and historical user data corresponding to candidate users are obtained, and similar users corresponding to the target user are determined from multiple candidate users, including: Based on the current time, determine the historical user data corresponding to the target user and the historical user data corresponding to the candidate user within the first historical time period; wherein, the first historical time period is not adjacent to the current time, that is, there is a second historical time period between the first historical time period and the current time, and the first historical time period, the second historical time period and the current time are continuous time. The first AI model is used to extract the historical user data corresponding to the target user to obtain the first historical feature vector; The first AI model is used to extract the historical user data corresponding to the candidate user to obtain the second historical feature vector; Obtain the cosine similarity between the first historical feature vector and the second historical feature vector, and arrange the candidate users in descending order of cosine similarity to obtain the arranged candidate users; Based on the ranked candidate users, a preset number of candidate users are selected sequentially as similar users to the target user.

[0007] In one possible implementation, target user data corresponding to the target user is collected, and features of the target user data are extracted using a first AI model to obtain a user feature vector corresponding to the target user, including: Collect target user data corresponding to the target user; the target user data includes basic user information and historical marketing content behavior sequences. The first AI model is used to extract features from the target user data to obtain the user feature vector corresponding to the target user.

[0008] In one possible implementation, marketing content data corresponding to candidate marketing content in the candidate marketing content pool is obtained, and features of the marketing content data are extracted using a second AI model to obtain a marketing feature vector corresponding to the marketing content data, including: Obtain marketing content data corresponding to candidate marketing content in the candidate marketing content pool; the marketing content data includes the price, category, brand, and text description of the marketing content; The text description is preprocessed to obtain the semantic feature vector corresponding to the text description; Based on the price, category, brand, and semantic feature vector of the marketing content, a second AI model is used to extract the features of the marketing content data to obtain the marketing feature vector corresponding to the marketing content data.

[0009] In one possible implementation, the first AI model and the second AI model are configured as convolutional neural networks.

[0010] In one possible implementation, the user feature vector corresponding to the target user is fused with the marketing feature vector corresponding to the marketing content data to obtain a fused feature vector, including: The user feature vector corresponding to the target user is concatenated with the marketing feature vector corresponding to the marketing content data to form a single vector, or the user feature vector corresponding to the target user is weighted and fused with the marketing feature vector corresponding to the marketing content data to obtain a fused feature vector.

[0011] In one possible implementation, a third AI model is used to identify the fused feature vector, determine the conversion probability corresponding to the candidate marketing content, and perform TOPN marketing content recommendation based on the conversion probabilities corresponding to all candidate marketing content, including: A third AI model is used to identify the fused feature vector, and the conversion probability corresponding to each candidate marketing content output by the third AI model is determined. The candidate marketing content is arranged in descending order of conversion probability, and the top N candidate marketing content is selected as the target marketing content. The target marketing content is then recommended to the target users to complete the TOPN marketing content recommendation.

[0012] In one possible implementation, after performing TOPN marketing content recommendations, the method further includes: After performing TOPN marketing content recommendations, obtain the target marketing content that the target user actually clicked; wherein, if the target user did not click on any recommended target marketing content, the target marketing content that the target user actually clicked is set to empty; For the target marketing content that the target user actually clicks on, the transition probability corresponding to the target marketing content that was actually clicked is set to 1, and the first sample data is constructed based on the fused feature vector and the transition probability corresponding to the target marketing content that was actually clicked on; Set the transition probability of the target marketing content that was not actually clicked to 0, and construct the second sample data based on the fused feature vector and transition probability of the target marketing content that was not actually clicked. After each data training cycle, the third AI model is updated based on the first sample data and the second sample data to obtain the updated third AI model. In the subsequent recommendation process, the updated third AI model is used to calculate the conversion probability.

[0013] The second aspect of this application provides an AI-based marketing content recommendation system, including: a candidate marketing content determination module, a first feature extraction module, a second feature extraction module, a feature fusion module, and a marketing content recommendation module; The candidate marketing content determination module is used to obtain similar users corresponding to any target user, and determine a candidate marketing content pool based on the similar users. The first feature extraction module is used to collect target user data corresponding to the target user, and use a first AI model to extract features from the target user data to obtain a user feature vector corresponding to the target user. The second feature extraction module is used to obtain marketing content data corresponding to candidate marketing content in the candidate marketing content pool, and to use the second AI model to extract features of the marketing content data to obtain the marketing feature vector corresponding to the marketing content data. The feature fusion module is used to fuse the user feature vector corresponding to the target user with the marketing feature vector corresponding to the marketing content data to obtain a fused feature vector; The marketing content recommendation module is used to identify the fused feature vector using a third AI model, determine the conversion probability corresponding to the candidate marketing content, and perform TOPN marketing content recommendation based on the conversion probabilities corresponding to all candidate marketing content.

[0014] Beneficial effects: This application provides an AI-based marketing content recommendation method and system. It determines a candidate marketing content pool by identifying similar users of the target user, and can recommend marketing content that the target user might be interested in, beyond those not tagged with interest tags. Then, a first AI model extracts features from the target user's data to obtain a user feature vector, which effectively represents the target user's interest direction and trends. Next, a second AI model extracts features from the marketing content data corresponding to the candidate marketing content in the pool, obtaining a marketing feature vector, which more comprehensively expresses the marketing content. The user feature vector and the marketing feature vector are then fused to obtain a fused feature vector. Finally, a third AI model identifies the fused feature vector to obtain the conversion probability corresponding to the candidate marketing content, and performs TOPN marketing content recommendation based on the conversion probability. By combining interest trends and the interests of similar users, it avoids unchanging recommendations and effectively improves recommendation accuracy. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of an AI-based marketing content recommendation method proposed in one embodiment of this application; Figure 2 This is a structural diagram of an AI-based marketing content recommendation system proposed in one embodiment of this application; Figure labeling: 201-Candidate marketing content determination module, 202-First feature extraction module, 203-Second feature extraction module, 204-Feature fusion module, 205-Marketing content recommendation module. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] like Figure 1 As shown in the embodiment of this application, a marketing content recommendation method based on AI (Artificial Intelligence) is provided, including: S101. For any target user, obtain the similar users corresponding to the target user, and determine the candidate marketing content pool based on the similar users.

[0019] In one possible implementation, for any target user, similar users corresponding to the target user are obtained, and a candidate marketing content pool is determined based on the similar users, including: For any given target user, determine the city location corresponding to that target user.

[0020] Based on the city location corresponding to the target user, multiple candidate users with the same city location are identified.

[0021] Assuming the number of candidate users is set to 1000, then 1000 candidate users with the same city location can be randomly selected.

[0022] Obtain historical user data corresponding to the target user and historical user data corresponding to the candidate users, and determine similar users corresponding to the target user from multiple candidate users.

[0023] In one possible implementation, historical user data corresponding to the target user and historical user data corresponding to candidate users are obtained, and similar users corresponding to the target user are determined from multiple candidate users, including: Based on the current time, determine the historical user data corresponding to the target user and the historical user data corresponding to the candidate user within the first historical time period; wherein, the first historical time period is not adjacent to the current time, that is, there is a second historical time period between the first historical time period and the current time, and the first historical time period, the second historical time period and the current time are continuous time.

[0024] The first AI model is used to extract the historical user data corresponding to the target user to obtain the first historical feature vector.

[0025] The first AI model is used to extract the historical user data corresponding to the candidate user to obtain the second historical feature vector.

[0026] Obtain the cosine similarity between the first historical feature vector and the second historical feature vector, and arrange the candidate users in descending order of cosine similarity to obtain the arranged candidate users.

[0027] Based on the ranked candidate users, a preset number of candidate users are selected sequentially as similar users to the target user.

[0028] Determining a candidate marketing content pool based on similar users may include: identifying G marketing content items recently clicked by each similar user, and constructing the identified marketing content into a candidate marketing content pool, where G is a pre-set collection quantity.

[0029] By constructing the candidate marketing content pool as described above, marketing content that may be of interest to target users can be recommended, without being limited to marketing content under preset interest tags.

[0030] In this application embodiment, the marketing content may refer to a certain product, and the product has a price, category, brand and text description.

[0031] S102. Collect target user data corresponding to the target user, and use the first AI model to extract features from the target user data to obtain the user feature vector corresponding to the target user.

[0032] In one possible implementation, target user data corresponding to the target user is collected, and features of the target user data are extracted using a first AI model to obtain a user feature vector corresponding to the target user, including: Collect target user data corresponding to the target user; the target user data includes basic user information and historical marketing content behavior sequences.

[0033] User basic information can include the user's age, gender, province, and interest tags. The number of interest tags is set to a fixed number (e.g., 3). If the number of interest tags exceeds the preset fixed number, the number of marketing content accessed by the target user under each interest tag can be determined, and the user's basic information is constructed using the top three interest tags with the most marketing content accessed. If the number of interest tags is less than the fixed number, a string of all zeros with the same length as the interest tag data is used instead. A string of all zeros is a string where all elements are zero.

[0034] The historical marketing content behavior sequence can include the category, price, brand, and text description of the marketing content clicked by the target user within a historical time period. The number of marketing content collected is a preset fixed number M, such as M can be set to 10 or 20. If the number of marketing content clicked by the user is less than M, zero-padding can be used to make the number of data corresponding to the marketing content in the historical marketing content behavior sequence the same.

[0035] To achieve effective data identification, each data type can be set to the same length but with different values. For example, age can be set to a three-digit number (such as 009 years old, 018 years old, or 100 years old to represent different ages); gender can be set to two values, 1 for male and 0 for female; province can be represented by a unique code, and each province's code has the same length; interest tags can also be represented by a unique code, and each interest tag's code has the same length.

[0036] The category corresponding to the marketing content can be a smaller category than the interest tag. For example, if the interest tag is "mobile phone," then the category corresponding to the marketing content can be a specific type of mobile phone (such as foldable screen phone, curved screen phone, flat screen phone, etc.). Each category is represented using a unique code, and all categories have the same code length. The price can be represented by the number of digits corresponding to the highest price among all marketing content items. For example, if the highest price is 10,000 yuan, then the highest price has 5 digits, and can be represented as 00001, 00011, 00166, 06166, etc., by padding lower prices with zeros to make all prices have the same number of digits as the highest price. All brands are represented using a fixed-length unique code, such as by assigning different unique codes to different brands in advance. The text descriptions in the historical marketing content behavior sequence can be extracted into fixed-length semantic feature vectors using existing semantic feature extraction techniques.

[0037] In this way, the target user data can be expressed as data of fixed length, and then the features of the target user data can be extracted by the first AI model to obtain the user feature vector corresponding to the target user.

[0038] In one possible implementation, the first AI model is configured as a convolutional neural network. Since a convolutional neural network is used, the target user data can be represented in the form of a vector or in the form of a data matrix.

[0039] Before using the first AI model, it can be trained to improve the accuracy of feature extraction. For example, multiple different historical user feature vectors can be collected in advance. Then, the K-means clustering algorithm can be used to cluster the historical user feature vectors, and a fixed label can be assigned to each class of historical user feature vectors. Using the historical user feature vectors as input data and the fixed labels corresponding to the historical user feature vectors as the expected output data, the first AI model can be trained to obtain the trained first AI model. Then, in the subsequent feature extraction process, the trained first AI model can be used to identify the target user data, and the features output by the fully connected layer of the trained first AI model can be used as the user feature vector corresponding to the target user.

[0040] S103. Obtain marketing content data corresponding to candidate marketing content in the candidate marketing content pool, and use the second AI model to extract features of the marketing content data to obtain the marketing feature vector corresponding to the marketing content data.

[0041] In one possible implementation, marketing content data corresponding to candidate marketing content in the candidate marketing content pool is obtained, and features of the marketing content data are extracted using a second AI model to obtain a marketing feature vector corresponding to the marketing content data, including: Obtain marketing content data corresponding to candidate marketing content in the candidate marketing content pool; the marketing content data includes the price, category, brand, and text description of the marketing content.

[0042] The text description is preprocessed to obtain the semantic feature vector corresponding to the text description. That is, existing semantic feature extraction techniques can be used to extract the semantic feature vector of fixed length.

[0043] Based on the price, category, brand, and semantic feature vector of the marketing content, a second AI model is used to extract the features of the marketing content data to obtain the marketing feature vector corresponding to the marketing content data. That is, the price, category, brand, and semantic feature vector of the marketing content are combined into a data vector or matrix and input into the second AI model. The features output by the fully connected layer of the second AI model are used as the marketing feature vector corresponding to the marketing content data.

[0044] The processing of marketing content data can be similar to the processing of historical marketing content behavior sequences, so that the marketing content data can be expressed as data of fixed length, so that the second AI model can extract the features of the marketing content data and obtain the marketing feature vector corresponding to the marketing content data.

[0045] In one possible implementation, the second AI model is configured as a convolutional neural network. This second AI model also requires pre-training before use, and the training method is similar to that of the first AI model. However, the second AI model uses marketing content data that has been converted into semantic feature vectors for clustering. Similarly, the trained second AI model can be used to... Optionally, this application provides a method for unifying the length of semantic feature vectors. When the lengths of semantic feature vectors corresponding to different marketing content data are different, zeros can be padded in front of the semantic feature vectors to make all semantic feature vectors have the same length.

[0046] S104. The user feature vector corresponding to the target user is fused with the marketing feature vector corresponding to the marketing content data to obtain a fused feature vector.

[0047] In one possible implementation, the user feature vector corresponding to the target user is fused with the marketing feature vector corresponding to the marketing content data to obtain a fused feature vector, including: The user feature vector corresponding to the target user is concatenated with the marketing feature vector corresponding to the marketing content data to form a single vector, or the user feature vector corresponding to the target user is weighted and fused with the marketing feature vector corresponding to the marketing content data to obtain a fused feature vector.

[0048] S105. Use a third AI model to identify the fused feature vector, determine the conversion probability corresponding to the candidate marketing content, and perform TOPN marketing content recommendation based on the conversion probabilities corresponding to all candidate marketing content.

[0049] In one possible implementation, a third AI model is used to identify the fused feature vector, determine the conversion probability corresponding to the candidate marketing content, and perform TOPN marketing content recommendation based on the conversion probabilities corresponding to all candidate marketing content, including: A third AI model is used to identify the fused feature vector, and the conversion probability corresponding to each candidate marketing content output by the third AI model is determined.

[0050] The candidate marketing content is arranged in descending order of conversion probability, and the top N candidate marketing content is selected as the target marketing content. The target marketing content is then recommended to the target users to complete the TOPN marketing content recommendation.

[0051] Optionally, the third AI model can be set as an LSTM (Long Short-Term Memory) model or a BP (Back Propagation) neural network model to achieve feature vector recognition. In this embodiment, an LSTM model is preferably used as the third AI model, thus combining the first and second AI models to achieve spatial feature extraction and recognition.

[0052] For some new users, it is not possible to effectively train a third AI model at the initial stage of use. Therefore, a user with the highest similarity in the same city can be matched to the new user, and the third AI model corresponding to the user with the highest similarity can be used as the third AI model corresponding to the new user, thereby solving the cold start problem to some extent.

[0053] Optionally, to reduce resource consumption, only the model parameters of the third AI model corresponding to each user can be saved. Then, when a user uses the third AI model, a new third AI model can be generated and the user's corresponding model parameters can be applied, thereby reducing resource utilization.

[0054] This application provides an AI-based marketing content recommendation method. It determines a candidate marketing content pool by identifying similar users of the target user, and recommends marketing content that the target user might be interested in beyond interest tags. Then, a first AI model extracts features from the target user's data to obtain a user feature vector, which effectively represents the target user's interest direction and trends. Next, a second AI model extracts features from the marketing content data corresponding to the candidate marketing content in the pool, obtaining a marketing feature vector, which more comprehensively expresses the marketing content. The user feature vector and the marketing feature vector are then fused to obtain a fused feature vector. Finally, a third AI model identifies the fused feature vector to obtain the conversion probability corresponding to the candidate marketing content, and performs TOPN marketing content recommendation based on the conversion probability. By combining interest trends and the interests of similar users, the method avoids unchanging recommendations and effectively improves recommendation accuracy.

[0055] In one possible implementation, after performing TOPN marketing content recommendations, the method further includes: After performing TOPN marketing content recommendations, the target marketing content actually clicked by the target user is obtained. If the target user does not click on any recommended target marketing content, the target marketing content actually clicked by the target user is set to empty. For example, if the user exits the application interface directly, the target marketing content actually clicked by the target user can be set to empty.

[0056] For the target marketing content that the target user actually clicks on, the transition probability corresponding to the target marketing content that was actually clicked is set to 1, and the first sample data is constructed based on the fused feature vector and transition probability corresponding to the target marketing content that was actually clicked on.

[0057] The transition probability corresponding to the target marketing content that was not actually clicked is set to 0, and a second sample data is constructed based on the fused feature vector and transition probability corresponding to the target marketing content that was not actually clicked.

[0058] After each data training cycle, the third AI model is updated based on the first and second sample data to obtain the updated third AI model. This updated third AI model is then used to calculate the conversion probability in subsequent recommendation processes. The data training cycle is the training period set by the administrator for the third AI model.

[0059] This application's embodiments update the third AI model through the actual operations of target users. This allows the third AI model to continuously learn user behavior, quickly adapt to market changes and shifts in user interests, ensuring the long-term effectiveness and continuous improvement of recommendation results. It can significantly improve the click-through rate, conversion rate, and user satisfaction of marketing content, reduce marketing costs, and create higher business value for enterprises.

[0060] Optionally, to ensure the accuracy of transition probability identification, embodiments of this application provide a method for updating a third AI model, which may include: For the model parameters of the third AI model (such as the connection weights between network layers), the model parameters are randomly initialized between the upper and lower bounds, and the randomly initialized model parameters are encoded into vectors to obtain a population containing multiple individuals. Based on the first sample data and the second sample data, obtain the fitness of each individual; After applying the model parameters from the individual to the third AI model, the fused feature vectors from the first and second sample data are used as inputs, and the corresponding transition probabilities are used as the expected outputs. The cross-entropy loss function value corresponding to the individual is obtained, and the cross-entropy loss function value is added to a preset constant (such as 1 or 0.001) and the reciprocal is taken to obtain the fitness of the individual.

[0061] The individual with the highest fitness is selected as the optimal individual, the individual with the greatest average distance in the population is selected as the first target individual, and the remaining individuals are selected as the second target individuals. The method for obtaining the population average distance is as follows: for any individual, obtain the Euclidean distance between that individual and all other individuals, and then calculate the average of the Euclidean distances between that individual and all other individuals to obtain the population average distance corresponding to that individual.

[0062] The probability of obtaining a decision is: ; In the formula, For decision probability, It is the hyperbolic tangent function. The first random number between (0,1) Let be the fitness of the i-th second target individual, i = 1, 2, ..., NP-2, where NP represents the total number of second target individuals. This represents the fitness of the optimal individual; A decision factor is randomly generated between (0,1), and the relationship between the decision factor and the decision probability is obtained. If the decision factor is less than the decision probability, then the second target individual is updated as follows:

[0063] In the formula, For the i-th second target individual during the t-th update process, For the updated second target individual , For adaptive search balance coefficients, The second random number between (0,1) The third random number between (-1, 1) The fourth random number between (-1, 1) For the primary target individual, The average position of all individuals during the t-th update process is the average value of the model parameters of all individuals in the same dimension of the average position individual. It is an exponential function with the natural constant e as its base. This is the preset maximum number of updates; If the decision factor is greater than or equal to the decision probability, then the second target individual is updated as follows: ; In the formula, The fifth random number between (-1, 1) Let cos be the sixth random number between (0,1), and let cos be the cosine function. Pi The optimal individual.

[0064] The optimal individual is updated as follows:

[0065]

[0066] In the formula, This is the optimal individual after the update. For the j-th individual, As random individuals, and Each element in each dimension is governed by obeying The normal distribution is randomly generated. Denotes the normal distribution factor. This represents the mean value of all individuals, meaning that each element in a dimension is the mean of all individuals in the same dimension, and T represents the transpose. Optional, can be used for The fitness level is judged, if The fitness is greater than The fitness level is then determined by... As the optimal individual after the update, otherwise... As the optimal individual after the update.

[0067] The first target individual is updated as follows:

[0068] In the formula, For the first target individual after the update, b represents the spiral shape constant and is set to 0.4; The variable spiral control factor is randomized between (-1, 1). The seventh random number between (0, 1); The updated best individual, the first target individual, and the second target individual are then merged back into a single population. Determine whether the current update count is greater than or equal to the preset maximum update count. If so, based on the re-merged population, re-obtain the optimal individual and use the model parameters in the optimal individual as the target parameters of the third AI model to obtain the updated third AI model. Otherwise, return to the step of obtaining the fitness of each individual.

[0069] Optionally, after each individual is updated, boundary violation handling can be performed to ensure that the model parameters in the individual are within their corresponding upper and lower limits. For example, if a model parameter in a certain dimension of an individual exceeds the limit, the model parameter in that dimension can be modified to the value before the update; it can also be randomly generated between its upper and lower limits; or the out-of-limit model parameter can be set to the nearest limit value.

[0070] The third AI model update method provided in this application embodiment can perform global and local balanced updates on the second target individuals, which account for a large proportion of the total number. It also has more refined search in the later stages of the algorithm, ensuring update speed and effect. The update of the optimal individual and the first target individual can ensure that the algorithm has a strong global search capability and can always avoid getting trapped in local optima. Compared with the prior art, it has a better update effect, which enables the updated third AI model to learn user behavior more accurately, quickly adapt to market changes and user interest shifts, and ensure the long-term effectiveness and continuous improvement of recommendation effect.

[0071] like Figure 2As shown in the figure, this application provides an AI-based marketing content recommendation system, including: a candidate marketing content determination module 201, a first feature extraction module 202, a second feature extraction module 203, a feature fusion module 204, and a marketing content recommendation module 205. The candidate marketing content determination module 201 is used to, for any target user, obtain similar users corresponding to the target user, and determine a candidate marketing content pool based on the similar users.

[0072] The first feature extraction module 202 is used to collect target user data corresponding to the target user, and use a first AI model to extract features from the target user data to obtain a user feature vector corresponding to the target user.

[0073] The second feature extraction module 203 is used to obtain marketing content data corresponding to candidate marketing content in the candidate marketing content pool, and to use the second AI model to extract features of the marketing content data to obtain the marketing feature vector corresponding to the marketing content data.

[0074] The feature fusion module 204 is used to fuse the user feature vector corresponding to the target user with the marketing feature vector corresponding to the marketing content data to obtain a fused feature vector.

[0075] The marketing content recommendation module 205 is used to identify the fused feature vector using a third AI model, determine the conversion probability corresponding to the candidate marketing content, and perform TOPN marketing content recommendation based on the conversion probabilities corresponding to all candidate marketing content.

[0076] The AI-based marketing content recommendation system provided in this application embodiment can execute the above-described method and technical solution. Its principle and beneficial effects are similar, and will not be repeated here.

[0077] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0078] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, electronic devices, and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart... Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0079] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0080] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0081] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0082] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0083] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. An AI-based marketing content recommendation method, characterized in that, include: For any target user, obtain the similar users corresponding to the target user, and determine the candidate marketing content pool based on the similar users; Collect target user data corresponding to the target user, and use a first AI model to extract features from the target user data to obtain a user feature vector corresponding to the target user; Obtain marketing content data corresponding to candidate marketing content in the candidate marketing content pool, and use a second AI model to extract features from the marketing content data to obtain a marketing feature vector corresponding to the marketing content data; The user feature vector corresponding to the target user is fused with the marketing feature vector corresponding to the marketing content data to obtain a fused feature vector; A third AI model is used to identify the fused feature vector, determine the conversion probability corresponding to the candidate marketing content, and perform TOPN marketing content recommendation based on the conversion probabilities corresponding to all candidate marketing content.

2. The AI-based marketing content recommendation method according to claim 1, characterized in that, For any given target user, obtain similar users corresponding to the target user, and determine a candidate marketing content pool based on the similar users, including: For any given target user, determine the city location corresponding to that target user; Based on the city location corresponding to the target user, identify multiple candidate users with the same city location; Obtain historical user data corresponding to the target user and historical user data corresponding to the candidate users, and determine similar users corresponding to the target user from multiple candidate users.

3. The AI-based marketing content recommendation method according to claim 2, characterized in that, Obtain historical user data corresponding to the target user and historical user data corresponding to candidate users, and determine similar users corresponding to the target user from multiple candidate users, including: Based on the current time, determine the historical user data corresponding to the target user and the historical user data corresponding to the candidate user within the first historical time period; wherein, the first historical time period is not adjacent to the current time, that is, there is a second historical time period between the first historical time period and the current time, and the first historical time period, the second historical time period and the current time are continuous time. The first AI model is used to extract the historical user data corresponding to the target user to obtain the first historical feature vector; The first AI model is used to extract the historical user data corresponding to the candidate user to obtain the second historical feature vector; Obtain the cosine similarity between the first historical feature vector and the second historical feature vector, and arrange the candidate users in descending order of cosine similarity to obtain the arranged candidate users; Based on the ranked candidate users, a preset number of candidate users are selected sequentially as similar users to the target user.

4. The AI-based marketing content recommendation method according to claim 1, characterized in that, Collect target user data corresponding to the target user, and use a first AI model to extract features from the target user data to obtain a user feature vector corresponding to the target user, including: Collect target user data corresponding to the target user; the target user data includes basic user information and historical marketing content behavior sequences. The first AI model is used to extract features from the target user data to obtain the user feature vector corresponding to the target user.

5. The AI-based marketing content recommendation method according to claim 4, characterized in that, Obtain marketing content data corresponding to candidate marketing content in the candidate marketing content pool, and use a second AI model to extract features from the marketing content data to obtain a marketing feature vector corresponding to the marketing content data, including: Obtain marketing content data corresponding to candidate marketing content in the candidate marketing content pool; the marketing content data includes the price, category, brand, and text description of the marketing content; The text description is preprocessed to obtain the semantic feature vector corresponding to the text description; Based on the price, category, brand, and semantic feature vector of the marketing content, a second AI model is used to extract the features of the marketing content data to obtain the marketing feature vector corresponding to the marketing content data.

6. The AI-based marketing content recommendation method according to claim 5, characterized in that, The first AI model and the second AI model are configured as convolutional neural networks.

7. The AI-based marketing content recommendation method according to claim 1, characterized in that, The user feature vector corresponding to the target user is fused with the marketing feature vector corresponding to the marketing content data to obtain a fused feature vector, including: The user feature vector corresponding to the target user is concatenated with the marketing feature vector corresponding to the marketing content data to form a single vector, or the user feature vector corresponding to the target user is weighted and fused with the marketing feature vector corresponding to the marketing content data to obtain a fused feature vector.

8. The AI-based marketing content recommendation method according to claim 1, characterized in that, A third AI model is used to identify the fused feature vector, determine the conversion probability corresponding to the candidate marketing content, and perform TOPN marketing content recommendation based on the conversion probabilities of all candidate marketing content, including: A third AI model is used to identify the fused feature vector, and the conversion probability corresponding to each candidate marketing content output by the third AI model is determined. The candidate marketing content is arranged in descending order of conversion probability, and the top N candidate marketing content is selected as the target marketing content. The target marketing content is then recommended to the target users to complete the TOPN marketing content recommendation.

9. The AI-based marketing content recommendation method according to claim 8, characterized in that, After performing TOPN marketing content recommendations, the process also includes: After performing TOPN marketing content recommendations, obtain the target marketing content that the target user actually clicked; wherein, if the target user did not click on any recommended target marketing content, the target marketing content that the target user actually clicked is set to empty; For the target marketing content that the target user actually clicks on, the transition probability corresponding to the target marketing content that was actually clicked is set to 1, and the first sample data is constructed based on the fused feature vector and the transition probability corresponding to the target marketing content that was actually clicked on; Set the transition probability of the target marketing content that was not actually clicked to 0, and construct the second sample data based on the fused feature vector and transition probability of the target marketing content that was not actually clicked. After each data training cycle, the third AI model is updated based on the first sample data and the second sample data to obtain the updated third AI model. In the subsequent recommendation process, the updated third AI model is used to calculate the conversion probability.

10. An AI-based marketing content recommendation system, characterized in that, include: The module includes a candidate marketing content determination module, a first feature extraction module, a second feature extraction module, a feature fusion module, and a marketing content recommendation module. The candidate marketing content determination module is used to obtain similar users corresponding to any target user, and determine a candidate marketing content pool based on the similar users. The first feature extraction module is used to collect target user data corresponding to the target user, and use a first AI model to extract features from the target user data to obtain a user feature vector corresponding to the target user. The second feature extraction module is used to obtain marketing content data corresponding to candidate marketing content in the candidate marketing content pool, and to use the second AI model to extract features of the marketing content data to obtain the marketing feature vector corresponding to the marketing content data. The feature fusion module is used to fuse the user feature vector corresponding to the target user with the marketing feature vector corresponding to the marketing content data to obtain a fused feature vector; The marketing content recommendation module is used to identify the fused feature vector using a third AI model, determine the conversion probability corresponding to the candidate marketing content, and perform TOPN marketing content recommendation based on the conversion probabilities corresponding to all candidate marketing content.

Images