Target user determination method, apparatus, storage medium, and electronic device

Abstract

The present disclosure provides a target user determination method and device, a storage medium and an electronic device; and relates to the technical field of artificial intelligence. The method comprises: acquiring multi-dimensional feature data of a plurality of users, and generating a user feature vector and an advertisement feature vector according to the multi-dimensional feature data; inputting the user feature vector and the advertisement feature vector into a user representation model to generate a first seed user vector and a first candidate user vector; inputting the first seed user vector and the first candidate user vector into a similarity estimation model to generate a second seed user vector and a second candidate user vector, and calculating the similarity between the second seed user vector and the second candidate user vector; and determining a target user in the candidate users according to the similarity. The present disclosure can accurately and efficiently determine the similar groups of the seed users, and further improve the accuracy of advertisement delivery.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and more specifically, to a method for determining a target user, a device for determining a target user, a computer-readable storage medium, and an electronic device. Background Technology

[0002] With the development of internet technology, advertising has become an important part of the online world. In advertising scenarios, there is often a need to expand the target audience.

[0003] For example, offline advertising relies heavily on manual audience segmentation based on users' historical behavior and advertisers' needs, which is inefficient. Furthermore, manual audience segmentation is highly dependent on the historical behavior of users associated with specific items, easily leading to a "Matthew effect" in ad placement. For instance, niche but high-quality long-tail content may be considered to have a lower priority, thus failing to receive effective advertising.

[0004] Therefore, in order to ensure the effective delivery of advertisements, it is essential to provide a precise and efficient method for identifying target users.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this disclosure is to provide a method, device, computer-readable storage medium, and electronic device for determining target users, thereby overcoming to some extent the problem that the efficiency and accuracy of expanding similar user groups are low due to the limitations and defects of related technologies.

[0007] According to a first aspect of this disclosure, a method for determining a target user is provided, comprising:

[0008] Acquire multidimensional feature data of multiple users, and generate user feature vectors and advertising feature vectors based on the multidimensional feature data, wherein the multiple users include seed users and candidate users;

[0009] The user feature vector and the advertisement feature vector are input into the user representation model to generate a user representation vector, which includes a first seed user vector and a first candidate user vector.

[0010] The first seed user vector and the first candidate user vector are input into the similarity prediction model to generate the second seed user vector and the second candidate user vector, and the similarity between the second seed user vector and the second candidate user vector is calculated.

[0011] The target user is determined from the candidate users based on the similarity.

[0012] In one exemplary embodiment of this disclosure, the seed users include one of the seed users who have previously received advertisements and the seed users who are to receive advertisements; before obtaining the multidimensional feature data of multiple users, the process includes:

[0013] Obtain the first ad feature vector of the ad to be delivered and the second ad feature vector of the historical ads;

[0014] Calculate the similarity between the first ad feature vector and the second ad feature vector, and determine the target ad that is similar to the ad to be delivered in the historical ad delivery based on the similarity;

[0015] The seed users for the ads to be delivered are constructed based on the seed users of the target ads.

[0016] In one exemplary embodiment of this disclosure, the multidimensional feature data includes user feature data and advertising feature data; the step of generating user feature vectors and advertising feature vectors based on the multidimensional feature data includes:

[0017] The user feature data is vectorized to generate the user feature vector;

[0018] The advertising feature data is vectorized to generate the advertising feature vector.

[0019] In one exemplary embodiment of this disclosure, the user feature data includes at least one of discrete feature data and continuous feature data; the step of vectorizing the user feature data to generate the user feature vector includes:

[0020] Statistical analysis is performed on each of the discrete feature data to obtain multiple first feature data;

[0021] Encode each of the first feature data to generate multiple first feature vectors;

[0022] Normalize each of the continuous feature data to generate multiple second feature vectors;

[0023] The user feature vector is generated by concatenating the plurality of first feature vectors and the plurality of second feature vectors.

[0024] In one exemplary embodiment of this disclosure, the user representation model includes at least a first fully connected layer and a first attention layer; the step of inputting the user feature vector and the advertising feature vector into the user representation model to generate a user representation vector includes:

[0025] The first fully connected layer is used to perform feature dimensionality reduction on the user feature vector and the advertisement feature vector to obtain the user intermediate vector and the advertisement intermediate vector.

[0026] Based on the user intermediate vector and the advertisement intermediate vector, the first seed user vector and the first candidate user vector are generated through the first attention layer.

[0027] In one exemplary embodiment of this disclosure, the similarity prediction model includes two network branches, each network branch including at least a second fully connected layer and a second attention layer; the step of inputting the first seed user vector and the first candidate user vector into the similarity prediction model to generate a second seed user vector and a second candidate user vector includes:

[0028] The second fully connected layer is used to perform feature dimensionality reduction on the first seed user vector and the first candidate user vector to obtain the seed user intermediate vector and the candidate user intermediate vector.

[0029] Based on the seed user intermediate vector and the candidate user intermediate vector, the second seed user vector and the second candidate user vector are generated through the second attention layer.

[0030] In one exemplary embodiment of this disclosure, the method further includes:

[0031] Obtain the first training sample, where each first training sample is a user feature vector and an advertising feature vector of a candidate user;

[0032] The user feature vector and advertising feature vector of each candidate user are input into the user representation model to obtain the corresponding user representation vector and advertising representation vector;

[0033] A first objective function is constructed based on the user representation vector and the advertisement representation vector;

[0034] Based on the first objective function, the parameters of the user representation model are iteratively updated. When the iteration termination condition is met, the training of the user representation model is completed.

[0035] In one exemplary embodiment of this disclosure, the method further includes:

[0036] Obtain a second training sample, each of which is a vector pair consisting of a seed user vector and a target user vector;

[0037] The similarity between the seed user vector and the target user vector is determined using the similarity prediction model.

[0038] A second objective function is constructed based on the similarity between the seed user vector and the target user vector;

[0039] Based on the second objective function, the parameters of the similarity prediction model are iteratively updated. When the iteration termination condition is met, the training of the similarity prediction model is completed.

[0040] According to a second aspect of this disclosure, a target user determination device is provided, comprising:

[0041] The feature vector generation module is used to acquire multidimensional feature data of multiple users and generate user feature vectors and advertising feature vectors based on the multidimensional feature data, wherein the multiple users include seed users and candidate users;

[0042] The representation vector generation module is used to input the user feature vector and the advertising feature vector into the user representation model to generate a user representation vector, wherein the user representation vector includes a first seed user vector and a first candidate user vector.

[0043] The similarity calculation module is used to input the first seed user vector and the first candidate user vector into the similarity prediction model, generate the second seed user vector and the second candidate user vector, and calculate the similarity between the second seed user vector and the second candidate user vector.

[0044] The target user determination module is used to determine the target user among the candidate users based on the similarity.

[0045] According to a third aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the preceding claims.

[0046] According to a fourth aspect of this disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the method described in any of the preceding methods by executing the executable instructions.

[0047] The exemplary embodiments disclosed herein may have some or all of the following beneficial effects:

[0048] In the target user determination method provided in the exemplary embodiments of this disclosure, multidimensional feature data of multiple users is acquired, and user feature vectors and advertising feature vectors are generated based on the multidimensional feature data. The user feature vectors and advertising feature vectors are input into a user representation model to generate a first seed user vector and a first candidate user vector. The first seed user vector and the first candidate user vector are input into a similarity prediction model to generate a second seed user vector and a second candidate user vector, and the similarity between the second seed user vector and the second candidate user vector is calculated. The target user is determined from the candidate users based on the similarity. On the one hand, by fully learning and combining different feature domains through the user representation model to generate user representation vectors, these vectors can more accurately reflect the user's potential interests, facilitating the precise identification of similar groups of seed users. Moreover, the seed user vectors and candidate user vectors output by the similarity prediction model can more accurately calculate the similarity between seed users and candidate users, further improving the accuracy of similar group expansion and thus improving the accuracy of advertising placement. On the other hand, target users can be determined directly through the user representation model and the similarity prediction model without manual selection, improving the efficiency of similar group expansion.

[0049] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0050] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0051] Figure 1 A schematic diagram of an exemplary system architecture for applying the target user determination method and apparatus in the embodiments of this disclosure is shown;

[0052] Figure 2 A flowchart illustrating a target user determination method according to an embodiment of the present disclosure is shown schematically;

[0053] Figure 3 This illustration schematically shows a structural diagram of a user representation model according to an embodiment of the present disclosure;

[0054] Figure 4 This illustration schematically shows a structural diagram of a similarity prediction model according to an embodiment of the present disclosure;

[0055] Figure 5A flowchart illustrating an offline advertising targeted delivery method according to an embodiment of the present disclosure is shown schematically;

[0056] Figure 6 A block diagram of a target user determination device according to an embodiment of the present disclosure is shown schematically;

[0057] Figure 7 The schematic diagram illustrates a structural schematic of a computer system suitable for implementing an electronic device according to embodiments of the present disclosure. Detailed Implementation

[0058] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0059] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0060] Figure 1 A schematic diagram of an exemplary system architecture for a target user determination method and apparatus to which embodiments of the present disclosure can be applied is shown.

[0061] like Figure 1As shown, system architecture 100 may include one or more of terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables. Terminal devices 101, 102, and 103 may be various electronic devices with displays, including but not limited to desktop computers, laptops, smartphones, and tablets. It should be understood that... Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, there can be any number of terminal devices, networks, and servers. For example, server 105 could be a server cluster composed of multiple servers.

[0062] The target user determination method provided in this embodiment can be executed by server 105. Correspondingly, the target user determination device can be set in server 105. The server can send user feature vectors, advertising feature vectors, user representation vectors, and target users to the terminal device, and the terminal device can display them to the staff. However, it is easy for those skilled in the art to understand that the target user determination method provided in this embodiment can also be executed by terminal devices 101, 102, and 103. Correspondingly, the target user determination device can also be set in terminal devices 101, 102, and 103. For example, after execution by the terminal device, the user feature vectors, advertising feature vectors, user representation vectors, and target users can be directly displayed on the display screen of the terminal device for display to the staff. This exemplary embodiment does not impose any special limitations on this.

[0063] The technical solutions of the embodiments of this disclosure are described in detail below:

[0064] This disclosure uses offline advertising in the logistics industry as an example to illustrate its implementation. Currently, offline advertising primarily relies on manual audience segmentation based on users' historical behavior information and advertisers' needs, which is inefficient. Furthermore, manual audience segmentation heavily depends on the historical behavior information of the user corresponding to the item (material), easily leading to a "Matthew effect" in advertising placement. For example, niche but high-quality long-tail content may be considered to have a low placement priority, thus failing to receive effective advertising.

[0065] To address one or more of the aforementioned issues, this example implementation provides a method for identifying target users. It establishes a look-alike model based on an attention mechanism, enabling precise expansion of the target audience for offline logistics advertising. Specifically, the look-alike technology can expand from a small group of seed users using big data analysis and machine learning to identify a similar user base. This involves automatically expanding based on shared attributes of the seed user group to broaden the potential user reach and improve marketing / advertising effectiveness.

[0066] It should be stated that the methods and approaches for obtaining multidimensional feature data of users in the exemplary embodiments of this disclosure are compliant and legal. (Reference) Figure 2 As shown, the target user determination method may include the following steps S210 to S240:

[0067] Step S210. Obtain multidimensional feature data of multiple users, and generate user feature vectors and advertising feature vectors based on the multidimensional feature data, wherein the multiple users include seed users and candidate users;

[0068] Step S220. Input the user feature vector and the advertisement feature vector into the user representation model to generate a user representation vector, wherein the user representation vector includes a first seed user vector and a first candidate user vector;

[0069] Step S230. Input the first seed user vector and the first candidate user vector into the similarity prediction model to generate the second seed user vector and the second candidate user vector, and calculate the similarity between the second seed user vector and the second candidate user vector;

[0070] Step S240. Determine the target user among the candidate users based on the similarity.

[0071] In the target user determination method provided in the exemplary embodiments of this disclosure, multidimensional feature data of multiple users is acquired, and user feature vectors and advertising feature vectors are generated based on the multidimensional feature data. The user feature vectors and advertising feature vectors are input into a user representation model to generate a first seed user vector and a first candidate user vector. The first seed user vector and the first candidate user vector are input into a similarity prediction model to generate a second seed user vector and a second candidate user vector, and the similarity between the second seed user vector and the second candidate user vector is calculated. The target user is determined from the candidate users based on the similarity. On the one hand, by fully learning and combining different feature domains through the user representation model to generate user representation vectors, these vectors can more accurately reflect the user's potential interests, facilitating the precise identification of similar groups of seed users. Moreover, the seed user vectors and candidate user vectors output by the similarity prediction model can more accurately calculate the similarity between seed users and candidate users, further improving the accuracy of similar group expansion and thus improving the accuracy of advertising placement. On the other hand, target users can be determined directly through the user representation model and the similarity prediction model without manual selection, improving the efficiency of similar group expansion.

[0072] The steps described above in this example implementation will now be explained in more detail.

[0073] In step S210, multidimensional feature data of multiple users are obtained, and user feature vectors and advertising feature vectors are generated based on the multidimensional feature data, wherein the multiple users include seed users and candidate users.

[0074] In the logistics industry, offline advertising refers to distributing advertising materials to warehouses, which then deliver the ads to users. Offline advertising can take the form of package advertising, advertising on outer packaging, etc., and the delivery process relies on the logistics packages carrying orders in the warehouse. For example, an advertising leaflet can be pasted on the outer packaging of a package; when a user receives the package and browses or clicks on a product through a link on the leaflet, this can be considered a successful offline advertising delivery. When using a look-alike model to determine the target audience for advertising, seed users can be identified, and then a similar group of users can be identified. A seed user can be a single user or a group of users. For example, seed users could be 1 million users who clicked / shared / purchased a product. Candidate users could be multiple users in the user pool who have not previously received ads for that product, such as 20 million users aged 15 to 50.

[0075] Multidimensional user characteristic data can include user characteristic data and advertising characteristic data. User characteristic data can be further divided into attribute characteristic data, behavioral characteristic data, and other dimensions. Advertising characteristic data can refer to material information in a user's logistics order. Attribute characteristic data can include user profile information such as age, gender, education, and occupation, as well as information such as the user's frequently used address and interests. Behavioral characteristic data can include user browsing, clicking, inquiries, adding to cart, and favorites behaviors. Advertising characteristic data can include product SKU (stockkeeping unit) and product category data.

[0076] After obtaining multidimensional feature data of seed users and candidate users, the feature data of each dimension can be vectorized to obtain user feature vectors and advertising feature vectors. For example, user feature data can be vectorized to generate user feature vectors, and advertising feature data can be vectorized to generate advertising feature vectors. For instance, user profile information and multiple behavioral characteristics can be used to characterize users, and material information, product SKUs, and product categories contained within these behavioral characteristics can be used to characterize advertisements.

[0077] Taking user feature data as an example, user feature data can include at least one of discrete feature data and continuous feature data. Specifically, user browsing, clicking, inquiries, adding to cart, and favorites behaviors are continuous feature data, while user age, gender, education level, and product category are discrete feature data. For example, the values ​​of each discrete feature can be statistically analyzed to determine the frequency of each feature data occurrence, which is then used as the first feature data. This first feature data is then encoded to obtain a first feature vector. For instance, the first feature data can be encoded as a one-hot encoding feature. For continuous feature data, normalization can be performed to generate a second feature vector. Normalizing continuous features can transform features of different magnitudes into features of the same magnitude, avoiding the impact of different numerical levels on the model. For example, each feature dimension can be linearly mapped to a target range. If mapped to [0, 1] or [-1, 1], normalization can also be performed using the standard deviation; this disclosure does not limit this approach.

[0078] In this exemplary implementation, a pipeline mechanism can be used to serialize the vectorization process of feature data across various dimensions, thereby concatenating multiple first feature vectors and multiple second feature vectors to generate feature vectors for each user. It can be seen that the feature vectors of each user consist of normalized continuous feature data and encoded discrete feature data. The pipeline mechanism is a batch processing technique that can improve data processing efficiency.

[0079] Similarly, when vectorizing ad feature data, statistics can be performed on all ad feature values ​​for a user, such as using the mode or mean of each ad feature as the processed feature data. User purchasing power can be characterized by statistically analyzing data such as average favorites, add-to-cart activity, and average order price. User loyalty can be measured by statistically analyzing average browsing time and order frequency over a certain period. Likewise, categorical features in ad feature data can be encoded, numerical features can be normalized, and then aggregated through a pipeline mechanism to obtain ad feature vectors. The process of generating ad feature vectors is similar to that of generating user feature vectors, and will not be elaborated upon here.

[0080] To increase the diversity of advertising features, Chinese word segmentation can be performed on user browsing, adding to cart, and favorited product information and product SKUs, dividing them into finer-grained range attributes to form new features. For example, word segmentation can be based on dictionaries, statistics, or rules; this disclosure does not impose specific limitations. For instance, by segmenting product SKUs, the color features of the product can be obtained.

[0081] Importantly, in the exemplary embodiments disclosed herein, the determination of target users can include two scenarios: audience expansion and potential customer mining. Audience expansion refers to expanding the historical audience in the context of existing advertising campaigns. Potential customer mining refers to mining users from the user pool in the context of new advertising campaigns to identify potential customers. Correspondingly, seed users can be either seed users from past advertising campaigns or seed users for upcoming advertising campaigns. In the context of existing advertising campaigns, seed user groups for different third-level product categories can be established based on the advertiser's successful historical advertising information. For example, multiple seed user groups can be established for mobile phone advertisements from different brands. However, in the context of new advertising campaigns, there is no historical advertising information to refer to. Therefore, similarity recall of advertising materials can be performed, and seed user groups of similar products advertised can be used as seed user groups for upcoming advertising campaigns.

[0082] For example, a first ad feature vector for the ad to be delivered and a second ad feature vector for historical ads can be obtained, and the similarity between the first and second ad feature vectors can be calculated to determine the target ad from the historical ads that is similar to the ad to be delivered. Finally, seed users for the ad to be delivered can be constructed based on the seed users of the target ad. For example, the first ad feature vector can be obtained based on features from multiple domains such as product SKU and product category in the ad to be delivered, and the second ad feature vector can be obtained based on features from multiple domains such as product SKU and product category in historical ads. The similarity between the two vectors is obtained by calculating the distance between the first and second ad feature vectors. For example, the cosine distance, Euclidean distance, etc., between the first and second ad feature vectors can be calculated. By calculating the similarity between the ad to be delivered and multiple historical ads, the historical ad with the highest similarity to the ad to be delivered can be selected as the target ad. It is understandable that the seed users of the target ad are more likely to be interested in the products in the ad. Therefore, the seed users of the target ad can be used as the seed users of the ad to be placed, so as to use the seed users to discover potential customers in the new ad placement scenario.

[0083] In this example, by accumulating a seed user group from successful historical advertising campaigns and using this seed user group to provide knowledge guidance in new advertising scenarios, the lack of human experience can be compensated for, and the precise discovery of potential customers can be achieved.

[0084] In step S220, the user feature vector and the advertisement feature vector are input into the user representation model to generate a user representation vector, which includes a first seed user vector and a first candidate user vector.

[0085] Since users include seed users and candidate users, the user feature vectors generated based on the multidimensional feature data of each user can include seed user feature vectors and candidate user feature vectors. In one example implementation, the user representation model can be an improved YouTube DNN (Deep Semantic Node) model, which includes at least a first fully connected layer and a first attention layer. For example, the first fully connected layer can be used to perform feature dimensionality reduction on the user feature vectors and ad feature vectors to obtain intermediate user vectors and intermediate ad vectors. Further, based on the intermediate user vectors and intermediate ad vectors, a first seed user vector and a first candidate user vector can be generated through the first attention layer.

[0086] refer to Figure 3The diagram shows a schematic of a YouTube DNN model incorporating an Attention mechanism. This model contains a three-layer DNN structure: an input layer 310, a hidden layer 320, and an output layer 330. The input layer 310 processes the input features; for example, it concatenates or averages the user and ad features, then inputs the processed user and ad features into a fully connected layer 1 (301) for spatial transformation. The hidden layer 320 includes fully connected layers 1 (301), 2 (303), and 3 (304). Each fully connected layer is connected to an activation function, such as ReLU (Rectified Linear Units). A first attention layer (302) is added between fully connected layers 1 (301) and 2 (303), allowing for sufficient learning and combination of different feature domains. The output layer 330 outputs a user feature vector and an ad feature vector. Understandably, in order to utilize the attention layer to obtain local feature saliency, the attention layer is usually added to a shallow fully connected layer. In other examples, the attention layer can also be added to a deep fully connected layer, such as adding a first attention layer (302) between fully connected layer 2 (303) and fully connected layer 3 (304), which is not limited in this disclosure.

[0087] For example, user feature vectors and advertising feature vectors obtained through the Pipeline mechanism can be input as follows: Figure 3 In the YouTube DNN model shown, specifically, the user feature vector and the ad feature vector are average pooled and then fed into a fully connected layer 1 (301). The user feature vector and the ad feature vector after average pooling in the fully connected layer 1 (301) can be used for feature dimensionality reduction to obtain the user intermediate vector and the ad intermediate vector. Then, the user intermediate vector and the ad intermediate vector are input into the first attention layer (302). The first attention layer (302) can capture the composition of the ad intermediate vector and weight the ad intermediate vectors that are similar to the user intermediate vector, thereby improving the expressive power of the ad intermediate vector. Correspondingly, after outputting two multidimensional vectors from the first attention layer (302), the two multidimensional vectors can be spatially transformed through the fully connected layer 2 (303) and the fully connected layer 3 (304) in sequence to obtain the user representation vector and the ad representation vector. The user representation vector may include the first seed user vector and the first candidate user vector.

[0088] In this example, an attention mechanism is integrated into the YouTube DNN model. The attention mechanism enables deep models to automatically focus on key features amidst a wealth of information. The output of the model's last hidden layer is used as the user representation vector. Low-order features of different dimensions can be aggregated in this space, thus accurately reflecting the user's potential interests and facilitating precise audience expansion in the future.

[0089] In step S230, the first seed user vector and the first candidate user vector are input into the similarity prediction model to generate a second seed user vector and a second candidate user vector, and the similarity between the second seed user vector and the second candidate user vector is calculated.

[0090] refer to Figure 4 The diagram illustrates a structural schematic of a similarity prediction model. This model employs a dual-tower structure, comprising two network branches. Each branch includes at least a second fully connected layer (401) and a second attention layer (402), used to process the seed user representation vector and the candidate user representation vector, respectively. For example, the second fully connected layers can be used to perform feature dimensionality reduction on the first seed user vector and the first candidate user vector, yielding intermediate vectors for the seed user and candidate users. Further, based on the intermediate vectors for the seed user and candidate users, the second attention layer can generate a second seed user vector and a second candidate user vector. Finally, by calculating the similarity between the second seed user vector and the second candidate user vector, the target user among the candidate users is identified, and offline advertising is delivered to the target user.

[0091] For example, the first seed user vector can be input into the first branch of the similarity prediction model, and spatial transformation can be performed through the second fully connected layer (401) in the first branch and the activation function connected to the second fully connected layer (401). Then, the dimensionality-reduced first seed user vector can be input into the second attention layer (402) to obtain the second seed user vector. The activation function can be the PReLU (Parametric Rectified Linear Unit) activation function. Similarly, the first candidate user vector can be input into the second branch of the similarity prediction model for the same processing to obtain the second candidate user vector; the specific processing steps are not detailed here. It is understandable that the attention mechanism can select more valuable vector features from the seed user representation vector compared to the current candidate users, thus improving the expressive power of the seed user representation vector and enabling more accurate calculation of the similarity between seed users and candidate users.

[0092] After obtaining the second seed user vector and the second candidate user vector, the similarity between the two vectors can be calculated to determine candidate users similar to the seed user. For example, the cosine distance, Euclidean distance, Mahalanobis distance, etc., can be calculated between the two vectors to obtain a similarity score. It should be noted that the similarity scores between each candidate user and the current seed user can be normalized to a range of 0 to 1. This facilitates the filtering of candidate users based on the normalized similarity scores to identify multiple target users similar to the seed user. In other examples, the similarity between each candidate user and multiple seed users can be calculated, and the multiple similarity scores can be weighted and averaged. The result is then used as the final similarity score for each candidate user, and the target user among the candidate users can be determined based on the similarity score.

[0093] In one example implementation, before predicting target users online, a user representation model and a similarity prediction model can be pre-trained. For the user representation model, a YouTube DNN model incorporating an attention mechanism is used as an example. Exemplarily, first training samples can be obtained, each consisting of a candidate user's user feature vector and an advertisement feature vector. The user feature vector and advertisement feature vector of each candidate user can be input into the user representation model to obtain the corresponding user representation vector and advertisement representation vector. A first objective function is constructed based on these vectors. The parameters of the user representation model are iteratively updated based on this first objective function. When the iteration termination condition is met, the training of the model is complete. The first objective function can be an NCE (Noise Contrastive Estimation) loss function. It is understood that in the training samples, users who complete the product link jump are considered positive samples, those who are not reached or do not complete the product link jump after reaching the user are considered negative samples, and other samples in intermediate stages are invalid data.

[0094] For example, the first training samples can be divided into a training set, a validation set, and a test set in a ratio such as 7:1:2. The training set is used to train the model parameters, the validation set is used to observe performance changes during model training to avoid underfitting or overfitting, and the test set is used to evaluate the model's performance. When tuning the parameters of the user representation model, taking the YouTube DNN model with an attention mechanism as an example, this can include setting the model parameter initialization, maximum number of iterations, activation function selection, optimizer selection, early stopping conditions, and learning rate. During training of the model's parameters, training terminates when all parameters tend to converge or when a certain number of iterations is met. The process of tuning model parameters also includes tuning model features, which can be evaluated based on the similarity between features, and feature dimensionality should be reduced as much as possible to avoid overfitting.

[0095] For the similarity prediction model, during training, a second training sample is obtained, each consisting of a vector pair composed of a seed user vector and a target user vector. The similarity prediction model determines the similarity between the seed user vector and the target user vector. A second objective function is constructed based on this similarity. The parameters of the similarity prediction model are iteratively updated based on this second objective function. Training is complete when the iteration termination condition is met. The similarity prediction model is a multi-class classification task, requiring the prediction of the category of the seed user group corresponding to each candidate user. Therefore, the second objective function can be multiple sets of cross-entropy loss functions. Similarly, the second training samples can be divided into training, validation, and test sets in a ratio such as 7:1:2. The model's prediction result is the probability that each candidate user belongs to the seed user group. Training terminates when all parameters in the model converge or when a certain number of iterations is met.

[0096] After the user representation model and similarity prediction model are trained, the performance of each model can be evaluated using a test set. For example, metrics such as AUC (Area Under Curve), precision, recall, prec@k, and F1 score can be used for evaluation. AUC is a commonly used metric in ranking model evaluation, and the ROC curve is also known as the receiver operating characteristic curve. Taking the evaluation of positive samples as an example: AUC is the probability that the predicted value of a positive sample is greater than that of a negative sample in a given data set of positive and negative samples.

[0097] Pre represents the model's accuracy, where TP represents the data that were successfully predicted as positive, FP represents the data that were incorrectly predicted as positive, and TP+FP represents all data that were predicted as positive.

[0098]

[0099] Rec represents the model's recall rate, where TP represents the number of positive samples successfully predicted as positive, FN represents the number of positive samples incorrectly predicted as negative, and TP+FN represents all positive sample data.

[0100]

[0101] F1 is the harmonic mean of the model's precision and recall, where Pre is the model's precision and Rec is the model's recall.

[0102]

[0103] prec@k represents the percentage of users actually reached by the top k recommended ad categories.

[0104] In this example, an attention mechanism is introduced into the user representation model and the similarity prediction model. This attention mechanism enables the same set of seed user vectors to be weighted differently when predicting the interests of different candidate users, thereby improving the expressive power of the seed user vectors and thus improving the accuracy of the model prediction.

[0105] In step S240, the target user among the candidate users is determined based on the similarity.

[0106] For example, all candidate users can be sorted in descending order based on their similarity scores, and the top 20 (e.g., 20) can be selected as target users. Alternatively, all candidate users can be sorted in ascending order based on their similarity scores, and the bottom 20 (e.g., 20) can be selected as target users. A similarity threshold can also be preset, and candidate users meeting the threshold can be selected as target users. For example, the similarity threshold can be set to 0.5 or 0.7; this disclosure does not specifically limit this. For instance, candidate users with similarity scores greater than the similarity threshold can be selected as target users.

[0107] Once the target users are identified, advertising can be quantitatively targeted based on the number of target users. Cost threshold control can also be implemented by combining advertiser needs and similarity distribution. For example, when the similarity threshold is set to 0.5, only 1 / 3 of the target users have a similarity score greater than 0.7. In this case, advertising can be targeted only to target users with a similarity score greater than 0.7, thus adjusting the lower limit of advertising costs. During advertising, the advertising media format (such as waybill ads, self-service lockers, etc.) can be combined to target the ads to the target users' packages or community media, achieving precise audience selection and potential customer discovery, and reaching the target user group.

[0108] In this exemplary embodiment, based on successful pre-ad placement experience and combined with a large amount of historical behavioral information from user logistics scenarios, a YouTube DNN model incorporating an attention mechanism can be used to obtain accurate behavioral representations of seed users. Furthermore, during real-time online prediction, the similarity score between the behavioral representations of each candidate user in the user pool and the cluster centers of the clustered seed users can be calculated, and the target expanded user group can be accurately selected based on this score. For example, methods such as K-means clustering, mean-shift clustering, and density-based clustering can be used to reduce the computational load of the model; this embodiment does not limit this approach.

[0109] In one example implementation, reference Figure 5 As shown, the targeted delivery of advertisements can be completed according to steps S501 to S505.

[0110] Step S501. Establish a seed user group: Based on the successful experience of previous advertising campaigns, establish seed user groups for different third-level product categories. For old advertising scenarios, the seed user group can be obtained directly. For new advertising scenarios, similarity recall of advertising materials can be performed based on i2i (item-item similarity calculation), and the seed users of similar products can be used for potential customer mining for the current advertisement;

[0111] Step S502. Generate seed user representation vector and candidate user representation vector: Learn the high-order representation of user features by using a YouTube DNN model that integrates the Attention mechanism to obtain seed user representation vector and candidate user representation vector;

[0112] Step S503. Similarity Prediction: The seed user representation vector and the candidate user representation vector are used as inputs to the basic model (fully connected layer and PReLU activation function). After the vector space is transformed by the attention layer, the cosine similarity between the two vectors is calculated to obtain the similarity prediction score.

[0113] Step S504. Determine the target audience: After obtaining the estimated similarity scores between candidate users and seed users, multiple candidate users who meet the similarity threshold can be determined as the target audience pool for the current advertising campaign, based on the advertiser's budget or campaign volume requirements.

[0114] Step S505. Offline Advertising Placement: Each user corresponds to one or more types of offline advertising reach media. Under the constraint of the advertiser's funds, the actual advertising target audience and corresponding offline reach media with the maximum cumulative similarity can be obtained by solving a multidimensional programming problem to ensure that the current advertising gains are maximized.

[0115] In the target user determination method provided in the exemplary embodiments of this disclosure, multidimensional feature data of multiple users is acquired, and user feature vectors and advertising feature vectors are generated based on the multidimensional feature data. The user feature vectors and advertising feature vectors are input into a user representation model to generate a first seed user vector and a first candidate user vector. The first seed user vector and the first candidate user vector are input into a similarity prediction model to generate a second seed user vector and a second candidate user vector, and the similarity between the second seed user vector and the second candidate user vector is calculated. The target user is determined from the candidate users based on the similarity. On the one hand, by fully learning and combining different feature domains through the user representation model to generate user representation vectors, these vectors can more accurately reflect the user's potential interests, facilitating the precise identification of similar groups of seed users. Moreover, the seed user vectors and candidate user vectors output by the similarity prediction model can more accurately calculate the similarity between seed users and candidate users, further improving the accuracy of similar group expansion and thus improving the accuracy of advertising placement. On the other hand, target users can be determined directly through the user representation model and the similarity prediction model without manual selection, improving the efficiency of similar group expansion.

[0116] It should be noted that although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0117] Furthermore, in this exemplary embodiment, a target user identification device is also provided, which can be applied to a server or terminal device. (See reference) Figure 6 As shown, the target user determination device 600 may include a feature vector generation module 610, a representation vector generation module 620, a similarity calculation module 630, and a target user determination module 640, wherein:

[0118] The feature vector generation module 610 is used to acquire multidimensional feature data of multiple users and generate user feature vectors and advertising feature vectors based on the multidimensional feature data, wherein the multiple users include seed users and candidate users;

[0119] The representation vector generation module 620 is used to input the user feature vector and the advertisement feature vector into the user representation model to generate a user representation vector, wherein the user representation vector includes a first seed user vector and a first candidate user vector.

[0120] The similarity calculation module 630 is used to input the first seed user vector and the first candidate user vector into the similarity prediction model, generate the second seed user vector and the second candidate user vector, and calculate the similarity between the second seed user vector and the second candidate user vector.

[0121] The target user determination module 640 is used to determine the target user among the candidate users based on the similarity.

[0122] In one optional implementation, the seed users include one of the seed users who have previously received advertisements and the seed users who are to receive advertisements; the target user determination device 600 further includes:

[0123] The feature vector acquisition module is used to acquire the first feature vector of the advertisement to be delivered and the second feature vector of the historical advertisements.

[0124] The target advertisement determination module is used to calculate the similarity between the first advertisement feature vector and the second advertisement feature vector, and determine the target advertisements that are similar to the advertisement to be delivered in the historical advertisements based on the similarity.

[0125] The seed user determination module is used to construct the seed users of the advertisement to be delivered based on the seed users of the target advertisement.

[0126] In one optional implementation, the multidimensional feature data includes user feature data and advertising feature data; the feature vector generation module 610 includes:

[0127] The first vectorization module is used to perform vectorization processing on the user feature data to generate the user feature vector.

[0128] The second vectorization module is used to perform vectorization processing on the advertising feature data to generate the advertising feature vector.

[0129] In one optional implementation, the user feature data includes at least one of discrete feature data and continuous feature data; the first vectorization module includes:

[0130] The first feature data processing submodule is used to perform statistics on each of the discrete feature data to obtain multiple first feature data;

[0131] The first feature vector generation submodule is used to encode each first feature data and generate multiple first feature vectors;

[0132] The second feature vector generation submodule is used to normalize each of the continuous feature data and generate multiple second feature vectors.

[0133] The first vector generation submodule is used to concatenate the plurality of first feature vectors and the plurality of second feature vectors to generate the user feature vector.

[0134] In one optional implementation, the user representation model includes at least a first fully connected layer and a first attention layer; the representation vector generation module 620 includes:

[0135] The first feature dimensionality reduction submodule is used to perform feature dimensionality reduction on the user feature vector and the advertisement feature vector using the first fully connected layer to obtain the user intermediate vector and the advertisement intermediate vector.

[0136] The second vector generation submodule is used to generate the first seed user vector and the first candidate user vector through the first attention layer based on the user intermediate vector and the advertisement intermediate vector.

[0137] In one optional implementation, the similarity prediction model includes two network branches, each network branch including at least a second fully connected layer and a second attention layer; the similarity calculation module 630 includes:

[0138] The second feature dimensionality reduction submodule is used to perform feature dimensionality reduction on the first seed user vector and the first candidate user vector using the second fully connected layer to obtain the seed user intermediate vector and the candidate user intermediate vector.

[0139] The third vector generation submodule is used to generate the second seed user vector and the second candidate user vector through the second attention layer based on the seed user intermediate vector and the candidate user intermediate vector.

[0140] In an optional implementation, the target user determination device 600 further includes a first model training module, which is configured to acquire first training samples, each of which is a user feature vector and an advertising feature vector of a candidate user; input the user feature vector and advertising feature vector of each candidate user into the user representation model to obtain corresponding user representation vectors and advertising representation vectors; construct a first objective function based on the user representation vectors and the advertising representation vectors; iteratively update the parameters of the user representation model based on the first objective function; and complete the training of the user representation model when the iteration termination condition is met.

[0141] In an optional implementation, the target user determination device 600 further includes a second model training module, which is configured to acquire second training samples, each of which is a vector pair consisting of a seed user vector and a target user vector; determine the similarity between the seed user vector and the target user vector using the similarity prediction model; construct a second objective function based on the similarity between the seed user vector and the target user vector; iteratively update the parameters of the similarity prediction model based on the second objective function; and complete the training of the similarity prediction model when the iteration termination condition is met.

[0142] The specific details of each module in the aforementioned target user determination device have been described in detail in the corresponding target user determination method, so they will not be repeated here.

[0143] The modules in the above-described device can be general-purpose processors, including central processing units (CPUs), network processors, etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. Each module can also be implemented using software, firmware, etc. The processors in the above-described device can be independent processors or integrated together.

[0144] Exemplary embodiments of this disclosure also provide a computer-readable storage medium having a program product stored thereon capable of implementing the methods described above in this specification. In some possible embodiments, various aspects of this disclosure may also be implemented as a program product including program code that, when run on an electronic device, causes the electronic device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. This program product may be a portable compact disc read-only memory (CD-ROM) including program code and may run on an electronic device, such as a personal computer. However, the program product of this disclosure is not limited thereto. In this document, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0145] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0146] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0147] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0148] Program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing devices can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0149] Exemplary embodiments of this disclosure also provide an electronic device capable of implementing the above-described method. Referring below... Figure 7 To describe an electronic device 700 according to such an exemplary embodiment of the present disclosure. Figure 7 The electronic device 700 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0150] like Figure 7 As shown, the electronic device 700 can be represented as a general-purpose computing device. The components of the electronic device 700 may include, but are not limited to: at least one processing unit 710, at least one storage unit 720, a bus 730 connecting different system components (including storage unit 720 and processing unit 710), and a display unit 740.

[0151] Storage unit 720 stores program code that can be executed by processing unit 710, causing processing unit 710 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. For example, processing unit 710 can perform... Figure 2 and Figure 5 Any one or more of the method steps.

[0152] Storage unit 720 may include a readable medium in the form of a volatile storage unit, such as random access memory (RAM) 721 and / or cache memory 722, and may further include a read-only memory (ROM) 723.

[0153] The storage unit 720 may also include a program / utility 724 having a set (at least one) of program modules 725, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0154] Bus 730 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0155] Electronic device 700 can also communicate with one or more external devices 800 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 700, and / or with any device that enables electronic device 700 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 750. Furthermore, electronic device 700 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 760. As shown, network adapter 760 communicates with other modules of electronic device 700 via bus 730. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 700, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0156] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the method according to the exemplary embodiments of this disclosure.

[0157] Furthermore, the above figures are merely illustrative representations of the processes included in the methods according to exemplary embodiments of this disclosure, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0158] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0159] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for determining target users, characterized in that, include: Obtain the first ad feature vector of the ad to be delivered and the second ad feature vector of the historical ads; Calculate the similarity between the first ad feature vector and the second ad feature vector, and determine the target ad that is similar to the ad to be delivered in the historical ad delivery based on the similarity; construct the seed users of the ad to be delivered based on the seed users of the target ad. Acquire multidimensional feature data of multiple users, and generate user feature vectors and advertising feature vectors based on the multidimensional feature data, wherein the multiple users include seed users and candidate users; The user feature vector and the advertisement feature vector are input into a user representation model to generate a user representation vector. The user representation model includes at least a first fully connected layer, a first attention layer, a second fully connected layer, and a third fully connected layer. The first attention layer is used to learn and combine different feature domains. The process of inputting the user feature vector and the advertisement feature vector into the user representation model to generate the user representation vector includes: using the first fully connected layer to perform feature dimensionality reduction on the user feature vector and the advertisement feature vector to obtain a user intermediate vector and an advertisement intermediate vector; based on the user intermediate vector and the advertisement intermediate vector, the first attention layer captures the composition of the advertisement intermediate vector; the advertisement intermediate vectors similar to the user intermediate vector are weighted; and the two multidimensional vectors output by the first attention layer are sequentially transformed through the second fully connected layer and the third fully connected layer to obtain a user representation vector and an advertisement representation vector. The user representation vector includes a first seed user vector and a first candidate user vector. The first seed user vector and the first candidate user vector are input into the similarity prediction model to generate a second seed user vector and a second candidate user vector, and the similarity between the second seed user vector and the second candidate user vector is calculated; wherein, the similarity prediction model includes two network branches, and each network branch includes at least a second fully connected layer and a second attention layer; The target user is determined from the candidate users based on the similarity.

2. The target user determination method according to claim 1, characterized in that, The multidimensional feature data includes user feature data and advertising feature data; the step of generating user feature vectors and advertising feature vectors based on the multidimensional feature data includes: The user feature data is vectorized to generate the user feature vector; The advertising feature data is vectorized to generate the advertising feature vector.

3. The target user determination method according to claim 2, characterized in that, The user feature data includes at least one of discrete feature data and continuous feature data; the step of vectorizing the user feature data to generate the user feature vector includes: Statistical analysis is performed on each of the discrete feature data to obtain multiple first feature data; Encode each of the first feature data to generate multiple first feature vectors; Normalize each of the continuous feature data to generate multiple second feature vectors; The user feature vector is generated by concatenating the plurality of first feature vectors and the plurality of second feature vectors.

4. The target user determination method according to claim 1, characterized in that, The similarity prediction model includes two network branches, each branch including at least a second fully connected layer and a second attention layer; the step of inputting the first seed user vector and the first candidate user vector into the similarity prediction model to generate the second seed user vector and the second candidate user vector includes: The second fully connected layer is used to perform feature dimensionality reduction on the first seed user vector and the first candidate user vector to obtain the seed user intermediate vector and the candidate user intermediate vector. Based on the seed user intermediate vector and the candidate user intermediate vector, the second seed user vector and the second candidate user vector are generated through the second attention layer.

5. The target user determination method according to claim 1, characterized in that, The method further includes: Obtain the first training sample, where each first training sample is a user feature vector and an advertising feature vector of a candidate user; The user feature vector and advertising feature vector of each candidate user are input into the user representation model to obtain the corresponding user representation vector and advertising representation vector; Construct a first objective function based on the user representation vector and the advertisement representation vector; Based on the first objective function, the parameters of the user representation model are iteratively updated. When the iteration termination condition is met, the training of the user representation model is completed.

6. The target user determination method according to claim 1, characterized in that, The method further includes: Obtain the second training sample, each of the second training samples being a vector pair consisting of a seed user vector and a target user vector; The similarity between the seed user vector and the target user vector is determined using the similarity prediction model. A second objective function is constructed based on the similarity between the seed user vector and the target user vector; Based on the second objective function, the parameters of the similarity prediction model are iteratively updated. When the iteration termination condition is met, the training of the similarity prediction model is completed.

7. A target user determination device, characterized in that, include: A feature vector generation module is used to obtain a first ad feature vector of the ad to be delivered and a second ad feature vector of historical ads; calculate the similarity between the first ad feature vector and the second ad feature vector, and determine the target ads similar to the ad to be delivered in the historical ads based on the similarity; construct seed users for the ad to be delivered based on the seed users of the target ads; obtain multidimensional feature data of multiple users, and generate user feature vectors and ad feature vectors based on the multidimensional feature data, wherein the multiple users include seed users and candidate users; A representation vector generation module is used to input the user feature vector and the advertising feature vector into a user representation model to generate a user representation vector. The user representation model includes at least a first fully connected layer, a first attention layer, a second fully connected layer, and a third fully connected layer. The first attention layer is used to learn and combine different feature domains. The step of inputting the user feature vector and the advertising feature vector into the user representation model to generate the user representation vector includes: using the first fully connected layer to perform feature dimensionality reduction on the user feature vector and the advertising feature vector to obtain a user intermediate vector and an advertising intermediate vector; based on the user intermediate vector and the advertising intermediate vector, capturing the composition of the advertising intermediate vector through the first attention layer, weighting the advertising intermediate vectors similar to the user intermediate vector, and sequentially performing spatial transformation on the two multidimensional vectors output by the first attention layer through the second fully connected layer and the third fully connected layer to obtain the user representation vector and the advertising representation vector. The user representation vector includes a first seed user vector and a first candidate user vector. The similarity calculation module is used to input the first seed user vector and the first candidate user vector into the similarity prediction model, generate the second seed user vector and the second candidate user vector, and calculate the similarity between the second seed user vector and the second candidate user vector; wherein, the similarity prediction model includes two network branches, and each network branch includes at least a second fully connected layer and a second attention layer; The target user determination module is used to determine the target user among the candidate users based on the similarity.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-6.

9. An electronic device, characterized in that, include: processor; as well as Memory for storing the executable instructions of the processor; The processor is configured to execute the method of any one of claims 1-6 by executing the executable instructions.

Images