A user group matching method and device, electronic equipment and computer storage medium

Abstract

The application discloses a user group matching method and device, electronic equipment and computer storage medium. The user characteristics of a first user and a plurality of second users are obtained from G users. The similarity between the first user and each second user is calculated based on a collaborative filtering method. The feature image between the first user and each second user is generated based on the calculated similarity. Finally, the feature image is input into an image classification model to obtain target information. The user group information in the target information can indicate whether the first user and the second user belong to the same user group. The similarity problem of non-image problems can be converted into an image problem to improve the efficiency of outputting the target user group. In addition, the feature image is identified and classified by combining the image classification model to obtain the target information for indicating the target user group, which can improve the accuracy of outputting the target user group.

Description

Technical Field

[0001] This application belongs to the field of data service technology, and in particular relates to a user group matching method, apparatus, electronic device and computer storage medium. Background Technology

[0002] As business grows, the need for precision in identifying target user groups is increasing when launching new business initiatives. Accurately identifying target user groups can reduce the investment costs of launching new businesses and improve the success rate of activities. Existing methods for selecting target user groups mainly include: selecting user tags from a pre-defined tag library based on daily business needs, or using big data analysis, city-level searches, and business department analysis to analyze and extract data for specific scenarios.

[0003] Existing methods for selecting target user groups have certain limitations and are not very suitable for specific scenarios. Furthermore, the accuracy of user group selection is often affected by the depth and detail of the analysis, which leads to low accuracy in identifying target user groups. Summary of the Invention

[0004] This application provides a user group matching method, apparatus, electronic device, and computer storage, which can improve the efficiency and accuracy of outputting target user groups.

[0005] In a first aspect, embodiments of this application provide a user group matching method, which may include:

[0006] Obtain the user characteristics of the first user and the user characteristics of multiple second users among G users, where G is a positive integer and the multiple second users are users other than the first user among the G users;

[0007] Based on the collaborative filtering method, the similarity between the first user and each second user is calculated according to the user characteristics of the first user and the user characteristics of each second user.

[0008] Based on the similarity between the first user and multiple second users, feature images are generated between the first user and each of the second users.

[0009] The feature images of the first user and each of the second users are input into the image classification model to obtain target information. The target information includes user group information of the first user and each of the second users. The user group information is used to indicate whether the first user and the second user belong to the same user group.

[0010] In one embodiment, the aforementioned generation of feature images between the first user and each of the second users based on the similarity between the first user and multiple second users may include:

[0011] Based on the similarity between the first user and multiple second users, a similarity matrix is ​​generated between the first user and each second user. The similarity matrix includes the similarity between the first user and each second user.

[0012] Based on the similarity matrix between the first user and multiple second users, feature images are generated between the first user and each of the second users.

[0013] In one embodiment, the user characteristics of each of the G users mentioned above include user sub-features in k dimensions;

[0014] The aforementioned process of generating feature images between the first user and each of the second users based on the similarity matrix between the first user and multiple second users may include:

[0015] The similarity matrix between the first user and multiple second users is split into k similarity sub-matrices corresponding to k dimensions. Each similarity sub-matrix includes the similarity calculated based on the user sub-features of its corresponding dimension.

[0016] Based on k similarity sub-matrices, feature images are generated between the first user and each of the second users.

[0017] In one embodiment, each of the k similarity sub-matrices mentioned above includes N×M similarities, where N and M are positive integers;

[0018] The aforementioned generation of feature images between the first user and each of the second users based on k similarity sub-matrices can include:

[0019] Based on the N×M similarities in each similarity submatrix, determine the N×M pixel gray levels in the color channels corresponding to the similarity submatrix, and obtain the N×M pixel gray levels in k color channels. Different similarity submatrixes correspond to different color channels.

[0020] Based on the N×M pixel grayscale values ​​in k color channels, a feature image containing N×M pixels is generated. The pixel value of each pixel in the feature image is obtained by fusing the pixel grayscale values ​​in its corresponding k color channels.

[0021] In one embodiment, the aforementioned process of splitting the similarity matrix between the first user and multiple second users into k similarity sub-matrices corresponding to k dimensions may include:

[0022] Obtain user sub-features across k dimensions;

[0023] Based on the user sub-features in k dimensions, the similarity scores corresponding to the user sub-features in k dimensions are obtained from the similarity matrix between the first user and each second user, resulting in k similarity sub-matrices.

[0024] In one embodiment, the aforementioned generation of feature images between the first user and each of the second users based on the similarity between the first user and multiple second users may include:

[0025] Based on the similarity between the first user and multiple second users, the pixel value corresponding to each similarity is determined, wherein the pixel value corresponding to different similarities is different;

[0026] Based on the pixel values ​​of the pixels corresponding to each similarity, feature images between the first user and each second user are generated.

[0027] In one embodiment, before inputting the feature images between the first user and each of the second users into the image classification model to obtain the target information, the process may further include:

[0028] Acquire training image samples, which include historical feature images and labels corresponding to the historical feature images. The labels are used to indicate whether the first historical user and the second historical user belong to the same user group. The historical feature images are the feature images between the first historical user and the multiple second historical users among the multiple historical users. The multiple second historical users are the historical users other than the first historical user among the multiple historical users.

[0029] The training image samples are input into the image classification model to be trained, and the image classification model is obtained through training.

[0030] In one embodiment, before obtaining the user characteristics of the first user and the user characteristics of the multiple second users among the G users mentioned above, the process may further include:

[0031] Obtain user characteristic data and user product order data of the first user among G users, as well as user characteristic data and user product order data of multiple second users;

[0032] Based on the user characteristic data and user product order data of the first user among G users, and the user characteristic data and user product order data of multiple second users, generate the user characteristic and order matrix of the first user among G users and the user characteristic and order matrix of multiple second users. The user characteristic and order matrix includes multiple user characteristics and multiple user ordered products.

[0033] The aforementioned acquisition of user characteristics of the first user and multiple second users among G users may include:

[0034] From the user characteristics and ordering matrix of the first user among G users, and the user characteristics and ordering matrix of multiple second users, obtain the user characteristics of the first user among G users and the user characteristics of multiple second users among G users.

[0035] In one embodiment, after inputting the feature images between the first user and each of the second users into the image classification model to obtain the target information, the process may further include:

[0036] Based on the target information, the user characteristics and ordering matrix of the first user among G users, and the user characteristics and ordering matrix of multiple second users, determine the target user ordering products of the target user group. The target user group includes the first user and the second users whose user group information indicates that the first user and the second user belong to the same user group.

[0037] In one embodiment, before calculating the similarity between the first user and each of the second users based on the user characteristics of the first user and the user characteristics of each of the second users, the method may further include:

[0038] Feature engineering is performed on the user characteristics of the first user and the user characteristics of each second user. Feature engineering includes feature understanding, feature cleaning and enhancement, feature construction and feature selection.

[0039] In one embodiment, the image classification model mentioned above includes a generative adversarial network model based on a collaborative filtering method.

[0040] Secondly, embodiments of this application provide a user group matching device, which may include:

[0041] The acquisition module is used to acquire the user characteristics of the first user and the user characteristics of multiple second users among G users, where G is a positive integer and the multiple second users are users other than the first user among the G users.

[0042] The calculation module is used to calculate the similarity between the first user and each second user based on the user characteristics of the first user and the user characteristics of each second user, using the collaborative filtering method.

[0043] The generation module is used to generate feature images between the first user and each of the second users based on the similarity between the first user and multiple second users;

[0044] The input module is used to input the feature images between the first user and each of the second users into the image classification model to obtain target information. The target information includes user group information of the first user and each of the second users. The user group information is used to indicate whether the first user and the second users belong to the same user group.

[0045] Thirdly, embodiments of this application provide an electronic device, the device comprising:

[0046] processor;

[0047] Memory used to store processor-executable instructions;

[0048] The processor is configured to execute instructions to implement the user group matching method as shown in any embodiment of the first aspect.

[0049] Fourthly, embodiments of this application provide a computer storage medium on which a computer program is stored, which, when executed by a processor, implements the user group matching method as shown in any embodiment of the first aspect.

[0050] Fifthly, embodiments of this application also provide a computer program product comprising a computer program stored in a readable storage medium, wherein at least one processor of the device reads from the storage medium and executes the computer program, causing the device to perform the user group matching method shown in any embodiment of the first aspect.

[0051] This application provides a user group matching method, apparatus, electronic device, and computer storage medium. Compared with the prior art, this application has the following advantages:

[0052] This application discloses a user group matching method, apparatus, electronic device, and computer storage medium. It acquires user features of a first user and user features of multiple second users from a pool of G users. Based on a collaborative filtering method, it calculates the similarity between the first user and each of the second users according to their respective user features. Then, based on the similarity between the first user and the multiple second users, it generates feature images between the first user and each of the second users. These feature images are then input into an image classification model to obtain target information. The user group information in the target information indicates whether the first user and the second users belong to the same user group.

[0053] Thus, based on the collaborative filtering method, the similarity of user features between the first user and the second user is calculated, generating feature images between the first user and each second user. This transforms the non-image-based similarity problem into an image-based problem, thereby improving the efficiency of outputting the target user group. Furthermore, combining the feature images with an image classification model for recognition and classification yields target information indicating the target user group, improving the accuracy of outputting the target user group. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart illustrating a user group matching method provided in an embodiment of this application;

[0056] Figure 2 This is a flowchart illustrating a user group matching method provided in an embodiment of this application;

[0057] Figure 3 This is a schematic diagram of a similarity matrix provided in an embodiment of this application;

[0058] Figure 4 This is a flowchart illustrating another user group matching method provided in an embodiment of this application;

[0059] Figure 5 This is a schematic diagram of a multidimensional similarity matrix provided in an embodiment of this application;

[0060] Figure 6 This is a flowchart illustrating another user group matching method provided in an embodiment of this application;

[0061] Figure 7 This is a schematic diagram of a similarity matrix visualization provided in an embodiment of this application;

[0062] Figure 8 This is a schematic diagram of another similarity matrix visualization provided in an embodiment of this application;

[0063] Figure 9 This is a flowchart illustrating another user group matching method provided in an embodiment of this application;

[0064] Figure 10 This is a flowchart illustrating another user group matching method provided in an embodiment of this application;

[0065] Figure 11 This is a flowchart illustrating another user group matching method provided in an embodiment of this application;

[0066] Figure 12 This is a flowchart illustrating another user group matching method provided in an embodiment of this application;

[0067] Figure 13 This is a flowchart illustrating another user group matching method provided in an embodiment of this application;

[0068] Figure 14This is a flowchart illustrating another user group matching method provided in an embodiment of this application;

[0069] Figure 15 This is a schematic diagram of a feature engineering process provided in an embodiment of this application;

[0070] Figure 16 This is a flowchart illustrating another user group matching method provided in an embodiment of this application;

[0071] Figure 17 This is a schematic diagram of the structure of a user group matching device provided in an embodiment of this application;

[0072] Figure 18 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0073] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0074] It should be noted that, in this document, relational terms such as "first" and "second" are used merely 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 apparatus 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 apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0075] As can be seen from the background section, existing methods for selecting target user groups have certain universality and limitations, resulting in relatively low suitability for specific scenarios. Furthermore, the accuracy of user group selection is often affected by the depth and detail of the analysis, leading to low accuracy in identifying target user groups.

[0076] To address the problems existing in the prior art, embodiments of this application provide a user group matching method, apparatus, electronic device, and computer storage medium. By acquiring user features of a first user and multiple second users from a pool of G users, and using a collaborative filtering method, the similarity between the first user and each of the second users is calculated based on the user features of the first user and each of the second users. Then, based on the similarity between the first user and the multiple second users, feature images of the first user and each of the second users are generated. These feature images are input into an image classification model to obtain target information. The user group information in the target information indicates whether the first user and the second users belong to the same user group.

[0077] Thus, based on the collaborative filtering method, the similarity of user features between the first user and the second user is calculated, generating feature images between the first user and each second user. This transforms the non-image-based similarity problem into an image-based problem, thereby improving the efficiency of outputting the target user group. Furthermore, combining the feature images with an image classification model for recognition and classification yields target information indicating the target user group, improving the accuracy of outputting the target user group.

[0078] The user group matching method provided in the embodiments of this application will be described below. Figure 1 As shown in the embodiments of this application, the user group matching method includes the following steps:

[0079] S101: Obtain the user characteristics of the first user and the user characteristics of multiple second users among G users, where G is a positive integer and the multiple second users are users other than the first user among the G users;

[0080] S102: Based on the collaborative filtering method, the similarity between the first user and each second user is calculated according to the user characteristics of the first user and the user characteristics of each second user.

[0081] S103: Based on the similarity between the first user and multiple second users, generate feature images between the first user and each of the second users;

[0082] S104: Input the feature images between the first user and each of the second users into the image classification model to obtain target information. The target information includes user group information of the first user and each of the second users. The user group information is used to indicate whether the first user and the second user belong to the same user group.

[0083] The above describes a user group matching method provided in this application. It obtains user features of a first user and multiple second users from a pool of G users. Based on a collaborative filtering method, it calculates the similarity between the first user and each of the second users according to their respective user features. Then, based on the similarity between the first user and the multiple second users, it generates feature images between the first user and each of the second users. These feature images are then input into an image classification model to obtain target information. The user group information in the target information indicates whether the first user and the second users belong to the same user group.

[0084] Thus, based on the collaborative filtering method, the similarity of user features between the first user and the second user is calculated, generating feature images between the first user and each second user. This transforms the non-image-based similarity problem into an image-based problem, thereby improving the efficiency of outputting the target user group. Furthermore, combining the feature images with an image classification model for recognition and classification yields target information indicating the target user group, improving the accuracy of outputting the target user group.

[0085] In step S101, the user characteristics of the first user and multiple second users are obtained from among G users, where G is a positive integer and the multiple second users are users other than the first user among the G users. In one example, user business characteristic data and user product order data can be extracted from the business data and combined into a user characteristic and product order matrix. The user characteristic data includes different categories of data such as basic information, consumption information, communication information, family information, and traffic information.

[0086] In step S102, based on the collaborative filtering method, the similarity between the first user and each of the second users is calculated according to the user features of the first user and the user features of each second user. In one example, the collaborative filtering method is used to calculate the pairwise similarity of user features in the matrix. It iterates through each user, calculating the similarity of the current user with all other users on each feature. The similarity value between any two users on each feature is then obtained. For example, pairwise similarity calculations are performed on all users out of the obtained G users, and the similarity of the current user with the other G-1 users on each user feature is calculated. It can be understood that similarity can be represented by a similarity value. This allows for more accurate separation of key user features and improves the accuracy of the image classification model algorithm.

[0087] Collaborative filtering is a classic and commonly used recommendation algorithm. The basic idea behind collaborative filtering is to discover users' interests and hobbies based on their historical behavior data, segment users according to these interests, and recommend similar products to them. Collaborative filtering algorithms are mainly divided into two categories: item-based collaborative filtering algorithms, which recommend items similar to the user's previous likes; and user-based collaborative filtering algorithms, which recommend items liked by users with similar interests.

[0088] In S103, feature images are generated based on the similarity between the first user and multiple second users. In one example, the similarity matrix corresponding to the similarity between the first user and multiple second users is transformed into a feature similarity image. This approach can convert abstract user similarity into an image, and then use the advantageous algorithms of image solutions to process the feature images.

[0089] In S104, the feature images between the first user and each of the second users are input into the image classification model to obtain target information. The target information includes user group information for the first user and each of the second users, indicating whether the first user and the second users belong to the same user group. In one example, the user-dimensional feature images obtained through dimensional feature image processing are used as input, and a collaborative filtering-based generative adversarial network (CFGAN) framework is used to perform unsupervised classification of the images. Since each image represents a user, and each pixel in the image represents the feature similarity value between users, users in the same category are a set of users with high similarity. For example, feature image 1 (i.e., the feature image between user A and users B, C, and D), feature image 2 (i.e., the feature image between user B and users A, C, and D), feature image 3 (i.e., the feature image between user C and users A, B, and D), and feature image 4 (i.e., the feature image between user D and users A, B, and C) are input into the image classification model to obtain the feature image classification results. The user group information in the target information indicates that user A and user B belong to the same user group, and user C and user D belong to the same user group. It should be noted here that since the first user can be any one of the G users, and the feature image is the feature image between the first user and each of the second users, it can be understood that there are at most G feature images. The above G feature images are input into the image classification model for classification to obtain the user group classification result.

[0090] To more accurately represent the similarity between the first user and other second users on each feature, such as Figure 2 As shown, as an example, S103 may include:

[0091] S1031: Based on the similarity between the first user and multiple second users, generate a similarity matrix between the first user and each second user. The similarity matrix includes the similarity between the first user and each second user.

[0092] S1032: Generate feature images between the first user and each of the second users based on the similarity matrix between the first user and the multiple second users.

[0093] Based on the similarity between the first user and multiple second users, a similarity matrix between the first user and each second user is generated. Then, based on the similarity matrix between the first user and multiple second users, a feature image between the first user and each second user is generated, which can more accurately express the similarity between the first user and other second users on each feature.

[0094] In S1031, based on the similarity between the first user and multiple second users, a similarity matrix is ​​generated between the first user and each of the second users. The similarity matrix includes the similarity between the first user and each of the second users. In one example, such as... Figure 3 As shown, a feature similarity matrix is ​​generated for each user based on the similarity values ​​of any two users on each feature.

[0095] In S1032, feature images between the first user and each of the second users are generated based on the similarity matrix between the first user and the multiple second users.

[0096] To more accurately separate key user features and improve the accuracy of model algorithms, such as Figure 4 As shown, as an example, the user characteristics of each of the G users include user sub-features in k dimensions; correspondingly, S1032 may include:

[0097] S10321: The similarity matrix between the first user and multiple second users is split into k similarity sub-matrices corresponding to k dimensions. Each similarity sub-matrix includes the similarity calculated based on the user sub-features in its corresponding dimension.

[0098] S10322: Generate feature images between the first user and each of the second users based on k similarity sub-matrices.

[0099] By splitting the similarity matrix between the first user and multiple second users into k similarity sub-matrices corresponding to k dimensions, and then generating feature images between the first user and each second user based on the k similarity sub-matrices, the key features of users can be separated more accurately, and the accuracy of the model algorithm can be improved.

[0100] In S10321, the similarity matrix between the first user and multiple second users is split into k similarity sub-matrices corresponding to k dimensions. Each similarity sub-matrix includes the similarity calculated based on the user sub-features in its corresponding dimension. In one example, the user features of each of the G users include user sub-features in k dimensions. The similarity matrix between the first user and multiple second users is split into k similarity sub-matrices, where each similarity sub-matrix includes the similarity corresponding to some user features between the first user and multiple second users. For example, if there are 9 user features in the above similarity matrix, these 9 user features are split into 3 parts according to 3 dimensions: features 1, 2, 3 and features 4, 5, 6 and features 7, 8, 9. Based on the user features after splitting, the above similarity matrix is ​​split into three similarity sub-matrices. The first similarity sub-matrix includes the similarity between the first user and each second user on features 1, 2, and 3. The second similarity sub-matrix includes the similarity between the first user and each second user on features 4, 5, and 6. The third similarity sub-matrix includes the similarity between the first user and each second user on features 7, 8, and 9.

[0101] In one example, the user's feature similarity matrix is ​​first split into multiple similarity sub-matrices. These sub-matrices are then used to construct a multi-dimensional feature similarity matrix, and a multi-dimensional matrix is ​​output for each user, such as... Figure 5 As shown, the format is: Number of users * Number of features * Feature dimensions. Here, the number of users refers to the number of other users being compared; the number of features refers to the number of features under each dimension; and the feature dimensions are the multiple dimensions that are used to break down all user features as needed. For example, basic user information can be divided into a basic profile dimension; cost-related information such as consumption information and home data traffic can be divided into a user consumption and subscription dimension; and resource usage data in home data traffic can be divided into a user resource dimension.

[0102] For example, in a dataset of 100 users across 3 dimensions (10 features per dimension), the result is as follows: the first element of the first matrix represents the similarity score between user 1 and user 2 in the first feature of dimension 1. This process is repeated to generate a similarity matrix between user 1 and the other 99 users. Similarly, data from the other two dimensions is generated. Finally, a multi-dimensional similarity matrix is ​​generated between user 1 and the other 99 users across 3 dimensions. The similarity calculation result for all 100 users is a multi-dimensional similarity matrix for each of the 100 users across 3 dimensions and 10 features per dimension, compared to the other 99 users. The final result is a 3D matrix with n-1 other users (n is the number of users in the sample set, n-1 for comparison with other users), m features (the number of features in each dimension), such as 3 dimensions (basic information dimension, user consumption and ordering dimension, user resource dimension).

[0103] In S10322, feature images between the first user and each of the second users are generated based on k similarity sub-matrices. In one example, the optimal sub-matrix can be selected from the k similarity sub-matrices, and the feature images between the first user and each of the second users can be generated based on the optimal sub-matrix. Alternatively, the k similarity sub-matrices can be combined to generate feature images between the first user and each of the second users.

[0104] To transform abstract user similarity into images and improve the accuracy and efficiency of identifying target user groups, such as... Figure 6 As shown, as an example, each of the k similarity submatrices includes N×M similarities, where N and M are positive integers; correspondingly, S10322 can include:

[0105] S103221: Based on the N×M similarities in each similarity submatrix, determine the N×M pixel gray levels in the color channels corresponding to the similarity submatrix, and obtain the N×M pixel gray levels in k color channels. Different similarity submatrixes correspond to different color channels.

[0106] S103222: Generate a feature image containing N×M pixels based on the gray levels of N×M pixels in k color channels. The pixel value of each pixel in the feature image is obtained by fusing the gray levels of the pixels in its corresponding k color channels.

[0107] By determining the N×M pixel grayscale values ​​in the color channels corresponding to each similarity submatrix based on the N×M similarity values ​​in each submatrix, N×M pixel grayscale values ​​are obtained in k color channels. Then, based on these N×M pixel grayscale values ​​in the k color channels, a feature image containing N×M pixels is generated. This method can transform abstract user similarity into an image, improving the accuracy and efficiency of identifying target user groups.

[0108] In S103221, based on the N×M similarities in each similarity submatrix, the N×M pixel gray levels on the color channels corresponding to the similarity submatrix are determined, resulting in N×M pixel gray levels on k color channels. Different similarity submatrixes correspond to different color channels.

[0109] In one example, the process of transforming the abstract user similarity matrix into an image is similar to the red, green, and blue three-pipeline pixel matrix in image processing. First, each element (i.e., the similarity value) in each dimension matrix is ​​binned according to a specific rule for all users. The rule can be the range (or other rules, such as normalization, interquartile range, etc.). The range (maximum value - minimum value) is divided into 256 intervals, which correspond to 0 to 255 intervals from smallest to largest, forming 256 pixel gray levels from light to dark.

[0110] In step S103222, a feature image comprising N×M pixels is generated based on the N×M pixel grayscale values ​​across k color channels. The pixel value of each pixel in the feature image is obtained by fusing the pixel grayscale values ​​of its corresponding k color channels. In one example, the pixel value of each pixel is obtained by fusing the pixel grayscale values ​​of its corresponding k color channels, and the positional relationship of the pixel grayscale values ​​across the k color channels is corresponding. For example, if the target pixel grayscale value on the red channel is (1, 10) (i.e., the tenth column of the first row), then the target pixel grayscale value on the green and blue channels is also (1, 10). Based on the three target pixel grayscale values ​​at position (1, 10) across the three color channels, the pixel value of the target pixel at position (1, 10) in the feature image is obtained by fusing these values.

[0111] In one example, such as Figure 7 As shown, the red, green, and blue channels of the image represent the three business dimensions of the user. Each pixel in each channel represents the value of 256 bins, which calculate the similarity between a user's feature and the corresponding features of other users. Since each user has such a multi-dimensional matrix, each user corresponds to a dimensional feature image.

[0112] In another example, such as Figure 8 As shown, depending on the specific needs of different business scenarios, features can also be processed into more dimensions, such as extending features to location trajectory, social circles, family circles, etc. Here, we take four dimensions as an example, forming a four-dimensional matrix. Referring to a PNG image, we add transparency pipes to create a four-channel image.

[0113] For each user, an image is generated based on the similarity of multi-dimensional features with other users. The image sets of all users can be used as input, and an image classification algorithm is used to train the model to classify the images of multiple users based on their similarity.

[0114] To improve the accuracy of converting abstract user similarity into images, such as Figure 9 As shown, as an example, S10321 may include:

[0115] S103211: Obtain user sub-features across k dimensions;

[0116] S103212: Based on the user sub-features in k dimensions, obtain the similarity corresponding to the user sub-features in k dimensions in the similarity matrix between the first user and each second user, and obtain k similarity sub-matrices.

[0117] By obtaining user sub-features in k dimensions, and then, based on these k-dimensional user sub-features, extracting the similarity scores corresponding to the user sub-features in each of the k dimensions from the similarity matrix between the first user and each second user, k similarity sub-matrices are obtained. This improves the accuracy of converting abstract user similarity into images.

[0118] In S103211, user sub-features are obtained across k dimensions. For example, from 9 user features, features are obtained that are split into 3 parts based on 3 dimensions: features 1, 2, 3, features 4, 5, 6, and features 7, 8, 9.

[0119] In S103212, based on the user sub-features in k dimensions, the similarity scores corresponding to the user sub-features in each of the k dimensions are obtained from the similarity matrix between the first user and each of the second users, resulting in k similarity sub-matrices. For example, based on features 1, 2, 3, 4, 5, 6, 7, 8, 9, the similarity scores between the first user and each of the second users on features 1, 2, 3, 4, 5, 6, 7, 8, 9 are obtained from the similarity matrix.

[0120] To transform abstract user similarity into images and improve the accuracy and efficiency of identifying target user groups, such as... Figure 10 As shown, as an example, S103 may include:

[0121] S1033: Based on the similarity between the first user and multiple second users, determine the pixel value corresponding to each similarity, wherein the pixel value corresponding to different similarities is different;

[0122] S1034: Generate feature images between the first user and each second user based on the pixel values ​​of the pixels corresponding to each similarity.

[0123] By calculating the similarity between a first user and multiple second users, the pixel value corresponding to each similarity level is determined. Based on these pixel values, a feature image is generated between the first user and each second user. This method transforms abstract user similarity into an image, improving the accuracy and efficiency of identifying target user groups.

[0124] In S1033, based on the similarity between the first user and multiple second users, the pixel value corresponding to each similarity level is determined, where different similarities correspond to different pixel values. In one example, the pixel grayscale value corresponding to each similarity level is determined based on the similarity matrix between the first user and multiple second users.

[0125] In S1034, feature images between the first user and each of the second users are generated based on the pixel values ​​of the pixels corresponding to each similarity level. In one example, feature images between the first user and each of the second users are generated based on the pixel grayscale values ​​of the pixels corresponding to each similarity level. It can be understood that the feature images obtained from the pixel grayscale values ​​generated based on the above similarity matrix can be black and white images composed of pixels with different grayscale values.

[0126] To improve the accuracy of outputting target user groups through collaborative filtering-based generative adversarial networks, such as... Figure 11 As shown, as an example, before S104, it may also include:

[0127] S1101: Obtain training image samples, the training image samples include historical feature images and labels corresponding to the historical feature images. The labels are used to indicate whether the first historical user and the second historical user belong to the same user group. The historical feature images are the feature images between the first historical user and the multiple second historical users among the multiple historical users. The multiple second historical users are the historical users other than the first historical user among the multiple historical users.

[0128] S1102: Input the training image samples into the image classification model to be trained, and train the image classification model.

[0129] By inputting the acquired training image samples into the image classification model to be trained, an image classification model is obtained. This can improve the accuracy of outputting target user groups through generative adversarial networks based on collaborative filtering.

[0130] In S1101, training image samples are obtained. The training image samples include historical feature images and labels corresponding to the historical feature images. The labels are used to indicate whether the first historical user and the second historical user belong to the same user group. The historical feature images are the feature images between the first historical user and the multiple second historical users among the multiple historical users. The multiple second historical users are the historical users other than the first historical user among the multiple historical users.

[0131] In step S1102, training image samples are input into the image classification model to be trained, resulting in an image classification model. In one example, the image classification model includes a generative adversarial network model based on a collaborative filtering method.

[0132] In one example, before S104, it may also include:

[0133] The training image samples are input into the image classification model to be trained. The target information of the training image samples is obtained by the image classification model to be trained based on the historical feature images. The target information of the training image samples includes the user group information of the first user and each second user. The training image samples include the historical feature images of the first user and each second user, as well as the sample label information corresponding to the historical feature images. The sample label information includes the historical user group information of the first user and each second user.

[0134] The loss function value of the image classification model to be trained is determined based on the target information and sample label information of the training image samples.

[0135] The image classification model is iteratively trained based on the training image samples until the loss function value and / or the number of training iterations of the image classification model meet the preset stopping condition, thus obtaining the image classification model.

[0136] In one example, the specific process of iteratively training the image classification model to be trained is as follows:

[0137] The user-dimensional feature images, obtained through dimensional feature visualization, are used as input, and an unsupervised classification of the images is performed using a collaborative filtering-based generative adversarial network (CFGAN) model framework. Since each image represents a user, and each pixel in the image represents the feature similarity value between users, users in the same category are a set of users with high similarity.

[0138] A Generative Adversarial Network (GAN) consists of a generative model G and a discriminative model D, which compete in adversarial mode: D attempts to distinguish between real positive samples and positive samples faked by G, while G attempts to deceive D. The GAN training is complete when the adversarial forces are relatively balanced. The cost function of the GAN is shown in Equation (1):

[0139]

[0140] Where m represents the samples in the training set, U represents the user set, un represents the nth user, and φ and θ represent the parameters of the discriminant model D and the generative model G, respectively.

[0141] The training of the discrimination model aims to maximize the probability of correctly distinguishing between real positive samples and G-forged positive samples, as shown in formula (2):

[0142]

[0143] Where m+ and m- represent positive and negative samples, respectively, and mg represents a positive sample faked by the generative model G. In contrast to the discriminative model D, the generative model G aims to minimize the correct discrimination probability of the discriminative model D, i.e., to deceive the discriminative model.

[0144] As shown in formula (3): G draws negative samples from the sample pool according to the relevant probability distribution, so the parameter optimization of the generative model is as shown in formula (4), where M represents the sample set:

[0145]

[0146]

[0147] Since the samples are discrete during sampling, the optimal solution cannot be obtained directly through gradient descent as shown in the above formula. Instead, the policy gradient method based on reinforcement learning is used for approximate solution. K is the size of the list of sample pairs to be judged, as shown in formula (5):

[0148]

[0149] The process of using this formula to find the optimal solution for the feature dataset of sample users is the model generation process. During this process, the model is evaluated using TOP N ranking precision and recall, as well as the F1 score. The more robust Mean Absolute Percent Error (MAPER) is also used to evaluate the model output, reducing the impact of absolute errors caused by individual outliers. The training and evaluation process is iterated until the model output gets closer to the algorithm's optimal solution, ultimately outputting user feature image classifications, that is, classifying users according to image feature similarity.

[0150] To better identify the target customer group for the ordered products, such as Figure 12 As shown, as an example, before S101, it may also include:

[0151] S1201: Obtain user characteristic data and user product order data of the first user among G users, as well as user characteristic data and user product order data of multiple second users;

[0152] S1202: Based on the user feature data and user product order data of the first user among G users, and the user feature data and user product order data of multiple second users, generate the user feature and order matrix of the first user among G users and the user feature and order matrix of multiple second users. The user feature and order matrix includes multiple user features and multiple user ordered products.

[0153] Accordingly, S101 may include:

[0154] S1203: From the user characteristics and ordering matrix of the first user among G users and the user characteristics and ordering matrix of multiple second users, obtain the user characteristics of the first user among G users and the user characteristics and ordering matrix of multiple second users.

[0155] By acquiring user characteristic data and user product order data of the first user among G users, as well as user characteristic data and user product order data of multiple second users, a user characteristic and order matrix of the first user and a user characteristic and order matrix of multiple second users among G users can be generated, which can better output the target customer group corresponding to the ordered products.

[0156] In step S1201, user characteristic data and user product order data of the first user among G users, as well as user characteristic data and user product order data of multiple second users, are obtained. In one example, user business characteristic data and user product order data are extracted from the business analysis data.

[0157] In S1202, based on the user characteristic data and user product order data of the first user among G users, and the user characteristic data and user product order data of multiple second users, a user characteristic and order matrix of the first user and a user characteristic and order matrix of multiple second users are generated. The user characteristic and order matrix includes multiple user characteristics and multiple user ordered products.

[0158] In one example, user business characteristic data and user product order data are extracted from the data analytics and combined to form a user characteristic and product order matrix. User characteristic data includes different categories such as basic information, consumption information, communication information, family information, and traffic information. These are associated with user product order information by user and integrated into a user characteristic and product order matrix, as shown in Table 1. Each user contains n user characteristics and m products. Finally, a data sample is generated.

[0159] Table 1: User Characteristics and Product Ordering Matrix

[0160] 1 Mobile phone number String Mobile phone number 1+1 User feature 1 Type Explanation …… …… Type Explanation 1+n User feature n Type Explanation 1+n+1 User product 1 Type Explanation …… …… Type Explanation 1 + n + m User product m Type Explanation

[0161] In S1203, the user characteristics of the first user and the user characteristics of the multiple second users are obtained from the user characteristics and ordering matrix of the first user among the G users and the user characteristics and ordering matrix of the multiple second users.

[0162] In order to better deliver ordered products to the corresponding target customer groups, such as Figure 13 As shown, as an example, after S104, it may also include:

[0163] S1301: Based on the target information, the user characteristics and ordering matrix of the first user among G users, and the user characteristics and ordering matrix of multiple second users, determine the target user ordering products of the target user group. The target user group includes the first user and the second users whose user group information indicates that the first user and the second user belong to the same user group.

[0164] Based on user characteristics and the ordering matrix, the target user group can be identified as having the desired product ordering status, which enables better delivery of ordered products to the corresponding target customer group.

[0165] In step S1301, based on the target information, the user characteristics and ordering matrix of the first user among G users, and the user characteristics and ordering matrices of multiple second users, the target user group's target user ordering products are determined. The target user group includes the first user and the second users whose user group information indicates that the first user and the second user belong to the same user group. For example, after obtaining the target user group through the above steps, the users in the target user group have a high degree of similarity. Then, by using the user characteristics and product information in the product ordering matrix, the top N products within the same category are summarized (N can be set according to actual business needs). These are then matched with the products that the business department needs to promote, thus obtaining the user preference classification for the promoted products. This completes the matching of promoted products with the customer group, achieving recommendation based on a similarity-based feature image algorithm.

[0166] To more accurately separate key user features, such as Figure 14 As shown, as an example, before S102, it may also include:

[0167] S1401: Perform feature engineering on the user features of the first user and the user features of each second user. Feature engineering includes feature understanding, feature cleaning and enhancement, feature construction and feature selection.

[0168] By performing feature engineering on the user characteristics of the first user and the user characteristics of each second user, key user characteristics can be separated more accurately.

[0169] In S1401, feature engineering is performed on the user characteristics of the first user and the user characteristics of each second user. Feature engineering includes feature understanding, feature cleaning and enhancement, feature construction, and feature selection. In one example, feature understanding includes structured data identification and qualitative and quantitative data discrimination; feature cleaning and enhancement includes EDA exploration, normalization, and standardization; feature construction includes constructing mean, fluctuation, and chain ratio analysis data; and feature selection includes Pearson coefficient, hypothesis testing, and chi-square test. The specific feature engineering process is as follows: Figure 15 As shown.

[0170] The user group matching process in the above technical solution will be specifically described below using an embodiment, such as... Figure 16 As shown in the diagram, the specific steps are as follows:

[0171] S1601: User data extraction: User business attribute data, user product order data;

[0172] S1602: Feature Engineering: Feature understanding, feature cleaning and enhancement, feature construction, feature selection, etc.;

[0173] S1603: Similarity Calculation: Calculate the similarity between users using the coordinated filtering method;

[0174] S1604: Dimensional Feature Visualization: Constructing an abstract graph of user features through multidimensional user features;

[0175] S1605: Model Training: The model is trained using a generative adversarial network algorithm based on coordination filtering;

[0176] S1606: User Group Output: Generate a user group by combining the model output with the related marketing campaign products.

[0177] The solutions provided in the above embodiments can be applied to the following scenarios: In practical applications of business promotion, there are many scenarios that require planning and setting target customer groups. According to the method provided in this application, the existing pre-selection, preset customer group selection mode can be improved into an intelligent customer group matching mode that requires no manual intervention. After selecting a marketing campaign product, a suitable target customer group can be generated simultaneously. This can be applied to many business scenarios, such as new campaign planning to better develop promotions targeting specific customer groups, analysis and results summary of a marketing campaign's customer group, and assessment of the impact of outdated pre-terminated campaign planning on the customer group affected by the campaign's termination.

[0178] Customer selection in street vending marketing is a typical application scenario among many businesses that require customer segmentation. Sales personnel select marketing campaign products on a support platform on their personal PCs, and the system automatically generates matching customer groups (currently, this requires either using a tag library or importing from the front end). Subsequently, the system matches these customer groups with information such as their locations to determine the specific users to whom SMS messages will be sent for promotional activities.

[0179] The beneficial effects of the solution provided in the above embodiments are as follows: it can significantly improve existing business processes and efficiently and accurately locate the target customer groups needed for marketing activities. It greatly reduces personnel and time costs, saves human resources, and provides rapid and accurate support for business scenarios.

[0180] The solution provided in this application utilizes pre-designed and periodically extracted key customer features and product order matrices as model input data. It employs the approach of transforming the user feature matrix into user feature similarity images, combined with generative adversarial network algorithms adept at image processing, for modeling. The output is an accurate target customer group corresponding to the marketing campaign. This overcomes the shortcomings of existing marketing product promotion methods, such as low accuracy of target customer groups, time-consuming target customer group selection and analysis, and high resource investment. It can provide faster and more accurate support for marketing campaigns, improving campaign promotion effectiveness and future development potential.

[0181] The above describes a specific implementation of a user group matching method provided in this application. Based on the user group matching method provided in the above embodiments, this application also provides a specific implementation of a user group matching device, as described in the following embodiments.

[0182] like Figure 17 As shown in the embodiment of this application, a user group matching device 1700 is provided, which includes:

[0183] The acquisition module 1701 is used to acquire the user characteristics of the first user and the user characteristics of multiple second users among G users, where G is a positive integer and the multiple second users are users other than the first user among the G users.

[0184] The calculation module 1702 is used to calculate the similarity between the first user and each second user based on the user characteristics of the first user and the user characteristics of each second user, using a collaborative filtering method.

[0185] The generation module 1703 is used to generate feature images between the first user and each of the second users based on the similarity between the first user and multiple second users;

[0186] The input module 1704 is used to input the feature images between the first user and each of the second users into the image classification model to obtain target information. The target information includes user group information of the first user and each of the second users. The user group information is used to indicate whether the first user and the second user belong to the same user group.

[0187] In the user group matching device 1700 provided in this embodiment, the acquisition module 1701 acquires the user features of the first user and the user features of multiple second users from among G users. The calculation module 1702 calculates the similarity between the first user and each of the second users based on the collaborative filtering method, according to the user features of the first user and the user features of each of the second users. The generation module 1703 generates a feature image between the first user and each of the second users based on the similarity between the first user and the multiple second users. The input module 1704 inputs the feature image between the first user and each of the second users into an image classification model to obtain target information.

[0188] Thus, based on the collaborative filtering method, the similarity of user features between the first user and the second user is calculated, generating feature images between the first user and each second user. This transforms the non-image-based similarity problem into an image-based problem, thereby improving the efficiency of outputting the target user group. Furthermore, combining the feature images with an image classification model for recognition and classification yields target information indicating the target user group, improving the accuracy of outputting the target user group.

[0189] As another embodiment of this application, in order to more accurately express the similarity of the first user and other second users on each feature, the above-mentioned generation module 1703 may include:

[0190] The first generation unit 17031 is used to generate a similarity matrix between the first user and each of the second users based on the similarity between the first user and multiple second users. The similarity matrix includes the similarity between the first user and each of the second users.

[0191] The second generation unit 17032 is used to generate feature images between the first user and each of the second users based on the similarity matrix between the first user and multiple second users.

[0192] As another embodiment of this application, in order to more accurately separate user key features and improve the accuracy of the model algorithm, the first generation unit 17031 may include:

[0193] The splitting subunit 170311 is used to split the similarity matrix between the first user and multiple second users into k similarity submatrices corresponding to k dimensions. Each similarity submatrice includes the similarity calculated based on the user sub-features on its corresponding dimension.

[0194] Generating subunit 170312 is used to generate feature images between the first user and each of the second users based on k similarity submatrices.

[0195] As another embodiment of this application, in order to transform abstract user similarity into images and improve the accuracy and efficiency of determining target user groups, the above-mentioned generation subunit 170312 can be specifically used for:

[0196] Based on the N×M similarities in each similarity submatrix, determine the N×M pixel gray levels in the color channels corresponding to the similarity submatrix, and obtain the N×M pixel gray levels in k color channels. Different similarity submatrixes correspond to different color channels.

[0197] Based on the N×M pixel grayscale values ​​in k color channels, a feature image containing N×M pixels is generated. The pixel value of each pixel in the feature image is obtained by fusing the pixel grayscale values ​​in its corresponding k color channels.

[0198] As another embodiment of this application, in order to improve the accuracy of converting abstract user similarity into images, the above-mentioned splitting subunit 170311 can be specifically used for:

[0199] Obtain user sub-features across k dimensions;

[0200] Based on the user sub-features in k dimensions, the similarity scores corresponding to the user sub-features in k dimensions are obtained from the similarity matrix between the first user and each second user, resulting in k similarity sub-matrices.

[0201] As another embodiment of this application, in order to convert abstract user similarity into images and improve the accuracy and efficiency of identifying target user groups, the above-mentioned generation module 1703 may include:

[0202] The determining unit 17033 is used to determine the pixel value of the pixel corresponding to each similarity based on the similarity between the first user and multiple second users, wherein the pixel value corresponding to different similarities is different;

[0203] The third generation unit 17034 is used to generate feature images between the first user and each second user based on the pixel values ​​of the pixels corresponding to each similarity.

[0204] As another embodiment of this application, in order to improve the accuracy of outputting target user groups through collaborative filtering-based generative adversarial networks, the user group matching device 1700 may further include:

[0205] The second acquisition module 1705 is used to acquire training image samples. The training image samples include historical feature images and labels corresponding to the historical feature images. The labels are used to indicate whether the first historical user and the second historical user belong to the same user group. The historical feature images are feature images between the first historical user and the multiple second historical users among multiple historical users. The multiple second historical users are historical users other than the first historical user among multiple historical users.

[0206] The model training module 1706 is used to input training image samples into the image classification model to be trained, and to train the image classification model.

[0207] As another embodiment of this application, in order to better output the target customer group corresponding to the ordered products, the user group matching device 1700 may further include:

[0208] The third acquisition module 1706 is used to acquire user characteristic data and user product order data of the first user among G users, as well as user characteristic data and user product order data of multiple second users.

[0209] The second generation module 1707 is used to generate a user feature and ordering matrix for the first user and a user feature and ordering matrix for the second user among the G users, based on the user feature data and user product ordering data of the first user among the G users, and the user feature data and user product ordering data of the multiple second users. The user feature and ordering matrix includes multiple user features and multiple user ordered products.

[0210] The aforementioned acquisition module 1701 can be specifically used for:

[0211] From the user characteristics and ordering matrix of the first user among G users, and the user characteristics and ordering matrix of multiple second users, obtain the user characteristics of the first user among G users and the user characteristics of multiple second users among G users.

[0212] As another embodiment of this application, in order to better deliver ordered products to the corresponding target customer groups, the user group matching device 1700 may further include:

[0213] Product determination module 1708 is used to determine the target user ordering product of the target user group based on target information, the user characteristics and ordering matrix of the first user among G users and the user characteristics and ordering matrix of multiple second users. The target user group includes the first user and the second user whose user group information indicates that the first user and the second user belong to the same user group.

[0214] As another embodiment of this application, in order to more accurately separate key user features, the user group matching device 1700 may further include:

[0215] The feature engineering module 1709 is used to perform feature engineering on the user features of the first user and the user features of each second user. Feature engineering includes feature understanding, feature cleaning and enhancement, feature construction and feature selection.

[0216] Based on the user group matching method and apparatus provided in the above embodiments, this application also provides an electronic device 1800, such as... Figure 18 As shown:

[0217] It includes a processor 1801, a memory 1802, and a computer program stored in the memory 1802 and executable on the processor 1801. When the computer program is executed by the processor 1801, it implements the various processes of the above-described user group matching method embodiments and achieves the same technical effect.

[0218] Specifically, the processor 1801 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0219] Memory 1802 may include mass storage for data or instructions. For example, and not limitingly, memory 1802 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 1802 may include removable or non-removable (or fixed) media. Where appropriate, memory 1802 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 1802 is non-volatile solid-state memory.

[0220] In certain embodiments, the memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Thus, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of this application.

[0221] The processor 1801 implements any of the user group matching methods in the above embodiments by reading and executing computer program instructions stored in the memory 1802.

[0222] In one example, the electronic device may also include a communication interface 1803 and a bus 1810. As an example, such as... Figure 18 As shown, the processor 1801, memory 1802, and communication interface 1803 are connected through bus 1810 and complete communication with each other.

[0223] The communication interface 1803 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0224] Bus 1810 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 1810 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0225] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the user group matching method embodiments described above and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0226] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0227] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0228] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0229] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this application. It should be understood that each block in 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, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0230] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A user group matching method, characterized in that, include: Obtain the user features of the first user and the user features of multiple second users from a group of G users, where G is a positive integer and the multiple second users are users other than the first user from the group of G users; the user features of each user in the group of G users include user sub-features in k dimensions; Based on the collaborative filtering method, the similarity between the first user and each of the second users is calculated according to the user characteristics of the first user and the user characteristics of each of the second users. Based on the similarity between the first user and multiple second users, a similarity matrix is ​​generated between the first user and each of the second users; The similarity matrix includes N×M similarity values, where N and M are positive integers; Based on the similarity matrix between the first user and the plurality of second users, a feature image is generated between the first user and each of the second users, including: splitting the similarity matrix between the first user and the plurality of second users into k similarity sub-matrices corresponding to the k dimensions, each of the similarity sub-matrices including similarity calculated based on user sub-features in its corresponding dimension; determining N×M pixel grayscale values ​​in the color channel corresponding to each of the similarity sub-matrices based on the N×M similarity values ​​in each of the similarity sub-matrices, thereby obtaining N×M pixel grayscale values ​​in the plurality of color channels, with different similarity sub-matrices corresponding to different color channels; Based on the N×M pixel grayscale values ​​on the plurality of color channels, a feature image comprising N×M pixels is generated, wherein the pixel value of each pixel in the feature image is obtained by fusing the pixel grayscale values ​​on the plurality of color channels corresponding to it; The feature images of the first user and each of the second users are input into an image classification model to obtain target information. The target information includes user group information of the first user and each of the second users. The user group information is used to indicate whether the first user and the second user belong to the same user group.

2. The method according to claim 1, characterized in that, The step of generating feature images between the first user and each of the second users based on the similarity matrix between the first user and the plurality of second users includes: Based on the k similarity sub-matrices, feature images between the first user and each of the second users are generated.

3. The method according to claim 2, characterized in that, The step of splitting the similarity matrix between the first user and multiple second users into k similarity sub-matrices corresponding to the k dimensions includes: Obtain user sub-features across the k dimensions; Based on the user sub-features in the k dimensions, the similarity corresponding to the user sub-features in the k dimensions is obtained in the similarity matrix between the first user and each of the second users, resulting in k similarity sub-matrices.

4. The method according to claim 1, characterized in that, The step of generating feature images between the first user and each of the second users based on the similarity between the first user and multiple second users includes: Based on the similarity between the first user and multiple second users, the pixel value corresponding to each similarity is determined, wherein different similarities correspond to different pixel values; Based on the pixel values ​​of the pixels corresponding to each of the similarities, a feature image between the first user and each of the second users is generated.

5. The method according to claim 1, characterized in that, Before inputting the feature images between the first user and each of the second users into the image classification model to obtain target information, the method further includes: Acquire training image samples, which include historical feature images and labels corresponding to the historical feature images. The labels are used to indicate whether a first historical user and a second historical user belong to the same user group. The historical feature images are feature images between the first historical user and multiple second historical users among multiple historical users. The multiple second historical users are historical users other than the first historical user among the multiple historical users. The training image samples are input into the image classification model to be trained, and the image classification model is obtained through training.

6. The method according to claim 1, characterized in that, Before obtaining the user characteristics of the first user and the user characteristics of multiple second users among the G users, the process also includes: Obtain user characteristic data and user product order data of the first user among G users, as well as user characteristic data and user product order data of multiple second users; Based on the user characteristic data and user product order data of the first user among the G users, and the user characteristic data and user product order data of multiple second users, a user characteristic and order matrix of the first user and a user characteristic and order matrix of multiple second users are generated among the G users. The user characteristic and order matrix includes multiple user characteristics and multiple user ordered products. The process of obtaining the user characteristics of the first user and the user characteristics of multiple second users among G users includes: From the user characteristics and ordering matrix of the first user among the G users and the user characteristics and ordering matrix of the multiple second users, obtain the user characteristics of the first user among the G users and the user characteristics of the multiple second users.

7. The method according to claim 6, characterized in that, After inputting the feature images between the first user and each of the second users into the image classification model to obtain the target information, the method further includes: Based on the target information, the user characteristics and ordering matrix of the first user among the G users, and the user characteristics and ordering matrix of multiple second users, the target user group's target user ordering products are determined. The target user group includes the first user and the second user whose user group information indicates that the first user and the second user belong to the same user group.

8. The method according to claim 1, characterized in that, Before calculating the similarity between the first user and each of the second users based on the user characteristics of the first user and the user characteristics of each of the second users, the method further includes: Feature engineering is performed on the user characteristics of the first user and the user characteristics of each of the second users. The feature engineering includes feature understanding, feature cleaning and enhancement, feature construction and feature selection.

9. The method according to claim 1, characterized in that, The image classification model includes a generative adversarial network model based on collaborative filtering.

10. A user group matching device for performing the method as described in claim 1, characterized in that, include: The acquisition module is used to acquire the user characteristics of the first user and the user characteristics of multiple second users among G users, where G is a positive integer and the multiple second users are users other than the first user among the G users. The calculation module is used to calculate the similarity between the first user and each of the second users based on the user characteristics of the first user and the user characteristics of each of the second users, using a collaborative filtering method. The generation module is used to generate feature images between the first user and each of the second users based on the similarity between the first user and the plurality of second users; The input module is used to input the feature images between the first user and each of the second users into the image classification model to obtain target information. The target information includes user group information of the first user and each of the second users, and the user group information is used to indicate whether the first user and the second user belong to the same user group.

11. An electronic device, characterized in that, The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the user group matching method as described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the user group matching method as described in any one of claims 1-9.

13. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device performs the user group matching method as described in any one of claims 1-9.

Images