Similar group expansion method and device

Abstract

The present application discloses a similar population expansion method and device. A specific implementation of the method includes: inputting the portrait information, the first time period operation sequence and the second time period operation sequence of each user among the target users into a user representation model to obtain a user feature vector corresponding to each user, wherein the user representation model is used to predict the user's pending operation based on the user's portrait information, the first time period operation sequence and the second time period operation sequence; based on the similarity of the user feature vectors, determining multiple candidate users similar to the seed user among the target users; inputting the user feature vector of each user among the multiple candidate users and the operation sequence of the user up to the current preset time period into a recognition model to obtain at least one similar user, wherein the recognition model is trained with the seed user as a positive sample and the non-seed user as a negative sample. The present application improves the accuracy of determining the similar population of the seed user.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technology, and in particular to a method and device for expanding a similar group of people. Background Art

[0002] Look-alike (similar crowd expansion) is a technology that uses seed users to find similar crowds, which can improve the accuracy of crowd targeting. By expanding similar crowds, users can find the target crowd they want to find and improve the efficiency of crowd targeting. Currently, similar crowd expansion is generally performed through similarity-based methods or model-based methods. Summary of the invention

[0003] The embodiments of the present application provide a method and device for expanding a similar group of people.

[0004] In a first aspect, an embodiment of the present application provides a similar population expansion method, including: inputting the portrait information, the first time period operation sequence and the second time period operation sequence of each user among the target users into a user representation model to obtain a user feature vector corresponding to each user, wherein the user representation model is used to predict the user's upcoming operation based on the user's portrait information, the first time period operation sequence and the second time period operation sequence, and the user feature vector is the intermediate data obtained by the user representation model during the prediction process; based on the similarity of the user feature vectors, determining multiple candidate users among the target users who are similar to the seed user; inputting the user feature vector of each user among the multiple candidate users and the operation sequence of the user up to the current preset time period into a recognition model to obtain at least one similar user, wherein the recognition model is trained with seed users as positive samples and non-seed users as negative samples.

[0005] In some embodiments, the user representation model is trained through the following steps: taking the first time period operation sequence and the second time period operation sequence of the target user as input, determining the dimensional feature vectors of the items involved in the target user's operation information in different dimensions through a word vector model; using the dimensional feature vectors of the items involved in different dimensions as the initial weight parameters of the embedding layer in the initial user representation model; for each operation information in the first time period operation sequence of each target user, taking the user's portrait information, the operation information up to the operation information in the first time period operation sequence and the second time period operation sequence as input, and taking the next operation information of the operation information in the first time period operation sequence as the expected output, the user representation model is trained.

[0006] In some embodiments, the dimensions include item dimensions, brand dimensions to which the items belong, category dimensions to which the items belong, and store dimensions to which the items belong; the first time period operation sequence represents the user's operation information in the item dimension, and the second time period operation sequence represents the operation information in the brand dimension, category dimension, and store dimension; the embedding layer in the user representation model includes embedding layers corresponding to the portrait information, the first time period operation sequence, and the second time period operation sequence in sequence; and the above-mentioned use of the dimensional feature vectors of the items involved in different dimensions as the initial weight parameters of the embedding layer in the initial user representation model includes: determining the dimensional feature vectors of the items involved in the item dimension as the initial weight parameters of the embedding layer corresponding to the first time period operation sequence; determining the dimensional feature vectors of the items involved in the corresponding brand dimension, category dimension, and store dimension as the initial weight parameters of the embedding layer corresponding to the second time period operation sequence.

[0007] In some embodiments, during the training process of the user representation model, only the portrait information and the weight parameters of the embedding layer corresponding to the operation sequence of the first period are updated, and the weight parameters of the embedding layer corresponding to the operation sequence of the second period are kept unchanged.

[0008] In some embodiments, the above-mentioned inputting the portrait information, the first time period operation sequence and the second time period operation sequence of each target user into the user representation model to obtain the user feature vector corresponding to each user includes: for each target user, in the user representation model, processing the sequence information of each operation information in the first time period operation sequence, the second time period operation sequence and the second time period operation sequence of the user through a multi-head attention mechanism to obtain the intermediate data corresponding to the first time period operation sequence and the intermediate data corresponding to the second time period operation sequence; through the attention mechanism, fusing the intermediate data corresponding to the first time period operation sequence, the intermediate data corresponding to the second time period operation sequence and the intermediate data corresponding to the portrait information to obtain the user feature vector corresponding to the user.

[0009] In some embodiments, the dimensional feature vectors of the items in different dimensions obtained by the word vector model are used as the initial weight parameters of the embedding layer in the initial prediction model; the user feature vector of each user among the multiple candidate users and the operation sequence of the user up to the current preset time period are input into the recognition model to obtain at least one similar user, including: for each user among the multiple candidate users, through the embedding layer in the prediction model, processing the operation sequence of the user up to the current preset time period to obtain corresponding intermediate data; through the attention mechanism, the user feature vector of the user and the intermediate data corresponding to the operation sequence up to the current preset time period are fused to predict whether the user is a similar user.

[0010] In some embodiments, during the training process of the prediction model, the weight parameters of the embedding layer of the prediction model are kept from being updated.

[0011] In the second aspect, an embodiment of the present application provides a similar population expansion device, including: a first obtaining unit, configured to input the portrait information, the first time period operation sequence and the second time period operation sequence of each user among the target users into a user representation model, and obtain a user feature vector corresponding to each user, wherein the user representation model is used to predict the user's upcoming operation based on the user's portrait information, the first time period operation sequence and the second time period operation sequence, and the user feature vector is the intermediate data obtained by the user representation model in the prediction process; a determination unit, configured to determine multiple candidate users among the target users who are similar to the seed user based on the similarity of the user feature vectors; a second obtaining unit, configured to input the user feature vector of each user among the multiple candidate users and the operation sequence of the user up to the current preset time period into a recognition model, and obtain at least one similar user, wherein the recognition model is trained with seed users as positive samples and non-seed users as negative samples.

[0012] In some embodiments, the above-mentioned device also includes: a training unit, configured to train the user representation model through the following steps: taking the first time period operation sequence and the second time period operation sequence of the target user as input, determining the dimensional feature vectors of the items involved in the target user's operation information in different dimensions through the word vector model; using the dimensional feature vectors of the items involved in different dimensions as the initial weight parameters of the embedding layer in the initial user representation model; for each operation information in the first time period operation sequence of each user among the target users, taking the user's portrait information, the operation information up to the operation information in the first time period operation sequence and the second time period operation sequence as input, and taking the next operation information of the operation information in the first time period operation sequence as the expected output, training to obtain the user representation model.

[0013] In some embodiments, the dimensions include item dimension, brand dimension to which the item belongs, category dimension to which the item belongs, and store dimension to which the item belongs; the first time period operation sequence represents the user's operation information in the item dimension, and the second time period operation sequence represents the operation information in the brand dimension, category dimension, and store dimension; the embedding layer in the user representation model includes embedding layers corresponding to the portrait information, the first time period operation sequence, and the second time period operation sequence in sequence; and the training unit is further configured to: determine the dimensional feature vector of the item involved in the item dimension as the initial weight parameter of the embedding layer corresponding to the first time period operation sequence; determine the dimensional feature vector of the item involved in the corresponding brand dimension, category dimension, and store dimension as the initial weight parameter of the embedding layer corresponding to the second time period operation sequence.

[0014] In some embodiments, during the training process of the user representation model, only the portrait information and the weight parameters of the embedding layer corresponding to the operation sequence of the first period are updated, and the weight parameters of the embedding layer corresponding to the operation sequence of the second period are kept unchanged.

[0015] In some embodiments, the first obtaining unit is further configured to: for each target user, in the user representation model, process the sequence information of each operation information in the first time period operation sequence, the second time period operation sequence and the second time period operation sequence of the user through a multi-head attention mechanism to obtain the intermediate data corresponding to the first time period operation sequence and the intermediate data corresponding to the second time period operation sequence; through the attention mechanism, fuse the intermediate data corresponding to the first time period operation sequence, the intermediate data corresponding to the second time period operation sequence and the intermediate data corresponding to the portrait information to obtain the user feature vector corresponding to the user.

[0016] In some embodiments, the dimensional feature vectors of the items in different dimensions obtained by the word vector model are used as the initial weight parameters of the embedding layer in the initial prediction model; the second obtaining unit is further configured to: for each user among multiple candidate users, process the operation sequence of the user up to the current preset time period through the embedding layer in the prediction model to obtain corresponding intermediate data; through the attention mechanism, the user feature vector of the user and the intermediate data corresponding to the operation sequence up to the current preset time period are fused to predict whether the user is a similar user.

[0017] In some embodiments, during the training process of the prediction model, the weight parameters of the embedding layer of the prediction model are kept from being updated.

[0018] In a third aspect, an embodiment of the present application provides a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation manner of the first aspect is implemented.

[0019] In a fourth aspect, an embodiment of the present application provides an electronic device, comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by one or more processors, the one or more processors implement the method described in any implementation manner of the first aspect.

[0020] The similar population expansion method and device provided in the embodiment of the present application obtain a user feature vector corresponding to each user by inputting the portrait information, the first time period operation sequence and the second time period operation sequence of each user among the target users into a user representation model, wherein the user representation model is used to predict the user's upcoming operation based on the user's portrait information, the first time period operation sequence and the second time period operation sequence, and the user feature vector is the intermediate data obtained by the user representation model in the prediction process; based on the similarity of the user feature vectors, multiple candidate users similar to the seed user among the target users are determined; the user feature vector of each user among the multiple candidate users and the operation sequence of the user up to the current preset time period are input into a recognition model to obtain at least one similar user, wherein the recognition model is trained with seed users as positive samples and non-seed users as negative samples, thereby providing a similar population expansion method and improving the accuracy of determining the similar population of the seed user. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0022] Figure 1 is an exemplary system architecture diagram in which an embodiment of the present application may be applied;

[0023] Figure 2 is a flow chart of an embodiment of a similar population expansion method according to the present application;

[0024] Figure 3 A partial structural diagram of a user representation model according to an embodiment of the present application

[0025] Figure 4 is a schematic diagram of an application scenario of the similar group expansion method according to this embodiment;

[0026] Figure 5 is a flow chart of another embodiment of the similar population expansion method according to the present application;

[0027] Figure 6 is a structural diagram of an embodiment of a similar group expansion device according to the present application;

[0028] Figure 7 It is a structural diagram of a computer system suitable for implementing the embodiments of the present application. DETAILED DESCRIPTION

[0029] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It should also be noted that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0030] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0031] Figure 1 An exemplary architecture 100 to which the similarity group expansion method and apparatus of the present application can be applied is shown.

[0032] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The communication connection between the terminal devices 101, 102, 103 constitutes a topological network, and the network 104 is used to provide a medium for the communication link between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0033] Terminal devices 101, 102, 103 may be hardware devices or software that support network connection for data interaction and data processing. When terminal devices 101, 102, 103 are hardware, they may be various electronic devices that support network connection, information acquisition, interaction, display, processing and other functions, including but not limited to smart phones, tablet computers, e-book readers, laptop computers and desktop computers, etc. When terminal devices 101, 102, 103 are software, they may be installed in the electronic devices listed above. They may be implemented as multiple software or software modules for providing distributed services, or they may be implemented as a single software or software module. No specific limitation is made here.

[0034] The server 105 may be a server that provides various services, such as obtaining operation information sent by the target user through the terminal devices 101, 102, and 103, and determining a similar group of people to the seed user based on the target user's portrait information, the first time period operation sequence, and the second time period operation sequence. As an example, the server 105 may be a cloud server.

[0035] It should be noted that the server can be hardware or software. When the server is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or it can be implemented as a single server. When the server is software, it can be implemented as multiple software or software modules (for example, software or software modules used to provide distributed services), or it can be implemented as a single software or software module. No specific limitation is made here.

[0036] It should also be noted that the similar group expansion method provided in the embodiments of the present disclosure can be executed by a server, or by a terminal device, or by a server and a terminal device in cooperation with each other. Accordingly, the various parts (such as various units) included in the similar group expansion device can be all set in the server, or all set in the terminal device, or can be set in the server and the terminal device respectively.

[0037] It should be understood that Figure 1 The number of terminal devices, networks, and servers in the system is only illustrative. Depending on the implementation requirements, there may be any number of terminal devices, networks, and servers. When the electronic device on which the similarity crowd expansion method is running does not need to transmit data with other electronic devices, the system architecture may only include the electronic device (such as a server or terminal device) on which the similarity crowd expansion method is running.

[0038] Continue to refer Figure 2 , a process 200 of an embodiment of a similar group expansion method is shown, comprising the following steps:

[0039] Step 201 , input the portrait information, the first time period operation sequence and the second time period operation sequence of each target user into a user representation model to obtain a user feature vector corresponding to each user.

[0040] In this embodiment, the execution subject of the similar group expansion method (for example Figure 1 The server in the example may input the portrait information, the operation sequence of the first time period, and the operation sequence of the second time period of each target user into the user representation model to obtain the user feature vector corresponding to each user.

[0041] The target user may be a user of a similar group to be determined as a seed user. In this embodiment, the execution subject may predetermine all target users. The seed user, as a reference standard for determining whether the target user is a similar group, may be any predetermined user.

[0042] Taking the crowd targeting operation of the shopping platform for advertising as an example, the seed users may be the representative users who meet the advertising characteristics predetermined by the shopping platform, and the target users may be all users who use the above shopping platform.

[0043] The user's portrait information represents various attribute information of the user, including but not limited to basic attributes such as gender, age, education, occupation, and advanced attribute information such as user purchasing power, promotion sensitivity, and category preference. The first time period operation sequence and the second time period operation sequence represent the operation information performed by the user in the first time period and the second time period, respectively. The duration corresponding to the first time period and the second time period can be specifically set according to the actual situation. As an example. The time length represented by the first duration is shorter, and the time length represented by the second time period is longer. Taking the target user as a user using the above-mentioned shopping platform as an example, the first time period operation sequence can be the operation information performed by the user on the above-mentioned shopping platform within a session (stage); the second time period operation sequence can be the operation information performed by the user on the above-mentioned shopping platform in the previous week. Among them, the operation information performed by the user can be various types of operations, including but not limited to browsing, clicking, purchasing, adding to purchase, returning goods and other operations.

[0044] The session division rule is that adjacent operations within a specific duration (for example, 30 minutes) belong to the same session, and operations exceeding the specific duration are divided into the next session. It can be understood that the user's operation sequence contains rich user interest feature information. In the same session, users often have clear purchase targets of interest, and when starting a new session, the user's interests may change. As an example, the user's most recent session can be regarded as the user's short-term interest, and all operation information within a week before the session corresponding to the short-term interest can be regarded as the user's long-term interest.

[0045] In this embodiment, the user representation model is used to predict the user's pending operation based on the user's portrait information, the first time period operation sequence and the second time period operation sequence. The user feature vector is the intermediate data obtained by the user representation model during the prediction process. As an example, the user representation model may include an input layer, an embedding layer, a multi-head self-attention layer, an attention fusion layer, and a fully connected layer.

[0046] In some optional implementations of this embodiment, the user representation model is trained through the following steps:

[0047] First, taking the target user's first time period operation sequence and the second time period operation sequence as input, the dimensional feature vectors of the items involved in the target user's operation information in different dimensions are determined through the word vector model.

[0048] As an example, the word vector model is a word2vec model. Specifically, the dimensional feature vectors of the items involved in the target user's operation information in different dimensions can be determined by a CBOW model or a skip-gram model. Among them, the items can be any items involved in the operation information performed by the user on the shopping platform.

[0049] The dimension of the item can be any dimension involved in the item, including but not limited to the item dimension, the brand dimension to which the item belongs, the category dimension to which the item belongs, and the store dimension to which the item belongs. The operation information in the first time period operation sequence and the second time period operation sequence can be recorded information under the same dimension, or can be recorded information under different dimensions. Taking the item as an example, when the dimensions are the same, the operation information in the first time period operation sequence and the second time period operation sequence can both correspond to the mobile phone model; when the dimensions are different, the operation information in the first time period operation sequence corresponds to the mobile phone model, and the operation information in the second time period operation sequence corresponds to the mobile phone brand. It can be understood that under the item dimension, such as the SKU (Stock Keeping Unit) dimension, the information corresponding to the operation sequence is more detailed; the information corresponding to the operation sequence in the brand dimension, category dimension, and store dimension is more general.

[0050] Second, the dimensional feature vectors of the items involved in different dimensions are used as the initial weight parameters of the embedding layer in the initial user representation model.

[0051] The initial user representation model is the user representation model before training. During the training process of the user representation model, the weight parameters of the embedding layer can be updated along with the training process. In order to increase the training speed of the model, the embedding layer in the user representation model can also keep the initial weight unchanged or keep part of the initial weight parameters unchanged.

[0052] In some optional implementations of this embodiment, the first time period operation sequence represents the user's operation information in the item dimension, and the second time period operation sequence represents the operation information in the brand dimension, the category dimension, and the store dimension. The embedding layer in the user representation model includes the embedding layers corresponding to the portrait information, the first time period operation sequence, and the second time period operation sequence in sequence.

[0053] The execution subject may perform the second step in the following manner: determining the dimensional feature vector of the item involved in the item dimension as the initial weight parameter of the embedding layer corresponding to the first time period operation sequence; determining the dimensional feature vector of the item involved in the corresponding brand dimension, category dimension, and store dimension as the initial weight parameter of the embedding layer corresponding to the second time period operation sequence. The initial weight parameter of the embedding layer corresponding to the portrait information may be an initialized weight parameter.

[0054] The operation information in the first period operation sequence is the operation information under the item dimension, which is richer and requires the user representation model to learn more information during the training process; while the operation sequence in the second period is the operation information under the corresponding brand dimension, category dimension, and store dimension, which is more general. In order to ensure the accuracy of the user representation model on the basis of improving the model training speed, during the training process of the user representation model, only the weight parameters of the embedding layer corresponding to the portrait information and the first period operation sequence are updated, and the weight parameters of the embedding layer corresponding to the second period operation sequence are kept unchanged.

[0055] Third, for each operation information in the first period operation sequence of each target user, the user's portrait information, the operation information up to this operation information in the first period operation sequence and the second period operation sequence are used as input, and the next operation information of this operation information in the first period operation sequence is used as the expected output to train a user representation model.

[0056] As an example, the first-period operation sequence of user A includes 10 operation information recorded in chronological order, and the current operation information is the 5th operation information. The input of the user representation model is the portrait information of user A, the first 5 operation information in the first-period operation sequence, and the second-period operation information. The expected output of the user representation model is the 6th operation information in the first-period operation sequence.

[0057] During the training process, the model loss can be determined based on the actual output and expected output of the user representation model, and then the parameters of the Embedding layer, Multi-head self-attention layer, Attention Merge layer, and fully connected layer are updated according to the model loss until the preset end conditions are met (for example, the model loss converges, the model accuracy reaches a preset threshold, the number of training times exceeds the preset number, and the training time exceeds the preset time).

[0058] In this embodiment, the execution subject can obtain the user feature vector of each user in the target user through the trained user representation model in advance. It can be understood that as time goes by, the user will generate a new operation sequence on the shopping platform. The execution subject can obtain the target new operation sequence based on the preset time interval and update the user feature vector of the target user. The preset time interval can be specifically set according to the actual situation. For example, the time interval can be one day or one week.

[0059] In some optional implementations of this embodiment, the execution subject may perform step 201 in the following manner:

[0060] First, for each target user, in the user representation model, the multi-head attention mechanism is used to process the sequence information of each operation information in the first period operation sequence, the second period operation sequence, and the second period operation sequence of the user to obtain the intermediate data corresponding to the first period operation sequence and the intermediate data corresponding to the second period operation sequence.

[0061] In this implementation, in the user representation model, the above-mentioned execution entity can first process the user's portrait information, the first time period operation sequence, and the second time period operation sequence through the embedding layers corresponding to the portrait information, the first time period operation sequence, and the second time period operation sequence in sequence, and process the output information of the embedding layer corresponding to the first time period operation sequence through the first multi-head attention layer to obtain the intermediate data corresponding to the first time period operation sequence; process the output information of the embedding layer corresponding to the second time period operation sequence and the sequence information of each operation information in the second time period operation sequence through the second multi-head attention layer to obtain the intermediate data corresponding to the second time period operation sequence.

[0062] Among them, the user's portrait information can be obtained through the embedding layer, splicing layer, and full connection layer corresponding to the portrait information to obtain the intermediate data corresponding to the portrait information.

[0063] Second, through the attention mechanism, the intermediate data corresponding to the operation sequence in the first period, the intermediate data corresponding to the operation sequence in the second period, and the intermediate data corresponding to the portrait information are fused to obtain the user feature vector corresponding to the user.

[0064] As an example, the above-mentioned execution entity can fuse the intermediate data corresponding to the operation sequence of the first time period, the intermediate data corresponding to the operation sequence of the second time period, and the intermediate data corresponding to the portrait information through the fusion attention layer, and obtain the user feature vector corresponding to the user after processing through the fully connected layer.

[0065] The model structure of the user representation model 300 can be as follows: Figure 3 As shown, the user representation model 300 includes the Embedding layers 301, 302, 303 corresponding to the portrait information, the first period operation sequence, and the second period operation sequence, a splicing layer 304, the Multi-head self-attention layers 305, 306 corresponding to the first period operation sequence and the second period operation sequence, an Attention Merge layer 307, and multiple fully connected layers.

[0066] Step 202: Based on the similarity of the user feature vectors, a plurality of candidate users similar to the seed user are determined among the target users.

[0067] In this embodiment, the execution subject may determine a plurality of candidate users similar to the seed user among the target users based on the similarity of the user feature vectors, wherein the user feature vectors of the seed user are also obtained based on the user representation model.

[0068] As an example, the execution subject may determine the similarity between the user feature vector of the target user and the user feature vector of the seed user according to Euclidean distance, standardized Euclidean distance, Mahalanobis distance, etc. Specifically, the execution subject may pre-set a distance threshold and determine the target user whose distance is less than the preset distance threshold as a candidate user.

[0069] As another example, in order to improve the screening speed of candidate users, the above-mentioned execution entity uses FAISS (Facebook Artificial Intelligence Similarity Search) to calculate the similarity between the user feature vector of the target user and the user feature vector of the seed user, and determines multiple candidate users among the target users who are similar to the seed user.

[0070] Step 203: Input the user feature vector of each user among the multiple candidate users and the operation sequence of the user within the current preset time period into the recognition model to obtain at least one similar user.

[0071] In this embodiment, the execution entity may input the user feature vector of each user among the multiple candidate users and the operation sequence of the user within a preset period up to the present into the recognition model to obtain at least one similar user.

[0072] The recognition model is trained with seed users as positive samples and non-seed users as negative samples. Non-seed users can be any user other than seed users. As an example, the recognition model may include network structure layers such as an embedding layer, an attention fusion layer, and a fully connected layer.

[0073] In this embodiment, in order to improve the training speed of the prediction model, the execution subject uses the dimension feature vectors of the items in different dimensions obtained by the word vector model as the initial weight parameters of the embedding layer in the initial prediction model. The dimensions include but are not limited to the item dimension, brand dimension, category dimension, and store dimension.

[0074] In some optional implementations of this embodiment, the execution subject may perform step 203 in the following manner:

[0075] First, for each user among the multiple candidate users, the operation sequence of the user up to the current preset time period is processed through the embedding layer in the prediction model to obtain corresponding intermediate data.

[0076] The time length corresponding to the preset time period can be set according to the actual situation. For example, the preset time period can be one day. The operation sequence includes operation information in different dimensions, including but not limited to the item dimension, brand dimension, category dimension, and store dimension.

[0077] Second, through the attention mechanism, the user feature vector of the user and the intermediate data corresponding to the operation sequence within the current preset time period are fused to predict whether the user is a similar user.

[0078] As an example, the execution subject may input the user feature vector of the user and the intermediate data corresponding to the operation sequence up to the current preset time period into the fusion attention layer for fusion, so as to predict whether the user is a similar user.

[0079] In some optional implementations of this embodiment, in order to further improve the training speed of the prediction model, during the training process of the prediction model, the weight parameters of the embedding layer of the prediction model are kept unchanged.

[0080] Continue to see Figure 4 , Figure 4 FIG4 is a schematic diagram 400 of an application scenario of the similar group expansion method according to this embodiment. Figure 4 In the application scenario, the server 401 first inputs the portrait information, the operation sequence of the first time period and the operation sequence of the second time period of each user among the target users into the user representation model 402 to obtain the user feature vector corresponding to each user, wherein the user representation model is used to predict the user's pending operation based on the user's portrait information, the operation sequence of the first time period and the operation sequence of the second time period, and the user feature vector is the intermediate data obtained by the user representation model in the prediction process. Then, based on the similarity of the user feature vectors, multiple candidate users similar to the seed user among the target users are determined. Finally, the user feature vector of each user among the multiple candidate users and the operation sequence of the user up to the current preset time period are input into the recognition model 403 to obtain at least one similar user. Among them, the recognition model is trained with seed users as positive samples and non-seed users as negative samples.

[0081] The method provided by the above-mentioned embodiments of the present disclosure obtains a user feature vector corresponding to each user by inputting the portrait information, the first time period operation sequence and the second time period operation sequence of each user among the target users into a user representation model, wherein the user representation model is used to predict the user's upcoming operation based on the user's portrait information, the first time period operation sequence and the second time period operation sequence, and the user feature vector is the intermediate data obtained by the user representation model in the prediction process; based on the similarity of the user feature vectors, multiple candidate users similar to the seed user among the target users are determined; the user feature vector of each user among the multiple candidate users and the operation sequence of the user up to the current preset time period are input into a recognition model to obtain at least one similar user, wherein the recognition model is trained with seed users as positive samples and non-seed users as negative samples, thereby providing a similar population expansion method and improving the accuracy of determining the similar population of the seed user.

[0082] In some optional implementations of this embodiment, the execution subject may also push preset information corresponding to the seed user to the determined similar person. As an example, the correspondence between the preset information and the seed user may be predetermined. The preset information may be any type of information, such as advertising information.

[0083] Continue to refer Figure 5 , shows a schematic process 500 of an embodiment of a similar population expansion method according to the present application, comprising the following steps:

[0084] Step 501, for each target user, in the user representation model, the sequence information of each operation information in the first period operation sequence, the second period operation sequence and the second period operation sequence of the user is processed through a multi-head attention mechanism to obtain the intermediate data corresponding to the first period operation sequence and the intermediate data corresponding to the second period operation sequence.

[0085] Step 502, through the attention mechanism, the intermediate data corresponding to the operation sequence of the first time period, the intermediate data corresponding to the operation sequence of the second time period and the intermediate data corresponding to the portrait information are integrated to obtain the user feature vector corresponding to the user.

[0086] Step 503: Based on the similarity of the user feature vectors, a plurality of candidate users similar to the seed user are determined among the target users.

[0087] Step 504: for each user among the multiple candidate users, the operation sequence of the user up to the current preset time period is processed through the embedding layer in the prediction model to obtain corresponding intermediate data.

[0088] Step 505: Through the attention mechanism, the user feature vector of the user and the intermediate data corresponding to the operation sequence within the current preset time period are integrated to predict whether the user is a similar user.

[0089] It can be seen from this embodiment that Figure 2 Compared with the corresponding embodiment, the process 500 of the similar group expansion method in this embodiment specifically illustrates the process of determining the user feature vector and the process of identifying the recognition model, which further improves the accuracy of determining the similar group of the seed user.

[0090] Continue to refer Figure 6 As an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of a similar group expansion device. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0091] like Figure 6 As shown, the similar population expansion device includes: including: a first obtaining unit 601, which is configured to input the portrait information, the first time period operation sequence and the second time period operation sequence of each user in the target user into the user representation model, and obtain the user feature vector corresponding to each user, wherein the user representation model is used to predict the user's upcoming operation based on the user's portrait information, the first time period operation sequence and the second time period operation sequence, and the user feature vector is the intermediate data obtained by the user representation model in the prediction process; a determination unit 602, which is configured to determine multiple candidate users similar to the seed user among the target users based on the similarity of the user feature vectors; a second obtaining unit 603, which is configured to input the user feature vector of each user among the multiple candidate users and the operation sequence of the user up to the current preset time period into the recognition model, and obtain at least one similar user, wherein the recognition model is trained with the seed user as the positive sample and the non-seed user as the negative sample.

[0092] In some embodiments, the above-mentioned device also includes: a training unit (not shown in the figure), which is configured to train the user representation model through the following steps: taking the first time period operation sequence and the second time period operation sequence of the target user as input, determining the dimensional feature vectors of the items involved in the target user's operation information in different dimensions through the word vector model; using the dimensional feature vectors of the items involved in different dimensions as the initial weight parameters of the embedding layer in the initial user representation model; for each operation information in the first time period operation sequence of each user among the target users, taking the user's portrait information, the operation information up to the operation information in the first time period operation sequence and the second time period operation sequence as input, and taking the next operation information of the operation information in the first time period operation sequence as the expected output, training to obtain the user representation model.

[0093] In some embodiments, the dimensions include item dimension, brand dimension to which the item belongs, category dimension to which the item belongs, and store dimension to which the item belongs; the first time period operation sequence represents the user's operation information in the item dimension, and the second time period operation sequence represents the operation information in the brand dimension, category dimension, and store dimension; the embedding layer in the user representation model includes embedding layers corresponding to the portrait information, the first time period operation sequence, and the second time period operation sequence in sequence; and the training unit (not shown in the figure) is further configured to: determine the dimensional feature vector of the item involved in the item dimension as the initial weight parameter of the embedding layer corresponding to the first time period operation sequence; determine the dimensional feature vector of the item involved in the corresponding brand dimension, category dimension, and store dimension as the initial weight parameter of the embedding layer corresponding to the second time period operation sequence.

[0094] In some embodiments, during the training process of the user representation model, only the portrait information and the weight parameters of the embedding layer corresponding to the operation sequence of the first period are updated, and the weight parameters of the embedding layer corresponding to the operation sequence of the second period are kept unchanged.

[0095] In some embodiments, the first obtaining unit 601 is further configured to: for each user among the target users, in the user representation model, process the sequence information of each operation information in the first time period operation sequence, the second time period operation sequence and the second time period operation sequence of the user through a multi-head attention mechanism to obtain the intermediate data corresponding to the first time period operation sequence and the intermediate data corresponding to the second time period operation sequence; through the attention mechanism, fuse the intermediate data corresponding to the first time period operation sequence, the intermediate data corresponding to the second time period operation sequence and the intermediate data corresponding to the portrait information to obtain the user feature vector corresponding to the user.

[0096] In some embodiments, the dimensional feature vectors of the items in different dimensions obtained by the word vector model are used as the initial weight parameters of the embedding layer in the initial prediction model; the second obtaining unit 603 is further configured to: for each user among multiple candidate users, process the operation sequence of the user up to the current preset time period through the embedding layer in the prediction model to obtain the corresponding intermediate data; through the attention mechanism, the user feature vector of the user and the intermediate data corresponding to the operation sequence up to the current preset time period are fused to predict whether the user is a similar user.

[0097] In some embodiments, during the training process of the prediction model, the weight parameters of the embedding layer of the prediction model are kept from being updated.

[0098] In this embodiment, the first obtaining unit in the similar population expansion device inputs the portrait information, the first time period operation sequence and the second time period operation sequence of each user among the target users into the user representation model to obtain the user feature vector corresponding to each user, wherein the user representation model is used to predict the user's upcoming operation based on the user's portrait information, the first time period operation sequence and the second time period operation sequence, and the user feature vector is the intermediate data obtained by the user representation model in the prediction process; the determination unit determines multiple candidate users among the target users who are similar to the seed user based on the similarity of the user feature vectors; the second obtaining unit inputs the user feature vector of each user among the multiple candidate users and the operation sequence of the user up to the current preset time period into the recognition model to obtain at least one similar user, wherein the recognition model is trained with the seed user as the positive sample and the non-seed user as the negative sample, thereby providing a similar population expansion device and improving the accuracy of determining the similar population of the seed user.

[0099] Reference below Figure 7 , which shows a device suitable for implementing the embodiments of the present application (eg Figure 1 Schematic diagram of the structure of a computer system 700 of devices 101, 102, 103, 105) shown. Figure 7 The device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0100] like Figure 7 As shown, the computer system 700 includes a processor (e.g., CPU, central processing unit) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage part 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the system 700 are also stored. The processor 701, ROM 702, and RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0101] The following components are connected to the I / O interface 705: an input section 706 including a keyboard, a mouse, etc.; an output section 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, a modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed, so that a computer program read therefrom is installed into the storage section 708 as needed.

[0102] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 709, and / or installed from a removable medium 711. When the computer program is executed by the processor 701, the above-mentioned functions defined in the method of the present application are executed.

[0103] It should be noted that the computer-readable medium of the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0104] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the client computer, partially on the client computer, as a separate software package, partially on the client computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the client computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0105] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the device, method and computer program product according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0106] The units involved in the embodiments described in the present application may be implemented by software or by hardware. The described units may also be set in a processor, for example, may be described as: a processor comprising a first obtaining unit, a determining unit, and a second obtaining unit. Among them, the names of these units do not constitute a limitation on the unit itself under certain circumstances. For example, the first obtaining unit may also be described as "a unit that inputs the portrait information of each user among the target users, the first time period operation sequence, and the second time period operation sequence into the user representation model to obtain the user feature vector corresponding to each user".

[0107] As another aspect, the present application also provides a computer-readable medium, which may be included in the device described in the above embodiment; or it may exist independently and not be assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by the device, the computer device: inputs the portrait information, the first time period operation sequence and the second time period operation sequence of each user in the target user into the user representation model, and obtains the user feature vector corresponding to each user, wherein the user representation model is used to predict the user's to-be-performed operation according to the user's portrait information, the first time period operation sequence and the second time period operation sequence, and the user feature vector is the intermediate data obtained by the user representation model in the prediction process; based on the similarity of the user feature vector, multiple candidate users similar to the seed user in the target user are determined; the user feature vector of each user in the multiple candidate users and the operation sequence of the user up to the current preset time period are input into the recognition model, and at least one similar user is obtained, wherein the recognition model is trained with the seed user as the positive sample and the non-seed user as the negative sample.

[0108] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above invention concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

Claims

1. A similar population expansion method, include: Inputting the portrait information, the operation sequence of the first time period and the operation sequence of the second time period of each target user into the user representation model to obtain a user feature vector corresponding to each user, wherein the user representation model is used to predict the user's to-be-performed operation according to the user's portrait information, the operation sequence of the first time period and the operation sequence of the second time period, and the user feature vector is the intermediate data obtained by the user representation model in the prediction process, and the time length corresponding to the second time period is longer than the time length corresponding to the first time period; Based on the similarity of the user feature vectors, determining a plurality of candidate users similar to the seed user among the target users; The user feature vector of each user among the multiple candidate users and the operation sequence of the user up to the current preset time period are input into the recognition model to obtain at least one similar user, wherein the recognition model is trained with the seed user as a positive sample and the non-seed user as a negative sample.

2. The method according to claim 1, in, The user representation model is trained by the following steps: Taking the first time period operation sequence and the second time period operation sequence of the target user as input, determining the dimensional feature vectors of the items involved in the operation information of the target user in different dimensions through a word vector model; The dimension feature vectors of the items involved in different dimensions are used as the initial weight parameters of the embedding layer in the initial user representation model; For each operation information in the first time period operation sequence of each user among the target users, the user's portrait information, the operation information up to this operation information in the first time period operation sequence and the second time period operation sequence are used as input, and the next operation information of this operation information in the first time period operation sequence is used as the expected output, to train the user representation model.

3. The method according to claim 2, in, The dimensions include item dimension, brand dimension, category dimension and store dimension; the first time period operation sequence represents the user's operation information in the item dimension, and the second time period operation sequence represents the operation information in the brand dimension, category dimension and store dimension; the embedding layer in the user representation model includes the embedding layers corresponding to the portrait information, the first time period operation sequence and the second time period operation sequence in sequence; as well as The step of using the dimension feature vectors of the items involved in different dimensions as initial weight parameters of the embedding layer in the initial user representation model includes: Determine the dimension feature vector of the item involved in the item dimension as the initial weight parameter of the embedding layer corresponding to the operation sequence of the first period; The dimensional feature vectors of the items involved in the corresponding brand dimension, category dimension, and store dimension are determined as the initial weight parameters of the embedding layer corresponding to the operation sequence in the second period.

4. The method according to claim 3, in, During the training process of the user representation model, only the portrait information and the weight parameters of the embedding layer corresponding to the operation sequence of the first period are updated, and the weight parameters of the embedding layer corresponding to the operation sequence of the second period are kept unchanged.

5. The method according to claim 3, in, The step of inputting the portrait information of each target user, the operation sequence of the first time period, and the operation sequence of the second time period into the user representation model to obtain the user feature vector corresponding to each user includes: For each user among the target users, in the user representation model, the sequence information of each operation information in the first period operation sequence, the second period operation sequence and the second period operation sequence of the user is processed by a multi-head attention mechanism to obtain the intermediate data corresponding to the first period operation sequence and the intermediate data corresponding to the second period operation sequence; Through the attention mechanism, the intermediate data corresponding to the operation sequence in the first period, the intermediate data corresponding to the operation sequence in the second period, and the intermediate data corresponding to the portrait information are fused to obtain the user feature vector corresponding to the user.

6. The method according to claim 5, in, Using the dimension feature vectors of the items in different dimensions obtained by the word vector model as the initial weight parameters of the embedding layer in the initial prediction model; The step of inputting a user feature vector of each user among the plurality of candidate users and an operation sequence of the user within a preset period up to the present into a recognition model to obtain at least one similar user comprises: For each user among the multiple candidate users, processing the operation sequence of the user within a preset time period up to the present through the embedding layer in the prediction model to obtain corresponding intermediate data; Through the attention mechanism, the user feature vector of the user and the intermediate data corresponding to the operation sequence within the current preset time period are fused to predict whether the user is a similar user.

7. The method according to claim 6, in, During the training process of the prediction model, the weight parameters of the embedding layer of the prediction model are kept unchanged.

8. A similar group expansion device, include: A first obtaining unit is configured to input the portrait information of each user among the target users, the operation sequence of the first time period and the operation sequence of the second time period into a user representation model to obtain a user feature vector corresponding to each user, wherein the user representation model is used to predict the user's to-be-performed operation according to the user's portrait information, the operation sequence of the first time period and the operation sequence of the second time period, the user feature vector is intermediate data obtained by the user representation model in the prediction process, and the time length corresponding to the second time period is longer than the time length corresponding to the first time period; a determining unit configured to determine a plurality of candidate users similar to the seed user among the target users based on the similarity of the user feature vectors; The second obtaining unit is configured to input the user feature vector of each user among the multiple candidate users and the operation sequence of the user within a current preset time period into a recognition model to obtain at least one similar user, wherein the recognition model is trained with the seed user as a positive sample and the non-seed user as a negative sample.

9. The device according to claim 8, in, The invention also includes: a training unit, configured to train the user representation model through the following steps: Taking the first time period operation sequence and the second time period operation sequence of the target user as input, determine the dimensional feature vectors of the items involved in the operation information of the target user in different dimensions through a word vector model; use the dimensional feature vectors of the items involved in different dimensions as the initial weight parameters of the embedding layer in the initial user representation model; for each operation information in the first time period operation sequence of each user among the target users, take the user's portrait information, the operation information up to the operation information in the first time period operation sequence and the second time period operation sequence as input, and take the next operation information of the operation information in the first time period operation sequence as the expected output, and train to obtain the user representation model.

10. The device according to claim 9, in, The dimensions include an item dimension, a brand dimension to which the item belongs, a category dimension to which the item belongs, and a store dimension to which the item belongs; the first time period operation sequence represents the user's operation information in the item dimension, and the second time period operation sequence represents the operation information in the brand dimension, the category dimension, and the store dimension; the embedding layer in the user representation model includes an embedding layer corresponding to the portrait information, the first time period operation sequence, and the second time period operation sequence in sequence; and The training unit is further configured to: The dimensional feature vectors of the items involved in the item dimension are determined as the initial weight parameters of the embedding layer corresponding to the operation sequence of the first period; the dimensional feature vectors of the items involved in the corresponding brand dimension, category dimension, and store dimension are determined as the initial weight parameters of the embedding layer corresponding to the operation sequence of the second period.

11. The device according to claim 10, in, During the training process of the user representation model, only the portrait information and the weight parameters of the embedding layer corresponding to the operation sequence of the first period are updated, and the weight parameters of the embedding layer corresponding to the operation sequence of the second period are kept unchanged.

12. The device according to claim 10, in, The first obtaining unit is further configured to: For each user among the target users, in the user representation model, the sequence information of each operation information in the first time period operation sequence, the second time period operation sequence and the second time period operation sequence of the user is processed through a multi-head attention mechanism to obtain the intermediate data corresponding to the first time period operation sequence and the intermediate data corresponding to the second time period operation sequence; through the attention mechanism, the intermediate data corresponding to the first time period operation sequence, the intermediate data corresponding to the second time period operation sequence and the intermediate data corresponding to the portrait information are fused to obtain the user feature vector corresponding to the user.

13. The device according to claim 12, in, Using the dimension feature vectors of the items in different dimensions obtained by the word vector model as the initial weight parameters of the embedding layer in the initial prediction model; The second obtaining unit is further configured to: For each user among the multiple candidate users, processing the operation sequence of the user within a preset time period up to the present through the embedding layer in the prediction model to obtain corresponding intermediate data; Through the attention mechanism, the user feature vector of the user and the intermediate data corresponding to the operation sequence within the current preset time period are fused to predict whether the user is a similar user.

14. The device according to claim 13, in, During the training process of the prediction model, the weight parameters of the embedding layer of the prediction model are kept unchanged.

15. A computer readable medium having a computer program stored thereon, in, When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

16. An electronic device, include: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

Images