Advertisement push method, device, electronic device and storage medium

Abstract

The present application provides an advertisement push method, device, electronic device and storage medium, the method comprising: obtaining advertisement materials of multiple first advertisements, for each first advertisement, using a pre-trained advertisement feature extraction model to extract a corresponding advertisement joint representation vector according to the advertisement material, determining the advertisement preference vector of the target user according to the second advertisement clicked by the target user, calculating the matching degree between the advertisement joint representation vectors of multiple first advertisements and the advertisement preference vector of the target user, and pushing the first advertisement whose matching degree meets the preset conditions to the target user. This solution can realize advertisement push based on advertisement materials without obtaining online advertisement delivery data, and can realize accurate push even for newly created advertisements, thereby increasing the click volume of pushed advertisements.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to an advertisement push method, device, electronic device and storage medium. Background Art

[0002] Current advertising recommendation algorithms usually use the system and ID features such as the creative ID and order ID of the advertisement to estimate the effectiveness of the advertisement, and then push the advertisement to relevant users who are interested in the advertisement based on the effectiveness, thereby increasing the number of clicks on the advertisement.

[0003] However, for newly created ads, since there is no online delivery data, the existing ad recommendation algorithm cannot achieve accurate delivery in the early stage of ad delivery, making the newly created ads ineffective for a period of time and only getting a small number of clicks. Summary of the invention

[0004] In order to overcome the problem in the related art that newly created advertisements cannot be accurately pushed, the present application provides an advertisement pushing method, device, electronic device and storage medium.

[0005] In a first aspect, an embodiment of the present application provides an advertisement push method, comprising:

[0006] Obtaining an advertisement material corresponding to each of the first advertisements among the plurality of first advertisements;

[0007] For each of the first advertisements, input the corresponding advertisement material into a pre-trained advertisement feature extraction model to obtain an output result, where the output result is an advertisement joint representation vector of the first advertisement;

[0008] determining an advertisement preference vector of the target user according to a second advertisement clicked by the target user;

[0009] determining a degree of match between an advertisement joint representation vector of each first advertisement in the plurality of first advertisements and the advertisement preference vector;

[0010] The first advertisements whose matching degree meets the preset condition among the plurality of the first advertisements are pushed to the target user.

[0011] As a possible implementation method, the training of the advertisement feature extraction model includes:

[0012] Acquire training data, wherein the training data includes advertising materials of a plurality of historical advertisements;

[0013] Obtaining click logs of the plurality of historical advertisements;

[0014] Based on the click log, construct an adjacency matrix between the multiple historical advertisements;

[0015] The preset feature extraction model is trained using the training data and the adjacency matrix to obtain an advertisement feature extraction model.

[0016] As a possible implementation manner, constructing an adjacency matrix between the plurality of historical advertisements based on the click log includes:

[0017] The click log contains the ID of the clicked advertisement and the ID of the user who clicked the advertisement. According to the advertisement click log, the plurality of historical advertisements are clustered, and the historical advertisements clicked by the same user are classified into one category.

[0018] Generate a corresponding relationship between the user and the historical advertisement according to the clustering result;

[0019] Based on the corresponding relationship, construct a correlation relationship graph between the multiple historical advertisements, wherein the correlation relationship graph has the historical advertisements as vertices, the correlation relationships between the historical advertisements as edges, and the correlation coefficients between the historical advertisements as edge weights;

[0020] An adjacency matrix corresponding to the correlation graph is generated.

[0021] As a possible implementation, the correlation coefficient between the historical advertisements is determined in the following manner:

[0022] The initial correlation coefficient between any two of the historical advertisements is determined to be 0. For any two of the historical advertisements, if it is determined according to the corresponding relationship that any two of the historical advertisements correspond to the same user, the correlation coefficient between the two historical advertisements is increased by 1.

[0023] As a possible implementation, the using the training data and the adjacency matrix to train a preset feature extraction model to obtain an advertisement feature extraction model includes:

[0024] The preset feature extraction model includes a feature extractor, a multi-layer perceptron and a triple loss function;

[0025] The training data is input into the feature extractor for feature extraction and then passed through the multi-layer perceptron for feature understanding to obtain an advertisement joint representation vector of the training data;

[0026] Using the adjacency matrix as the original true value, constructing a plurality of triplet data consisting of joint representations of advertisements of the historical advertisements;

[0027] The feature extraction model is trained using a plurality of the triplet data to obtain an advertisement feature extraction model.

[0028] As a possible implementation manner, determining the advertisement preference vector of the target user according to the second advertisement clicked by the target user includes:

[0029] Determining, according to the advertisement click log of the target user, a second advertisement clicked by the target user;

[0030] Obtaining an advertisement joint representation vector of the second advertisement;

[0031] The advertisement joint representation vector of the second advertisement is used as the advertisement preference vector of the user.

[0032] As a possible implementation manner, the method further includes:

[0033] The advertisement joint representation vector output by the advertisement feature extraction model and the corresponding advertisement are stored in a preset location in correspondence with each other;

[0034] The acquiring the advertisement joint representation vector of the second advertisement includes:

[0035] Detecting whether an advertisement joint representation vector corresponding to the second advertisement is stored in the preset position;

[0036] If it is determined that the advertisement joint characterization vector corresponding to the second advertisement is stored in the preset position, acquiring the advertisement joint characterization vector corresponding to the second advertisement stored in the preset position;

[0037] If it is determined that the advertisement joint representation vector corresponding to the second advertisement is not stored in the preset position, the advertisement material corresponding to the second advertisement is obtained, and the advertisement material corresponding to the second advertisement is input into the advertisement feature extraction model to obtain the corresponding output result, and the output result is used as the advertisement joint representation vector of the second advertisement.

[0038] In a second aspect, the embodiment of the present application further provides an advertisement push device, including:

[0039] An acquisition module, used to acquire an advertisement material corresponding to each of a plurality of first advertisements;

[0040] an advertisement vector determination module, configured to input the corresponding advertisement material into a pre-trained advertisement feature extraction model for each of the first advertisements, and obtain an output result, wherein the output result is an advertisement joint representation vector of the first advertisement;

[0041] A user vector determination module, configured to determine an advertisement preference vector of a target user according to a second advertisement clicked by the target user;

[0042] a calculation module, configured to determine a degree of match between an advertisement joint representation vector of each of the plurality of first advertisements and the advertisement preference vector;

[0043] The push module is used to push the first advertisements whose matching degree meets the preset conditions among the multiple first advertisements to the target user.

[0044] In a third aspect, an embodiment of the present application further provides an electronic device, comprising: a processor and a memory, wherein the processor is used to execute an advertisement push program stored in the memory to implement the advertisement push method described in the first aspect.

[0045] In a fourth aspect, an embodiment of the present application further provides a storage medium, wherein the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the advertisement push method described in the first aspect.

[0046] The technical solution provided by the present application may include the following beneficial effects: the present application provides an advertisement push method, which obtains advertisement materials of multiple first advertisements, and for each first advertisement, uses a pre-trained advertisement feature extraction model to extract the corresponding advertisement joint representation vector according to the advertisement material, determines the advertisement preference vector of the target user according to the second advertisement clicked by the target user, calculates the matching degree between the advertisement joint representation vectors of multiple first advertisements and the advertisement preference vector of the target user, and pushes the first advertisement whose matching degree meets the preset conditions to the target user. This solution can realize advertisement push based on advertisement materials without obtaining online advertisement delivery data, and can realize accurate push even for newly created advertisements, thereby increasing the click volume of pushed advertisements.

[0047] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0049] Figure 1 is a flow chart of an advertisement pushing method according to an exemplary embodiment;

[0050] Figure 2 is a flow chart of an advertisement pushing method according to another exemplary embodiment;

[0051] Figure 3 is a block diagram of an advertisement pushing device according to an exemplary embodiment;

[0052] Figure 4is a schematic diagram of an electronic device according to an exemplary embodiment;

[0053] Figure 5 is a schematic diagram showing a correlation diagram according to an exemplary embodiment. DETAILED DESCRIPTION

[0054] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0055] See also Figure 1 , is a flow chart of an advertisement push method provided by an embodiment of the present application, such as Figure 1 As shown, the advertisement push method provided in this embodiment may include:

[0056] S11. Obtain an advertisement material corresponding to each first advertisement among multiple first advertisements.

[0057] In an embodiment of the present invention, the method can be applied to a server for pushing advertisements, wherein an advertisement library is provided in the server, and advertisement materials corresponding to multiple advertisements are stored in the advertisement library. In the advertisement library, the advertisement materials can be stored corresponding to the corresponding advertisement IDs.

[0058] In actual applications, multiple advertisements may be determined from the advertisement library as first advertisements, and advertisement materials corresponding to each first advertisement may be obtained from the advertisement library, wherein the first advertisement may be a historically created advertisement or a newly created advertisement.

[0059] In actual applications, the advertising material may include text information and / or picture information of the advertisement, wherein the picture information may include a cover picture of the advertising video.

[0060] S12. For each of the first advertisements, the corresponding advertisement material is input into a pre-trained advertisement feature extraction model to obtain an output result, where the output result is an advertisement joint representation vector of the first advertisement.

[0061] In the embodiment of the present invention, the advertisement joint representation vector of the advertisement is also the embedding vector of the advertisement. The property of the embedding vector is to enable objects corresponding to vectors with similar distances to have similar meanings.

[0062] In an embodiment of the present application, the advertisement joint representation vector of an advertisement is obtained based on the advertisement copy information, image information and other advertising materials, without using the advertisement online delivery data. Therefore, even for a newly created advertisement, its corresponding advertisement joint representation vector can be determined.

[0063] In an embodiment of the present invention, the advertising feature extraction model is a pre-trained and stored model. After obtaining multiple advertising materials corresponding to multiple first advertisements, the advertising materials corresponding to each first advertisement can be respectively input into the advertising feature extraction model for feature extraction, thereby obtaining the advertising joint representation vector corresponding to each first advertisement.

[0064] As for how the advertising feature extraction model is trained, we will discuss it in detail below. Figure 2 This will be explained and will not be described in detail here.

[0065] S13. Determine an advertisement preference vector of the target user according to the second advertisement clicked by the target user.

[0066] Usually, users will only click on advertisements that they are interested in, so advertisements clicked by users can be considered as advertisements that meet user preferences. Therefore, in an embodiment of the present invention, the advertisement preference vector of the target user can be determined based on the second advertisement clicked by the target user.

[0067] In an embodiment of the present invention, when determining the advertising preference vector of a target user, the advertising click log of the target user can be obtained, and a second advertisement clicked by the target user can be determined based on the advertising click log of the target user, and an advertising joint characterization vector of the second advertisement can be obtained, and the advertising joint characterization vector of the second advertisement can be used as the advertising preference vector of the user.

[0068] The target user's ad click log may be obtained from user behavior data according to the target user's user ID. The obtained ad click log generally includes the target user's user ID and the ID of the ad clicked by the target user.

[0069] In the embodiment of the present invention, it is possible to determine one or more advertisements clicked by the target user based on the advertisement click log of the target user.

[0070] As an optional implementation manner, if an advertisement clicked by the target user is determined based on the advertisement click log of the target user, the advertisement is determined as the second advertisement clicked by the target user.

[0071] As another optional implementation method, if multiple advertisements clicked by the target user are determined based on the target user's advertisement click log, the one with the click time closest to the current time can be selected from the multiple advertisements clicked by the target user as the second advertisement clicked by the target user.

[0072] In an embodiment of the present invention, the advertisement joint characterization vector output by the advertisement feature extraction model and the corresponding advertisement may be stored in a preset location in correspondence, wherein the preset location may be set according to actual needs. When the advertisement joint characterization vector and the advertisement are stored in correspondence, in order to save storage space, the advertisement joint characterization vector and the ID of the corresponding advertisement may be stored in correspondence. Based on this, after determining the second advertisement clicked by the target user, the advertisement joint characterization vector of the second advertisement may be obtained in the following manner:

[0073] Detect whether an advertisement joint characterization vector corresponding to the second advertisement is stored in the preset position. If it is determined that an advertisement joint characterization vector corresponding to the second advertisement is stored in the preset position, obtain the advertisement joint characterization vector corresponding to the second advertisement stored in the preset position as the advertisement joint characterization vector of the second advertisement. If it is determined that the advertisement joint characterization vector corresponding to the second advertisement is not stored in the preset position, obtain the advertising material corresponding to the second advertisement, input the advertising material corresponding to the second advertisement into the advertising feature extraction model, obtain the corresponding output result, and use the output result as the advertisement joint characterization vector of the second advertisement.

[0074] S14. Determine a degree of match between the advertisement joint representation vector of each first advertisement in the plurality of first advertisements and the advertisement preference vector.

[0075] In an embodiment of the present invention, for each first advertisement, the degree of match between the advertisement joint representation vector and the advertisement preference vector of the first advertisement can be determined by calculating the cosine similarity between the advertisement joint representation vector and the advertisement preference vector of the first advertisement, wherein the larger the cosine similarity, the higher the degree of match, and the smaller the cosine similarity, the lower the degree of match.

[0076] S15. Pushing the first advertisements whose matching degree meets the preset conditions among the multiple first advertisements to the target user.

[0077] In the embodiment of the present invention, the preset conditions can be set according to actual needs.

[0078] As an embodiment, the preset condition is to sort multiple first advertisements in order of matching degree from large to small, select the first advertisements ranked in the top k positions as target advertisements, where k is an integer greater than or equal to 1, and after selecting k target advertisements according to the above preset conditions, push the target advertisements to the target users.

[0079] In an embodiment of the present invention, the advertising preference vector of the target user is the advertising joint representation vector of the second advertisement clicked by the target user. Therefore, the matching degree between the advertising joint representation vectors of multiple first advertisements and the advertising preference vector is calculated, that is, the similarity between the multiple first advertisements and the second advertisement clicked by the target user is calculated, and then the advertisement is pushed to the target user based on the calculated matching degree. This can achieve the goal of pushing advertisements similar to and related to the advertisements clicked by the target user.

[0080] In actual applications, the first advertisement whose matching degree meets the preset conditions is pushed to the target user, that is, the first advertisement whose matching degree meets the preset conditions is pushed to the terminal device used by the target user. Specifically, the user information of the target user, such as account information, can be obtained, and then the terminal device currently logged in by the target user's account information can be determined, and the terminal device is determined as the target terminal device, and then the first advertisement whose matching degree meets the preset conditions is pushed to the target terminal device, thereby achieving the first advertisement whose matching degree meets the preset conditions is pushed to the target user.

[0081] The application provides an advertisement push method, which obtains advertisement materials of multiple first advertisements, extracts the corresponding advertisement joint representation vector according to the advertisement materials using a pre-trained advertisement feature extraction model for each first advertisement, determines the advertisement preference vector of the target user according to the second advertisement clicked by the target user, calculates the matching degree between the advertisement joint representation vectors of multiple first advertisements and the advertisement preference vector of the target user, and pushes the first advertisement whose matching degree meets the preset conditions to the target user. This solution can realize advertisement push based on advertisement materials without obtaining online advertisement delivery data, and can realize accurate push even for newly created advertisements, thereby increasing the click volume of pushed advertisements.

[0082] See also Figure 2 , is a flowchart of an advertisement feature extraction model training provided by another embodiment of the present application, such as Figure 2 As shown, the training of the advertisement feature extraction model may include the following steps:

[0083] S21. Obtain training data, where the training data includes advertising materials of multiple historical advertisements.

[0084] In actual applications, advertising materials of multiple historical advertisements may be obtained from the advertisement library of the server as training data.

[0085] S22. Obtain click logs of the multiple historical advertisements.

[0086] In the embodiment of the present invention, the click logs of advertisements are usually stored in the server, so the click logs of multiple historical advertisements can be directly obtained from the server.

[0087] S23. Based on the click log, construct an adjacency matrix between the multiple historical advertisements.

[0088] In an embodiment of the present invention, the click log of an advertisement contains the ID of the clicked advertisement and the ID of the user who clicked the advertisement. Based on this, constructing an adjacency matrix between multiple historical advertisements based on the click logs of multiple historical advertisements may include: clustering multiple historical advertisements according to the click logs of multiple historical advertisements, classifying historical advertisements clicked by the same user into one category, generating a corresponding relationship between the user and the historical advertisements according to the clustering result, constructing a correlation relationship graph between multiple historical advertisements in the training data based on the corresponding relationship between the user and the historical advertisements, and then generating an adjacency matrix corresponding to the correlation relationship graph.

[0089] As an embodiment, the constructed correlation graph is an undirected graph, in which the IDs of historical advertisements in the training data are vertices, the correlations between historical advertisements are edges, and the correlation coefficients between historical advertisements are edge weights.

[0090] For example Figure 5 As shown, it is a schematic diagram of a correlation graph, which contains 4 vertices (A, B, C, D) and 5 edges, among which the weight of the edge between AB is 2, the weight of the edge between AC is 4, the weight of the edge between BC is 1, the weight of the edge between CD is 1, and the weight of the edge between AD is 1. The definition of the correlation graph is: G = (V, E), where G (graph): represents the correlation graph, V (Vertext): represents the matrix of vertices in the graph, and E (Edge): represents the matrix of edge weights in the graph. The adjacency matrix corresponding to the correlation graph is a matrix representing the adjacent relationship between vertices in the correlation graph, that is, the matrix E of edge weights.

[0091] by Figure 5 Taking the correlation diagram shown in the figure as an example, V is:

[0092] [ABCD]

[0093] E is:

[0094]

[0095] As an embodiment, the correlation coefficient between historical advertisements may be determined in the following manner:

[0096] The initial correlation coefficient between any two historical advertisements in the training data is determined to be 0. For any two historical advertisements, if it is determined that the two historical advertisements correspond to the same user based on the correspondence between the above-mentioned user and historical advertisements, the correlation coefficient between the two historical advertisements is increased by 1. For example, if historical advertisement A and historical advertisement B both correspond to user A, the correlation coefficient between historical advertisement A and historical advertisement B is 1. If historical advertisement A and historical advertisement B both correspond to user A and user B, the correlation coefficient between historical advertisement A and historical advertisement B is 2. By analogy, of course, if there is no user corresponding to both historical advertisement A and historical advertisement B, the correlation coefficient between historical advertisement A and historical advertisement B is 0. The larger the correlation coefficient between two historical advertisements, the more similar the two historical advertisements are.

[0097] Furthermore, in order to improve the accuracy of the correlation coefficient so that the correlation coefficient can more accurately reflect the similarity between historical advertisements, when determining the correspondence between the user and the historical advertisement, the interaction time between the user and the corresponding historical advertisement (that is, the time when the user clicks on the historical advertisement) is recorded at the same time. Then, when determining the similarity coefficient between historical advertisements, the correlation coefficient between the two historical advertisements can be increased by 1 only when it is determined that the two historical advertisements interact with the same user within a period of time (such as 4 days), otherwise, the correlation coefficient between the two historical advertisements is not increased by 1. For example, if user A interacted with historical advertisement A on November 20, 2020, and then interacted with historical advertisement B on November 22, 2020, then when determining the similarity coefficient between historical advertisement A and historical advertisement B, the similarity coefficient is increased by 1; if user A interacted with historical advertisement A on November 20, 2020, and then interacted with historical advertisement B on December 22, 2020, then when determining the similarity coefficient between historical advertisement A and historical advertisement B, the similarity coefficient is not increased by 1. This is because if the time interval between two historical advertisements interacting with the same user is too long, the two historical advertisements may not be related.

[0098] The adjacency matrix determined in the embodiment of the present invention can accurately reflect the correlation between the historical advertisements contained in the training data.

[0099] S24. Use the training data and the adjacency matrix to train the preset feature extraction model to obtain an advertising feature extraction model.

[0100] In an embodiment of the present invention, the preset feature extraction model may include a feature extractor, a multi-layer perceptron and a triple loss function.

[0101] As an embodiment, the preset feature extraction model includes two feature extractors, ResNet34 and SIF, where ResNet34 is used to extract features of image information and SIF is used to extract features of text information; the triplet loss function is Triplet loss. Based on this, Figure 3 As shown, when training the preset feature extraction model using training data, it may include:

[0102] The advertising material of the historical advertisement is input into the feature extraction model from the input side of the feature extraction model, and then the following method is performed for each historical advertisement: ResNet34 is used to extract features of the image in the advertising material, and SIF is used to extract features of the copy information, and the image features extracted by ResNet34 are input into the corresponding multi-layer perceptron for feature understanding to form the corresponding image feature vector, and the copy features extracted by SIF are input into the corresponding multi-layer perceptron for feature understanding to form the corresponding copy feature vector, and then the image feature vector and the copy feature vector are fused to obtain the advertisement joint representation vector corresponding to the historical advertisement. After the above method is executed for each historical advertisement in the training data, the advertisement joint representation vector corresponding to each historical advertisement is obtained. Then the adjacency matrix is ​​input into the model as the original true value of the model (i.e., Group truth), and then based on the original true value, a plurality of triple data consisting of the advertisement joint representation of the historical advertisement are constructed, and then the constructed triple data is used to train the feature extraction model, thereby obtaining the advertisement feature extraction model.

[0103] In an embodiment of the present invention, each triplet of data is composed of three data, which are respectively an advertisement joint representation vector of any historical advertisement in the training data, an advertisement joint representation vector of a historical advertisement associated with the any historical advertisement, and an advertisement joint representation vector of a historical advertisement not associated with the any historical advertisement. For example, if historical advertisement A is associated with historical advertisement B, and historical advertisement A is not associated with historical advertisement C, the advertisement joint representation of historical advertisement A, the advertisement joint representation of historical advertisement B, and the advertisement joint representation of historical advertisement C can be combined into a triplet of data. Among them, for any two historical advertisements, if the correlation coefficient between the two historical advertisements is not 0, it is determined that the two historical advertisements are associated, and if the correlation coefficient between the two historical advertisements is 0, it is determined that the two historical advertisements are not associated.

[0104] In an embodiment of the present invention, the advertisement feature extraction model obtained by training in the above manner takes the advertisement material corresponding to the advertisement as input and the advertisement joint representation vector of the advertisement as output. Since the adjacency matrix is ​​used as the Group truth during model training, the advertisement joint representation vector output by the advertisement feature extraction model is an embedding vector. The property of the embedding vector is that it can make objects corresponding to vectors with similar distances have similar meanings, that is, the advertisement joint representation vectors corresponding to similar advertisements are also similar. Therefore, the similarity between the advertisement joint representation vectors can reflect the similarity between the corresponding advertisements. The more similar the advertisement joint representation vectors of two advertisements are, the more similar the two advertisements are.

[0105] See also Figure 3 , is a schematic diagram of an advertisement push device provided by an embodiment of the present invention, such as Figure 3 As shown, the advertisement push device provided by the embodiment of the present invention may include:

[0106] An acquisition module 301 is used to acquire an advertisement material corresponding to each first advertisement among a plurality of first advertisements;

[0107] The advertisement vector determination module 302 is used to input the corresponding advertisement material into a pre-trained advertisement feature extraction model for each of the first advertisements to obtain an output result, where the output result is an advertisement joint representation vector of the first advertisement;

[0108] A user vector determination module 303, configured to determine an advertisement preference vector of a target user according to a second advertisement clicked by the target user;

[0109] A calculation module 304, configured to determine a matching degree between an advertisement joint representation vector of each of the plurality of first advertisements and the advertisement preference vector;

[0110] The push module 305 is used to push the first advertisements whose matching degree meets the preset conditions among the multiple first advertisements to the target user.

[0111] In an embodiment of the present invention, the training of the advertisement feature extraction model may include:

[0112] Acquire training data, wherein the training data includes advertising materials of a plurality of historical advertisements;

[0113] Obtaining click logs of the plurality of historical advertisements;

[0114] Based on the click log, construct an adjacency matrix between the multiple historical advertisements;

[0115] The preset feature extraction model is trained using the training data and the adjacency matrix to obtain an advertisement feature extraction model.

[0116] In the embodiment of the present invention, constructing an adjacency matrix between the plurality of historical advertisements based on the click log includes:

[0117] The click log contains the ID of the clicked advertisement and the ID of the user who clicked the advertisement. According to the advertisement click log, the plurality of historical advertisements are clustered, and the historical advertisements clicked by the same user are classified into one category.

[0118] Generate a corresponding relationship between the user and the historical advertisement according to the clustering result;

[0119] Based on the corresponding relationship, construct a correlation relationship graph between the multiple historical advertisements, wherein the correlation relationship graph has the historical advertisements as vertices, the correlation relationships between the historical advertisements as edges, and the correlation coefficients between the historical advertisements as edge weights;

[0120] An adjacency matrix corresponding to the correlation graph is generated.

[0121] In the embodiment of the present invention, the correlation coefficient between the historical advertisements is determined in the following manner:

[0122] The initial correlation coefficient between any two of the historical advertisements is determined to be 0. For any two of the historical advertisements, if it is determined according to the corresponding relationship that any two of the historical advertisements correspond to the same user, the correlation coefficient between the two historical advertisements is increased by 1.

[0123] In the embodiment of the present invention, the method of training a preset feature extraction model using the training data and the adjacency matrix to obtain an advertisement feature extraction model includes:

[0124] The preset feature extraction model includes a feature extractor, a multi-layer perceptron and a triple loss function;

[0125] The training data is input into the feature extractor for feature extraction and then passed through the multi-layer perceptron for feature understanding to obtain an advertisement joint representation vector of the training data;

[0126] Using the adjacency matrix as the original true value, constructing a plurality of triplet data consisting of joint representations of advertisements of the historical advertisements;

[0127] The feature extraction model is trained using a plurality of the triplet data to obtain an advertisement feature extraction model.

[0128] In the embodiment of the present invention, the advertisement vector determination module 302 is specifically configured to:

[0129] Determining, according to the advertisement click log of the target user, a second advertisement clicked by the target user;

[0130] Obtaining an advertisement joint representation vector of the second advertisement;

[0131] The advertisement joint representation vector of the second advertisement is used as the advertisement preference vector of the user.

[0132] In an embodiment of the present invention, the device may further include ( Figure 3 Not shown):

[0133] A storage module, used to store the advertisement joint representation vector output by the advertisement feature extraction model and the corresponding advertisement in a preset location;

[0134] The acquiring the advertisement joint representation vector of the second advertisement includes:

[0135] Detecting whether an advertisement joint representation vector corresponding to the second advertisement is stored in the preset position;

[0136] If it is determined that the advertisement joint characterization vector corresponding to the second advertisement is stored in the preset position, acquiring the advertisement joint characterization vector corresponding to the second advertisement stored in the preset position;

[0137] If it is determined that the advertisement joint representation vector corresponding to the second advertisement is not stored in the preset position, the advertisement material corresponding to the second advertisement is obtained, and the advertisement material corresponding to the second advertisement is input into the advertisement feature extraction model to obtain the corresponding output result, and the output result is used as the advertisement joint representation vector of the second advertisement.

[0138] See also Figure 4 , is a structural diagram of an electronic device provided in another embodiment of the present application.

[0139] like Figure 4 As shown, the electronic device provided in this embodiment includes: at least one processor 401, a memory 402, at least one network interface 403 and other user interfaces 404. The various components in the electronic device 400 are coupled together through a bus system 405. It can be understood that the bus system 405 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 405 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, various buses are marked as bus systems 405 in Figure X.

[0140] The user interface 404 may include a display, a keyboard, or a pointing device (eg, a mouse, a trackball, a touch pad, or a touch screen).

[0141] It can be understood that the memory 402 in the embodiment of the present invention can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DRRAM). The memory 402 described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0142] In some implementations, the memory 402 stores the following elements, executable units or data structures, or a subset thereof, or an extended set thereof: an operating system 4021 and a second application 4022 .

[0143] The operating system 4021 includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., which are used to implement various basic services and process hardware-based tasks. The second application 4022 includes various second applications, such as a media player (Media Player), a browser (Browser), etc., which are used to implement various application services. The program for implementing the method of the embodiment of the present invention can be included in the second application 4022.

[0144] In the embodiment of the present invention, by calling the program or instruction stored in the memory 402, specifically, the program or instruction stored in the second application 4022, the processor 401 is used to execute the method steps provided by each method embodiment, for example, including:

[0145] Obtain multiple advertising materials corresponding to multiple first advertisements;

[0146] For each of the first advertisements, input the corresponding advertisement material into a pre-trained advertisement feature extraction model to obtain an output result, where the output result is an advertisement joint representation vector of the first advertisement;

[0147] determining an advertisement preference vector of the target user according to a second advertisement clicked by the target user;

[0148] determining a degree of match between an advertisement joint representation vector of each first advertisement in the plurality of first advertisements and the advertisement preference vector;

[0149] The first advertisements whose matching degree meets the preset condition among the plurality of the first advertisements are pushed to the target user.

[0150] The method disclosed in the above embodiment of the present invention can be applied to the processor 401, or implemented by the processor 401. The processor 401 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit in the processor 401 or the instruction in the form of software. The above processor 401 can be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present invention can be directly embodied as a hardware decoding processor to execute, or the hardware and software units in the decoding processor can be executed. The software unit can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 402, and the processor 401 reads the information in the memory 402 and completes the steps of the above method in combination with its hardware.

[0151] It is understood that the embodiments described herein may be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit may be implemented in one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPDevice, DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), general purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions of the present application, or a combination thereof.

[0152] For software implementation, the technology of this article can be implemented by a unit that performs the functions of this article. The software code can be stored in a memory and executed by a processor. The memory can be implemented in the processor or outside the processor.

[0153] The embodiment of the present invention also provides a storage medium (computer-readable storage medium). The storage medium here stores one or more programs. The storage medium may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a read-only memory, a flash memory, a hard disk or a solid-state drive; the memory may also include a combination of the above-mentioned types of memory.

[0154] When one or more programs in the storage medium can be executed by one or more processors, the above-mentioned advertisement push method executed on the electronic device side can be implemented.

[0155] The processor is used to execute the program of the advertisement push method stored in the memory to implement the following steps of the advertisement push method executed on the electronic device side:

[0156] Obtain multiple advertising materials corresponding to multiple first advertisements;

[0157] For each of the first advertisements, input the corresponding advertisement material into a pre-trained advertisement feature extraction model to obtain an output result, where the output result is an advertisement joint representation vector of the first advertisement;

[0158] determining an advertisement preference vector of the target user according to a second advertisement clicked by the target user;

[0159] determining a degree of match between an advertisement joint representation vector of each first advertisement in the plurality of first advertisements and the advertisement preference vector;

[0160] The first advertisements whose matching degree meets the preset condition among the plurality of the first advertisements are pushed to the target user.

[0161] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0162] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0163] It should be noted that, in the description of this application, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of this application, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0164] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0165] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0166] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0167] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0168] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0169] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0170] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in the field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. An advertisement push method, characterized in that: include: Obtaining an advertisement material corresponding to each of the first advertisements among the plurality of first advertisements; For each of the first advertisements, input the corresponding advertisement material into a pre-trained advertisement feature extraction model to obtain an output result, where the output result is an advertisement joint representation vector of the first advertisement; determining an advertisement preference vector of the target user according to a second advertisement clicked by the target user; determining a degree of match between an advertisement joint representation vector of each of the plurality of first advertisements and the advertisement preference vector; Pushing the first advertisements whose matching degree meets the preset condition among the plurality of the first advertisements to the target user; The training of the advertisement feature extraction model includes: Acquire training data, wherein the training data includes advertising materials of a plurality of historical advertisements; Obtaining click logs of the plurality of historical advertisements; Based on the click log, construct an adjacency matrix between the multiple historical advertisements; The preset feature extraction model is trained using the training data and the adjacency matrix to obtain an advertisement feature extraction model.

2. The method according to claim 1, characterized in that The step of constructing an adjacency matrix between the plurality of historical advertisements based on the click log includes: The click log contains the ID of the clicked advertisement and the ID of the user who clicked the advertisement. According to the click log, the plurality of historical advertisements are clustered, and the historical advertisements clicked by the same user are classified into one category. Generate a corresponding relationship between the user and the historical advertisement according to the clustering result; Based on the corresponding relationship, construct a correlation relationship graph between the multiple historical advertisements, wherein the correlation relationship graph has the historical advertisements as vertices, the correlation relationships between the historical advertisements as edges, and the correlation coefficients between the historical advertisements as edge weights; An adjacency matrix corresponding to the correlation graph is generated.

3. The method according to claim 2, characterized in that The correlation coefficients between the historical advertisements are determined in the following manner: The initial correlation coefficient between any two of the historical advertisements is determined to be 0. For any two of the historical advertisements, if it is determined according to the corresponding relationship that any two of the historical advertisements correspond to the same user, the correlation coefficient between the two historical advertisements is increased by 1.

4. The method according to claim 1, characterized in that The step of training a preset feature extraction model using the training data and the adjacency matrix to obtain an advertisement feature extraction model includes: The preset feature extraction model includes a feature extractor, a multi-layer perceptron and a triple loss function; Inputting the training data into the feature extractor for feature extraction and then passing it through the multi-layer perceptron for feature understanding to obtain an advertisement joint representation vector of the training data; Using the adjacency matrix as the original true value, constructing a plurality of triplet data consisting of joint advertisement representations of the historical advertisements; The feature extraction model is trained using a plurality of the triplet data to obtain an advertisement feature extraction model.

5. The method according to claim 1, characterized in that The step of determining the advertisement preference vector of the target user according to the second advertisement clicked by the target user includes: Determining, according to the advertisement click log of the target user, a second advertisement clicked by the target user; Obtaining an advertisement joint representation vector of the second advertisement; The advertisement joint representation vector of the second advertisement is used as the advertisement preference vector of the user.

6. The method according to claim 4, characterized in that The method further comprises: The advertisement joint representation vector output by the advertisement feature extraction model and the corresponding advertisement are stored in a preset location in correspondence with each other; The acquiring the advertisement joint representation vector of the second advertisement includes: Detecting whether an advertisement joint representation vector corresponding to the second advertisement is stored in the preset position; If it is determined that the advertisement joint characterization vector corresponding to the second advertisement is stored in the preset position, acquiring the advertisement joint characterization vector corresponding to the second advertisement stored in the preset position; If it is determined that the advertisement joint representation vector corresponding to the second advertisement is not stored in the preset position, the advertisement material corresponding to the second advertisement is obtained, and the advertisement material corresponding to the second advertisement is input into the advertisement feature extraction model to obtain the corresponding output result, and the output result is used as the advertisement joint representation vector of the second advertisement.

7. An advertisement pushing device, characterized in that: include: An acquisition module, used to acquire an advertisement material corresponding to each of a plurality of first advertisements; an advertisement vector determination module, configured to input the corresponding advertisement material into a pre-trained advertisement feature extraction model for each of the first advertisements to obtain an output result, wherein the output result is an advertisement joint representation vector of the first advertisement; A user vector determination module, configured to determine an advertisement preference vector of a target user according to a second advertisement clicked by the target user; a calculation module, configured to determine a degree of match between an advertisement joint representation vector of each of the plurality of first advertisements and the advertisement preference vector; A push module, configured to push the first advertisements whose matching degree meets a preset condition among the plurality of the first advertisements to the target user; The training of the advertisement feature extraction model includes: Acquire training data, wherein the training data includes advertising materials of a plurality of historical advertisements; Obtaining click logs of the plurality of historical advertisements; Based on the click log, construct an adjacency matrix between the multiple historical advertisements; The preset feature extraction model is trained using the training data and the adjacency matrix to obtain an advertisement feature extraction model.

8. An electronic device, characterized in that: include: A processor and a memory, wherein the processor is used to execute an advertisement push program stored in the memory to implement the advertisement push method described in any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the advertisement push method described in any one of claims 1-6.

Images