Information delivery method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of computers, in particular to an information delivery method and device, electronic equipment and a storage medium, and aims to improve the information delivery accuracy. The method comprises the following steps: in response to an information delivery request for a target object, obtaining historical behavior characteristics of a plurality of pieces of historical delivery information and candidate content characteristics of each piece of candidate delivery information; performing feature intersection on each candidate content feature and each historical behavior feature to obtain an information intersection feature of each candidate delivery information; obtaining prediction feedback information based on the information cross features; and based on each piece of prediction feedback information, screening out target delivery information meeting a preset delivery condition from each piece of candidate delivery information, and delivering the target delivery information. According to the method, the influence of the sequential relationship of the historical behaviors on the feedback information of the object is fully combined, so that the current behavior preference of the object can be more accurately extracted, the target delivery information screened based on the obtained predicted feedback information is more accurate, and the information delivery accuracy is improved.

Description

Background Art

[0002] With the development of Internet technology, people can obtain a variety of information through network platforms and can provide some feedback on the information they are interested in. For example, people can click, appreciate, share, collect and download the information they are interested in, and make conversions such as purchases. When people obtain information, network platforms can deliver information that they may be interested in based on their preferences. Accurate information delivery can reduce the time people spend looking for the information they need, improve the user experience, and help information delivery parties to more quickly discover potential objects, thereby improving the efficiency of information delivery.

[0003] In the related art, the object's feedback behavior on the candidate delivery information is usually predicted based on the relationship between the attribute information of the candidate delivery information and the attribute information of the historical delivery information, as well as the object's historical feedback behavior on the historical delivery information. For example, if the object has clicked on multiple food information, it is predicted that the object has a higher click rate on food information.

[0004] However, the feedback behavior of the subject for a certain type of information often changes over time, that is, the subject may be interested in a certain type of information in a period of time, but not interested in it in a later period of time. Therefore, it is difficult to accurately analyze the subject's current preferences based solely on the attribute information and historical feedback behavior of the historical delivery information, and the predicted feedback behavior is prone to deviation. Ultimately, when selecting information from candidate information based on the predicted feedback behavior, it is difficult to select the information that the subject is really interested in, resulting in inaccurate information delivery.

[0005] Therefore, how to improve the accuracy of information delivery has become a problem that needs to be solved urgently. Summary of the invention

[0006] The embodiments of the present application provide an information delivery method, device, electronic device and storage medium to improve the accuracy of information delivery.

[0007] An information delivery method provided in an embodiment of the present application includes:

[0008] In response to an information delivery request for a target object, respectively obtain historical behavior features of each of a plurality of historical delivery information for the target object, and candidate content features of each of the candidate delivery information; wherein the historical behavior features at least include: attribute information of the corresponding historical delivery information, historical delivery time, and historical feedback behavior of the target object to the historical delivery information; the candidate content features at least include: attribute information of the corresponding candidate delivery information and an estimated delivery time;

[0009] Perform at least two types of feature cross processing on each candidate content feature and each historical behavior feature to obtain information cross features of each candidate delivery information;

[0010] Based on the cross-features of each information, obtaining the predicted feedback information of each candidate delivery information by the target object;

[0011] Based on the obtained prediction feedback information, target delivery information that meets the preset delivery conditions is screened out from the candidate delivery information, and the target delivery information is delivered to the target object.

[0012] An information delivery device provided in an embodiment of the present application includes:

[0013] an acquisition unit, configured to respectively acquire, in response to an information delivery request for a target object, a historical behavior feature of each of a plurality of historical delivery information for the target object, and a candidate content feature of each of the candidate delivery information; wherein the historical behavior feature at least includes: attribute information of the corresponding historical delivery information, historical delivery time, and historical feedback behavior of the target object to the historical delivery information; and the candidate content feature at least includes: attribute information of the corresponding candidate delivery information and an estimated delivery time;

[0014] A cross unit, used to perform at least two types of feature cross processing on each candidate content feature and each historical behavior feature, respectively, to obtain information cross features of each candidate delivery information;

[0015] A prediction unit, configured to obtain prediction feedback information of the target object for each of the candidate delivery information based on cross-features of each information;

[0016] The screening unit is used to screen out target delivery information that meets the preset delivery conditions from the candidate delivery information based on the obtained prediction feedback information, and deliver the target delivery information to the target object.

[0017] Optionally, the cross unit is specifically used for:

[0018] For each candidate content feature, the following operations are performed respectively:

[0019] A candidate content feature is subjected to two types of feature cross processing with each of the historical behavior features to obtain each time series attention weight and each time series content feature;

[0020] Multiplying each temporal attention weight by the corresponding temporal content feature to obtain each information interaction feature; wherein the temporal attention weight and the corresponding temporal content feature correspond to the same historical behavior feature;

[0021] The information interaction features are concatenated to obtain corresponding information cross features.

[0022] Optionally, the cross unit is specifically used for:

[0023] Perform dot product of the candidate content feature and each of the historical behavior features to obtain each of the temporal attention weights;

[0024] The one candidate content feature is multiplied by each of the historical behavior features to obtain each of the temporal content features.

[0025] Optionally, the acquiring unit is specifically used for:

[0026] The historical behavior features and the candidate content features are obtained in the following manner:

[0027] Encode the attribute information and the historical feedback behavior of a historical delivery information and the attribute information of a candidate delivery information respectively to obtain a historical encoding vector and a content encoding vector;

[0028] Based on the difference between the historical delivery time of the one historical delivery information and the expected delivery time of the one candidate delivery information, respectively encode the historical delivery time and the expected delivery time to obtain a historical time sequence encoding vector and a content time sequence encoding vector;

[0029] The historical behavior feature is obtained based on the historical coding vector and the historical time sequence coding vector, and the candidate content feature is obtained based on the content coding vector and the content time sequence coding vector.

[0030] Optionally, the acquiring unit is specifically used for:

[0031] Adding the historical coding vector and the historical time series coding vector to obtain the historical behavior feature;

[0032] The content coding vector and the content timing coding vector are added together to obtain the candidate content feature.

[0033] Optionally, the prediction unit is specifically used for:

[0034] The cross features are respectively input into the estimation network of the target estimation model to obtain output prediction feedback information.

[0035] Optionally, the prediction unit is specifically used for:

[0036] Acquire an initial training sample set, wherein each initial training sample in the initial training sample set at least includes: a sample historical behavior feature of each of a plurality of sample historical delivery information of a sample object, and a sample content feature of each of the sample delivery information;

[0037] Based on the initial training sample set, the initial estimation model is iteratively trained to obtain the target estimation model.

[0038] Optionally, the prediction unit is specifically used for:

[0039] During one training iteration, the following operations are performed:

[0040] Based on the feature extraction network of the initial estimation model, the content features of each sample in the selected initial training samples are subjected to feature cross processing with the historical behavior features of each sample, so as to obtain the sample cross features of each sample delivery information;

[0041] Inputting the cross features of each sample into the estimation network to perform information estimation, and obtaining sample feedback information of each sample delivery information;

[0042] A target loss function is constructed based on feedback information of each sample, and parameters of the initial estimation model are adjusted based on the target loss function.

[0043] Optionally, the predicted feedback information includes at least one of the following: click rate and conversion rate.

[0044] An electronic device provided in an embodiment of the present application includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any one of the above-mentioned information delivery methods.

[0045] An embodiment of the present application provides a computer-readable storage medium, which includes a computer program. When the computer program is executed on an electronic device, the computer program is used to enable the electronic device to execute the steps of any one of the above-mentioned information delivery methods.

[0046] An embodiment of the present application provides a computer program product, which includes a computer program, and the computer program is stored in a computer-readable storage medium; when a processor of an electronic device reads the computer program from the computer-readable storage medium, the processor executes the computer program, so that the electronic device performs the steps of any one of the above-mentioned information delivery methods.

[0047] The beneficial effects of this application are as follows:

[0048] The embodiments of the present application provide an information delivery method, device, electronic device and storage medium. During the information delivery process, candidate content features of each candidate delivery information are respectively subjected to at least two types of feature cross-processing with historical behavior features of multiple historical delivery information of the target object to obtain each information cross-feature. Through the deep-level feature cross-processing of different types, the expression ability of the obtained information cross-feature for the behavioral preference of the target object can be effectively improved. Moreover, since the candidate content features include the estimated delivery time of the candidate delivery information and the historical behavior features include the historical delivery time of the historical delivery information, the obtained information cross-feature fully captures the temporal relationship of the behavioral preference of the target object in different time periods. Furthermore, based on the information cross-feature, the obtained prediction feedback information can more accurately predict the current behavioral preference of the target object, and the obtained prediction feedback information is more accurate. Finally, based on the obtained prediction feedback information, the selected target delivery information is more accurate, thereby improving the accuracy of information delivery.

[0049] Other features and advantages of the present application will be described in the subsequent description, and partly become apparent from the description, or be understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained by the structures specifically pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0051] Figure 1 An optional schematic diagram of an application scenario in an embodiment of the present application;

[0052] Figure 2 This is a flowchart of an implementation method of information delivery in an embodiment of the present application;

[0053] Figure 3 This is a schematic diagram of an information delivery interface in an embodiment of the present application;

[0054] Figure 4 A schematic diagram of an encoding method in an embodiment of the present application;

[0055] Figure 5 is a schematic diagram of a feature cross processing method in an embodiment of the present application;

[0056] Figure 6 A schematic diagram of a click rate prediction process in an embodiment of the present application;

[0057] Figure 7A schematic diagram of the structure of a target estimation model in an embodiment of the present application;

[0058] Figure 8 This is a model performance comparison diagram in an embodiment of the present application;

[0059] Fig. 9 A heat map visualization result of mutual information in an embodiment of the present application;

[0060] Fig. 10A This is a rendering of the first model ablation experiment in the embodiment of the present application;

[0061] Fig. 10B This is a rendering of the second model ablation experiment in the embodiment of the present application;

[0062] Fig. 10C This is a rendering of the third model ablation experiment in the embodiment of the present application;

[0063] Fig. 10D This is a rendering of the fourth model ablation experiment in the embodiment of the present application;

[0064] Fig.11 A logical diagram of an information delivery method in an embodiment of the present application;

[0065] Fig.12 This is a schematic diagram of the structure of an information delivery device in an embodiment of the present application;

[0066] Fig.13 A schematic diagram of a hardware structure of an electronic device to which an embodiment of the present application is applied;

[0067] Fig.14 The present invention is a schematic diagram of the hardware structure of another electronic device to which the embodiments of the present application are applied. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the technical solution of the present application, rather than all of the embodiments. Based on the embodiments recorded in the application documents, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the technical solution of the present application.

[0069] The following is an introduction to some concepts involved in the embodiments of the present application.

[0070] Delivery information: refers to the general term for all information that is posted on the Internet and can be viewed and read. It can also be called network information, such as videos, news, e-books, online articles, information, and forum posts, etc. It can also refer to commodities. The delivery information in the embodiments of the present application is mainly divided into historical delivery information and candidate delivery information. Among them, candidate delivery information refers to information that has not yet been delivered to the target object. The target delivery information recommended to the target object is screened from multiple candidate delivery information by predicting the predicted feedback information of the target object. Historical delivery information refers to information that has been delivered to the target object. Therefore, the feedback on the candidate delivery information can be predicted based on the feedback of the target object on the historical delivery information.

[0071] Prediction feedback information: refers to the feedback of the target object predicted by information cross-features on the candidate delivery information, which may include click-through rate (CTR), conversion rate (CVR), post-click conversion rate and other information.

[0072] Click-through rate: It is defined as the ratio of clicked delivery information to the delivery information in the effective exposure stage, and is used to represent the probability of delivery information being clicked from the effective exposure stage.

[0073] Conversion rate: It is defined as the ratio of converted delivery information to clicked delivery information, and is used to represent the probability of delivery information being converted from being clicked.

[0074] Deep Neural Network (DNN): is a neural network with multiple hidden layers, and is currently the most popular and effective click-through rate prediction model. In the embodiment of the present application, after obtaining the information cross-feature, the information cross-feature can be input into the DNN for click-through rate prediction.

[0075] Embedding: A method of converting a discrete variable into a continuous vector representation. DNN generally converts discrete features into Embedding vectors, which are then concatenated or processed in other ways as the input layer of the model. The historical coding vector, content coding vector, historical time series coding vector, and content time series coding vector in the embodiments of the present application can be obtained through the Embedding layer.

[0076] Representation layer: It is a module that combines all feature vectors in the neural network model and generates a high-dimensional vector, which is input into the DNN network that predicts user click-through rate. Currently, a widely used feature combination method is to concatenate all feature vectors.

[0077] Attention unit: It is a neural network structure that has the ability to focus on a subset of its input (or features) and is often used as an important structure to improve the expressiveness and interpretability of the model.

[0078] The embodiments of the present application relate to artificial intelligence (AI) and machine learning technology, and are designed based on speech technology, natural language processing technology and machine learning (ML) in artificial intelligence.

[0079] Artificial intelligence is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence.

[0080] Artificial intelligence is the study of the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning and decision-making. Artificial intelligence technology mainly includes computer vision technology, natural language processing technology, and machine learning / deep learning. With the research and progress of artificial intelligence technology, artificial intelligence has been studied and applied in many fields, such as common smart homes, smart customer service, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, robots, smart medical care, etc. It is believed that with the development of technology, artificial intelligence will be applied in more fields and play an increasingly important role.

[0081] Machine learning is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Compared with data mining, which seeks mutual characteristics between big data, machine learning focuses more on algorithm design, allowing computers to automatically "learn" rules from data and use the rules to predict unknown data.

[0082] Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning and other technologies. Reinforcement learning (RL), also known as reinforcement learning, evaluation learning or enhanced learning, is one of the paradigms and methodologies of machine learning. It is used to describe and solve the problem of how an agent can maximize its rewards or achieve specific goals through learning strategies during its interaction with the environment.

[0083] In the embodiment of the present application, the training part of the target estimation model involves the technical field of machine learning. In the training part, the initial estimation model is trained by the machine learning technology, and the model parameters are continuously adjusted by the optimization algorithm to obtain the target estimation model.

[0084] The following is a brief introduction to the design concept of the embodiment of the present application:

[0085] With the development of Internet technology, people can obtain a variety of information through network platforms and can provide some feedback on the information they are interested in. For example, people can click, appreciate, share, collect and download the information they are interested in, and make conversions such as purchases. When people obtain information, network platforms can deliver information that they may be interested in based on their preferences. Accurate information delivery can reduce the time people spend looking for the information they need, improve the user experience, and help information delivery parties to more quickly discover potential objects, thereby improving the efficiency of information delivery.

[0086] In the related art, the object's feedback behavior on the candidate delivery information is usually predicted based on the relationship between the attribute information of the candidate delivery information and the attribute information of the historical delivery information, as well as the object's historical feedback behavior on the historical delivery information. For example, if the object has clicked on multiple food information, it is predicted that the object has a higher click rate on food information.

[0087] However, the feedback behavior of the subject for a certain type of information often changes over time, that is, the subject may be interested in a certain type of information in a period of time, but not interested in it in a later period of time. Therefore, it is difficult to accurately analyze the subject's current preferences based solely on the attribute information and historical feedback behavior of the historical delivery information, and the predicted feedback behavior is prone to deviation. Ultimately, when selecting information from candidate information based on the predicted feedback behavior, it is difficult to select the information that the subject is really interested in, resulting in inaccurate information delivery.

[0088] Therefore, how to improve the accuracy of information delivery has become a problem that needs to be solved urgently.

[0089] In view of this, the embodiments of the present application provide an information delivery method, device, electronic device and storage medium. During the information delivery process, the candidate content features of each candidate delivery information are respectively subjected to at least two types of feature cross-processing with the historical behavior features of multiple historical delivery information of the target object to obtain each information cross-feature. Through the deep-level feature cross-crossing of different types, the expression ability of the obtained information cross-feature for the behavioral preferences of the target object can be effectively improved. Moreover, since the candidate content features include the estimated delivery time of the candidate delivery information, and the historical behavior features include the historical delivery time of the historical delivery information, the obtained information cross-feature fully captures the temporal relationship of the behavioral preferences of the target object in different time periods. Furthermore, based on the information cross-feature, the obtained predicted feedback information can more accurately predict the current behavioral preferences of the target object, and the obtained predicted feedback information is more accurate. Finally, based on the obtained predicted feedback information, the screened target delivery information is more accurate, thereby improving the accuracy of information delivery.

[0090] The preferred embodiments of the present application are described below in conjunction with the drawings in the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application. In addition, the embodiments and features in the embodiments of the present application may be combined with each other if there is no conflict.

[0091] like Figure 1 As shown, it is a schematic diagram of an application scenario of an embodiment of the present application. The application scenario diagram includes two terminal devices 110 and a server 120.

[0092] In the embodiments of the present application, the terminal device includes but is not limited to mobile phones, tablet computers, laptop computers, desktop computers, e-book readers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals and other devices; the terminal device may be installed with a client related to information delivery, which may be software (such as a browser, social software, etc.), or a web page, a small program, etc. The server is a background server corresponding to the software or web page, small program, etc., or a server specifically used for information delivery, which is not specifically limited in the present application. The server may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and basic cloud computing services such as big data and artificial intelligence platforms.

[0093] It should be noted that the information delivery method in the embodiment of the present application can be executed by an electronic device, and the electronic device can be a server or a terminal device, that is, the method can be executed by the server or the terminal device alone, or by the server and the terminal device together. For example, when executed by the server and the terminal device together, the terminal device sends an information delivery request for the target object to the server, and the server responds to the information delivery request, and obtains the historical behavior characteristics of each of the multiple historical delivery information for the target object, and the candidate content characteristics of each candidate delivery information; each candidate content feature is subjected to at least two types of feature cross-processing with each historical behavior feature, and the information cross-feature of each candidate delivery information is obtained; based on each information cross-feature, the target object's respective predicted feedback information for each candidate delivery information is obtained; based on each predicted feedback information obtained, the target delivery information that meets the preset delivery conditions is screened out from each candidate delivery information, and the target delivery information is sent to the terminal device, and the terminal device delivers the target delivery information to the target object.

[0094] In an optional implementation, the terminal device and the server may communicate via a communication network.

[0095] In an optional implementation, the communication network is a wired network or a wireless network.

[0096] It should be noted that Figure 1 The figure is only an example. In fact, the number of terminal devices and servers is not limited and is not specifically limited in the embodiments of the present application.

[0097] In an embodiment of the present application, when there are multiple servers, the multiple servers can form a blockchain, and the servers are nodes on the blockchain; as in the information delivery method disclosed in the embodiment of the present application, the historical delivery information and candidate delivery information involved therein can be saved on the blockchain, etc.

[0098] In addition, the embodiments of the present application can be applied to various scenarios, including not only information delivery scenarios, but also including but not limited to cloud technology, artificial intelligence, smart transportation, assisted driving and other scenarios.

[0099] It is understandable that in the specific implementation of this application, data related to user information is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0100] The following describes the information delivery method provided by the exemplary embodiment of the present application in combination with the application scenarios described above and with reference to the accompanying drawings. It should be noted that the above application scenarios are only shown to facilitate understanding of the spirit and principles of the present application, and the implementation methods of the present application are not limited in this regard.

[0101] like Figure 2 As shown, it is an implementation flow chart of an information delivery method provided in an embodiment of the present application. Taking the execution subject as a server as an example, the specific implementation process of the method includes the following steps S21-S24:

[0102] S21: In response to the information delivery request for the target object, the server obtains the historical behavior characteristics of each of the multiple historical delivery information for the target object and the candidate content characteristics of each of the candidate delivery information;

[0103] The target object may refer to a user or an account used by a user, etc. The following mainly takes the target object as a user as an example for detailed introduction.

[0104] The information delivery request can be sent when the user opens an application (APP) or website, or when the user is browsing, and is not specifically limited here. Figure 3 As shown in the figure, when a user opens the Moments and refreshes the list, an advertisement can be placed on the Moments display interface. Online advertising is the most direct and transparent way for most Internet companies to realize traffic monetization. When an advertisement is displayed to a user or a user clicks on the advertisement or even activates the APP, places an order, etc., the system will automatically deduct fees from the advertiser, thereby realizing traffic monetization.

[0105] Among them, the historical behavior characteristics at least include the attribute information of the corresponding historical delivery information, the historical delivery time, and the historical feedback behavior of the target object to the historical delivery information. The attribute information may include the category, content and identification information of the historical delivery information. The historical delivery time is the time when the historical delivery information is delivered to the target object. The historical feedback behavior may include: yes / no click, yes / no download, yes / no purchase browsing time and other behaviors. The historical feedback behavior can represent the target object's interest in the historical delivery information. The historical behavior characteristics may also include personal attribute information such as the target object's age and gender, which can assist in further judging the target object's interests and hobbies.

[0106] For example, the attribute information of Advertisement 1 includes: category "home appliances", advertising ID "00011", historical delivery time "15:01, December 31, 2022", and historical feedback behavior "click"; the attribute information of Advertisement 2 includes: category "books", advertising ID "01251", historical delivery time "2:01:20, January 31, 2023", and historical feedback behavior "purchase".

[0107] The candidate content features include at least the attribute information and estimated delivery time of the corresponding candidate delivery information. The attribute information of the candidate delivery information may include category, content, and identification information, etc. The estimated delivery time is the time when the candidate delivery information is delivered to the target object. For example, the attribute information of Advertisement 3 includes: category "books", advertisement ID "11341", and estimated delivery time "15:15:32, February 11, 2023".

[0108] By mining the connection between historical behavior features and candidate behavior features, it is possible to determine whether the target object will perform operations on the candidate delivery information similar to the historical delivery information, and then determine whether to deliver the candidate delivery information.

[0109] The historical behavior features may directly include attribute information, historical delivery time, and historical feedback behavior. The candidate content features may directly include attribute information and estimated delivery time. Furthermore, the above information may be converted into an embedding vector through the Embedding layer.

[0110] Optionally, obtain historical behavior features and candidate content features through the following steps:

[0111] S31: Encode the attribute information and the historical feedback behavior of a historical delivery information and the attribute information of a candidate delivery information respectively to obtain a historical coding vector and a content coding vector;

[0112] S32: Based on the difference between the historical delivery time of a historical delivery information and the expected delivery time of a candidate delivery information, respectively encode the historical delivery time and the expected delivery time to obtain a historical time sequence encoding vector and a content time sequence encoding vector;

[0113] S33: Based on the historical coding vector and the historical time series coding vector, obtain the historical behavior feature, and based on the content coding vector and the content time series coding vector, obtain the candidate content feature.

[0114] Specifically, for each historical delivery information, the attribute information and historical feedback behavior are encoded into a historical coding vector through the Embedding layer, and for each candidate delivery information, the attribute information is encoded into a content coding vector through the Embedding layer. For the historical delivery time and the expected delivery time, considering that the temporal relationship between the historical delivery information and the candidate delivery information should be explicitly modeled, position coding is introduced to model the temporal relationship. The historical delivery time and the expected delivery time are input into the Embedding layer, and the corresponding position coding vector (historical temporal coding vector and content temporal coding vector) can be directly obtained. For example, the historical delivery time "January 1st 20:00" is encoded as "010120", and the expected delivery time "March 25th 21:00" is encoded as "032521".

[0115] For historical coding vectors and historical timing coding vectors, historical behavior features can be obtained by adding, multiplying, dot-producting, concatenating, and other operations on the above vectors. For content coding vectors and content timing coding vectors, candidate content features can also be obtained by adding, multiplying, dot-producting, concatenating, and other operations on the above vectors.

[0116] For example, after concatenating the historical coding vector [0 0 1 1 0 2 3] and the historical time series coding vector [1 1 4 7 8 5 2], we obtain the historical behavior feature [0 0 1 1 0 2 3 1 1 4 7 8 5 2].

[0117] like Figure 4 As shown, it is a schematic diagram of a coding method in an embodiment of the present application, wherein "shopping, 0011, click", "21:00 on January 23, 2023", "shopping, 1235", and "17:00 on August 28, 2023" are encoded respectively through the Embedding layer to obtain historical coding vector 1, historical time series coding vector 1, content coding vector 1, and content time series coding vector 1, and the historical coding vector 1 and the historical time series coding vector 1 are added to obtain the historical behavior feature 1, and the content coding vector 1 and the content time series coding vector 1 are added to obtain the candidate content feature 1.

[0118] Based on the above method, position encoding is introduced to directly encode the historical delivery time and the expected delivery time. Compared with the use of complex networks such as Recurrent Neural Network (RNN) and Long Short-Term Memory Network (Long Short-Term Memory) for modeling, it can reduce the calculation time, improve the calculation efficiency, reduce the coding complexity, and improve the model interpretability.

[0119] Optionally, step S32 may be implemented as the following steps:

[0120] The historical coding vector and the historical time series coding vector are added together to obtain the historical behavior feature; the content coding vector and the content time series coding vector are added together to obtain the candidate content feature.

[0121] Specifically, by adding vectors, comprehensive features can be obtained. For example, the historical coding vector [1 0 0 1 1] and the historical temporal coding vector [0 1 0 0 1], [1 0 0 1 1] ⊕ [0 1 0 0 1] = [1 1 0 1 2], that is, the historical behavior feature obtained is [1 1 0 1 2], the content coding vector [1 2 3 4 5] and the content temporal coding vector [1 0 0 0 1], [1 2 3 4 5] ⊕ [1 0 0 0 1] = [2 3 4 5 6], that is, the historical behavior feature obtained is [2 3 4 5 6]. Where ⊙ represents the multiplication of the elements at the corresponding positions (element-wise multiplication operation).

[0122] S22: The server performs at least two types of feature cross processing on each candidate content feature and each historical behavior feature, respectively, to obtain information cross features of each candidate delivery information;

[0123] Specifically, at least two types of feature cross processing are performed on each candidate content feature and each historical behavior feature. The feature cross processing may be performed in the form of addition, subtraction, dot multiplication, dot product, cross product, etc., which is not specifically limited here.

[0124] For example, for candidate content feature 1, perform dot product of candidate content feature 1 and historical behavior features 1-5 respectively to obtain intermediate features 1-5, then add candidate content feature 1 and historical behavior features 1-5 respectively to obtain intermediate features 6-10, and concatenate intermediate features 1-10 to obtain the information cross-features of the candidate delivery information corresponding to candidate content feature 1.

[0125] For another example, cross-product the candidate content feature 2 with the historical behavior features 6 and 7 respectively to obtain intermediate features 11 and 12, dot-product the candidate content feature 2 with the historical behavior features 6 and 7 respectively to obtain intermediate features 13 and 14, multiply the intermediate feature 11 and the intermediate feature 13 to obtain the intermediate feature 15, and multiply the intermediate feature 12 and the intermediate feature 14 to obtain the intermediate feature 16, and concatenate the intermediate features 15 and 16 to obtain the information cross-feature of the candidate delivery information corresponding to the candidate content feature 1.

[0126] It should be noted that the above only takes two feature cross-processing methods of candidate content features as an example to introduce the feature cross-processing method in the embodiment of the present application. In fact, other feature cross-processing methods can also be used, which are not specifically limited here.

[0127] Optionally, in step S22, the following operations are performed for each candidate content feature:

[0128] S221: Perform two types of feature cross processing on a candidate content feature and each historical behavior feature to obtain each time series attention weight and each time series content feature;

[0129] S222: multiplying each temporal attention weight by the corresponding temporal content feature to obtain each information interaction feature;

[0130] S223: Concatenate the information interaction features to obtain corresponding information cross features.

[0131] Specifically, the types of feature cross-processing can be addition, subtraction, dot multiplication, dot product, and cross product. Feature cross-processing can be performed based on any two of the above types. After the first type of feature cross-processing, the temporal attention weight is obtained, and after the second type of feature cross-processing, the temporal content weight is obtained.

[0132] The temporal attention weight and the corresponding temporal content feature correspond to the same historical behavior feature. For example, the candidate content feature 1 and the historical behavior feature 1 are subjected to the first type of feature cross-processing to obtain the temporal attention weight 1, and the candidate content feature 1 and the historical behavior feature 1 are subjected to the second type of feature cross-processing to obtain the temporal content feature 1. Then the temporal attention weight 1 corresponds to the temporal content feature 1.

[0133] Still taking candidate content feature 1 as an example, perform dot product of candidate content feature 1 and historical behavior features 1-5 respectively to obtain temporal attention weights 1-5, perform dot product of candidate content feature 1 and historical behavior features 1-5 respectively to obtain temporal content features 1-5, multiply temporal attention weight 1 and temporal content feature 1 to obtain information interaction feature 1, multiply temporal attention weight 2 and temporal content feature 2 to obtain information interaction feature 2, multiply temporal attention weight 3 and temporal content feature 3 to obtain information interaction feature 3, multiply temporal attention weight 4 and temporal content feature 4 to obtain information interaction feature 4, multiply temporal attention weight 5 and temporal content feature 5 to obtain information interaction feature 5, concatenate information interaction features 1-5 to obtain information cross-features.

[0134] In information recommendation systems, user historical behaviors, such as historical ad click behaviors and historical conversion behaviors, are very important features because they contain the user's potential interests. These features usually contain variable-length feature values, which can be converted into a fixed-dimensional representation. When modeling user historical behaviors, existing models generally use weighted summation to obtain a user historical behavior vector.

[0135] However, these methods have the following problems: (1) When modeling the interaction between user historical behavior (historical delivery information) and target advertisements (candidate delivery content), they often only focus on the user interest extraction layer or one part of the representation layer. There is little work on modeling the interaction relationship in both modules at the same time, which makes the existing methods often lack good expression capabilities; (2) When considering the temporal relationship of user historical behavior, existing methods often use RNN structure to implicitly model, which has low model interpretability and high computational complexity, making it unsuitable for the online environment of advertising systems.

[0136] In the embodiments of the present application, not only the temporal relationship between the historical delivery information and the candidate delivery information is modeled, but also the interactive relationship between the historical delivery information and the candidate delivery information is modeled at the user interest extraction layer (performing the first type of feature cross processing) and the representation layer (performing the second type of feature cross processing), which can effectively improve the model's expression ability.

[0137] Optionally, step S221 may be implemented as follows:

[0138] A candidate content feature is dot-producted with each historical behavior feature to obtain the attention weight of each time series; a candidate content feature is dot-producted with each historical behavior feature to obtain the content feature of each time series.

[0139] Specifically, in the embodiments of the present application, the two types of feature processing methods are dot product and dot multiplication. The order of the two processing methods is not specifically limited. It should be noted that, as described above, the candidate content features and historical behavior features in the present application are actually vectors, so dot product and dot multiplication operations can be performed.

[0140] For example, the candidate delivery information is t, and its vector is represented by v t (content coding vector), the position encoding corresponding to the candidate delivery information is (Content time sequence coding vector), the historical delivery information is i, and its vector is represented by e i (content encoding vector), the position encoding is (historical time series coding vector), where f(x) represents a position coding method, and the candidate content features Historical behavioral characteristics

[0141] like Figure 5 FIG. 1 is a schematic diagram of a feature cross-processing method in an embodiment of the present application. First, the candidate content features are and historical behavioral characteristics After dot product, we get the importance weight of historical delivery information i (Temporal Attention Weight):

[0142]

[0143] Then, the candidate content features and historical behavioral characteristics After point multiplication, we get the interaction vector f TIN (i,t)(time series content characteristics):

[0144]

[0145] Finally, and f TIN (i, t) are multiplied to obtain information interaction features.

[0146] S23: The server obtains the target object's respective prediction feedback information for each candidate delivery information based on the cross-features of each information;

[0147] The predicted feedback information includes at least one of the click-through rate and the conversion rate, and may also include information such as the click-through conversion rate (CTCVR) and the exposure rate. The predicted probability may be a probability value ranging from 0 to 1. The larger the value corresponding to the candidate delivery information, the greater the possibility that the target object is interested in the candidate delivery information after the candidate delivery information is recommended to the target object.

[0148] S24: The server selects target delivery information that meets preset delivery conditions from candidate delivery information based on the obtained prediction feedback information, and delivers the target delivery information to the target object.

[0149] Specifically, taking the predicted feedback information as the predicted click-through rate as an example, when filtering the target delivery information according to the predicted click-through rate, the predicted click-through rate of each candidate delivery information can be sorted, and the preset delivery condition can be to select a number of candidate delivery information whose sorting results are within a preset order range and deliver them to the target object, for example, selecting the first M candidate delivery information sorted in order from large to small, or selecting the last M candidate delivery information sorted in order from small to large, where M is a positive integer.

[0150] For example, for target object 1, there are 10 candidate delivery information, and the corresponding predicted click rates are: 0.8, 0.3, 0.7, 0.65, 0.55, 0.9, 0.48, 0.15, 0.4, and 0.05.

[0151] If M=3, three candidate delivery information with predicted click-through rates of 0.9, 0.8, and 0.7 respectively are recommended to target object 1.

[0152] Alternatively, N recommended content with predicted click rates greater than a preset probability threshold are selected and recommended to the target object 1, where N is a positive integer.

[0153] For example, if the preset probability threshold is 0.5, five candidate delivery information with predicted click rates of 0.9, 0.8, 0.7, 0.65, and 0.55 respectively are delivered to target object 1.

[0154] It should be noted that the several methods of filtering target delivery information according to predicted click-through rates listed in the above embodiments are only examples. In fact, any method of filtering according to predicted click-through rates is applicable to the embodiments of the present application.

[0155] In an embodiment of the present application, during the information delivery process, the candidate content features of each candidate delivery information are respectively subjected to at least two types of feature cross-processing with the historical behavior features of multiple historical delivery information of the target object to obtain each information cross-feature. Through different types of deep-level feature cross-processing, the expression ability of the obtained information cross-feature on the behavioral preferences of the target object can be effectively improved. Moreover, since the candidate content features include the estimated delivery time of the candidate delivery information, and the historical behavior features include the historical delivery time of the historical delivery information, the obtained information cross-feature fully captures the temporal relationship of the behavioral preferences of the target object in different time periods. Furthermore, based on the information cross-feature, the obtained predictive feedback information can more accurately predict the current behavioral preferences of the target object, and the obtained predictive feedback information is more accurate. Finally, based on the obtained predictive feedback information, the selected target delivery information is more accurate, thereby improving the accuracy of information delivery.

[0156] In an optional implementation, based on the cross-features of each information, the prediction feedback information of the target object for each candidate delivery information is obtained, which can be implemented by using a neural network model. The specific process is as follows:

[0157] The cross-features of each information are respectively input into the estimation network of the target estimation model to obtain the output prediction feedback information.

[0158] The estimation network may be a multi-layer perceptron, a DNN network, etc., which is not specifically limited here. Figure 6 As shown, information cross feature 1, information cross feature 2, information cross feature 3 and information cross feature 4 are respectively input into the multi-layer perceptron, and the corresponding click rates are 0.2, 0.7, 0.6 and 0.1 respectively.

[0159] In the embodiment of the present application, the target estimation model can be composed of three parts, such as Figure 7As shown, the first is the Embedding layer for encoding, which is used to encode the attribute information of the historical delivery information and the historical feedback behavior, the attribute information of the candidate delivery information, the historical delivery time and the expected delivery time, and obtain the historical coding vector, the content coding vector, the historical timing coding vector and the content timing coding vector;

[0160] Then comes the feature extraction network, which can also be called the Temporal Interest Network (TIN). In the TIN, the historical coding vector and the historical temporal coding vector are added to obtain the historical behavior feature, the content coding vector and the content temporal coding vector are added to obtain the candidate content feature, the historical behavior feature and the candidate content feature are dot-producted to obtain the temporal attention weight, the historical behavior feature and the candidate content feature are dot-multiplied to obtain the temporal content feature, and the temporal attention weight and the temporal content feature are multiplied to obtain the information interaction feature;

[0161] The last step is the estimation network, which concatenates the information interaction features to obtain the information cross features, and then inputs the information cross features into the estimation network to obtain the predicted feedback information of the candidate delivery information.

[0162] Optionally, the target estimation model is trained by:

[0163] An initial training sample set is obtained, wherein each initial training sample in the initial training sample set includes at least: sample historical behavior characteristics of each of a plurality of sample historical delivery information of the sample object, and sample content characteristics of each of the sample delivery information; based on the initial training sample set, an initial estimation model is iteratively trained to obtain a target estimation model.

[0164] The initial training samples may also include click-through rate labels of the sample delivery information. The click-through rate labels are determined based on the user's feedback behavior on the sample delivery information. If the user clicks on the sample delivery information, the labeled click-through rate is 1. If the user does not click on the sample delivery information, the labeled click-through rate is 0. These labeled click-through rates are the labels of the training samples. Based on the labels, the training samples can be divided into positive samples or negative samples. The initial estimation model can be trained based on these training samples.

[0165] For example, when user A browses short videos on short video software A, user A will generate feedback behaviors such as clicking or not clicking, liking or not liking for one of the short videos. At this time, user A or user A is the sample object, the short video currently browsed by user A is the sample delivery information, and the short videos previously browsed by user A can be used as sample historical delivery information.

[0166] Optionally, since the Embedding layer currently used for encoding is relatively mature, the Embedding layer used in the target estimation model can directly select the trained one. Therefore, during the iterative training process, the initial estimation model may not include the Embedding layer. During one iterative training process, the following operations are performed:

[0167] Based on the feature extraction network of the initial estimation model, the content features of each sample in the selected initial training samples are cross-processed with the historical behavior features of each sample to obtain the sample cross-features of each sample delivery information; each sample cross-feature is input into the estimation network for information estimation to obtain the sample feedback information of each sample delivery information; the target loss function is constructed based on the feedback information of each sample, and the parameters of the initial estimation model are adjusted based on the target loss function.

[0168] The target loss function may be a cross entropy loss function or other types of loss functions, which are not specifically limited here.

[0169] In an embodiment of the present application, according to actual processing needs, server training to obtain a target estimation model may be a periodic process, and training samples may be regenerated periodically to obtain a target estimation model through training.

[0170] The above describes in detail the input of the target estimation model in the embodiment of the present application, as well as the various parts included in the model. In order to verify the accuracy of the target estimation model prediction, offline verification was performed on data set 1 and data set 2 respectively:

[0171] 1. Comparison of model performance;

[0172] like Figure 8 As shown, this is a comparison chart of the model performance in the embodiments of the present application. Compared with models such as the Gated Recurrent Unit 4 Recommendation (GRU4Rec), the Deep Interest Evolution Network (DIEN), the Deep Session Interest Network (DEEP SESSION INTEREST NETWORK), the Self-Attentive Sequential Recommendation Model (SASRec) and the Bert4Rec, the TIN model (target estimation model) has achieved better logarithmic loss function (Logloss) and correlation coefficient index (GAUC index), verifying the effectiveness of the model.

[0173] right Figure 8The baselines and target estimation models presented in the paper are summarized in the following three points:

[0174] ① Whether to consider the target-aware temporal encoding (TTE) in the model, that is, whether to consider the relationship between the historical delivery time and the expected delivery time;

[0175] ② Whether to consider the interaction between historical behavior and target advertisements (Target-aware Attention, TA) in the user interest extraction module, that is, whether to perform the first feature cross-processing of historical behavior features and candidate content features;

[0176] ③ Whether to consider the interaction between historical behavior and target advertising in the representation part (Target-aware Representation, TR), that is, whether to perform the second feature cross-processing of historical behavior features and candidate content features.

[0177] If included, it is coded as √, otherwise it is coded as ×. Figure 8 As shown in , the target estimation model is the only model that simultaneously considers the temporal relationship of historical behaviors and explicitly models the second-order interaction of historical behavior features and target advertising features in both user interest extraction and representation.

[0178] 2. Model interpretability analysis;

[0179] (1) Definition of mutual information

[0180] Mutual information is used to measure whether a historical behavior i appears in the user click sequence when a given target ad t occurs. i The degree of correlation between the two variables, whether the user finally clicks on the ad y. The higher the mutual information of historical behavior, the more important it is to the target ad. Mutual information is defined as follows:

[0181]

[0182] Furthermore, considering the trend of the mutual information value corresponding to the historical behavior category c(i) changing with time series when the target advertising category c(t) is given, the category-time series mutual information value is defined as follows:

[0183]

[0184] Among them, c(t) and c(i) correspond to the product categories to which the target advertisement and historical behavior belong respectively, and p(i) represents the position of historical behavior i in the click sequence. Take product categories 674 and 558 in dataset 1 as an example and use them as target advertisement categories. Given the target advertisement category, the two-dimensional mutual information values ​​considering both the historical behavior category and the time series are calculated. The heat map visualization results of the mutual information are shown in Figure 2. Fig. 9 As shown in a and b. It can be found that the historical behavior that belongs to the same category as the target advertisement has the highest mutual information, and the mutual information value decays as the time interval between the historical behavior and the target advertisement increases ( Fig. 9 , 1 on the horizontal axis represents the historical behavior closest to the current ad timing).

[0185] The mutual information value under the above given target advertising category is used as the reference value (ground truth), and is compared with the interaction relationship between the historical behaviors learned by different models and the target advertisements, so as to quantify and compare the expressive power of the models.

[0186] (2) Model ablation experiment

[0187] The reasons for the improvement of the model effect are explored through ablation experiments. Specifically, in the ablation experiments in the last three rows of the table above, the necessity of the three core modules in the TIN model is verified. On both datasets, removing any part of the three core components of the TIN model will significantly reduce the model performance.

[0188] Taking product category 674 in dataset 1 as an example, the interactions between historical behavior categories and target advertising categories learned by TIN and three ablation experimental model variants are plotted, and the Pearson correlation coefficient between them and the mutual information is calculated. Fig. 10A -D, Fig. 10A This is the effect diagram of the TIN model.

[0189] exist Fig. 10B The temporal encoding of target perception is removed. Although the model can distinguish different types of historical behaviors, the trend that the importance decays with increasing time intervals is not learned. The reason is that the model does not consider the temporal relationship between historical behaviors.

[0190] exist Fig. 10CThe interaction between historical behaviors and target ads in the user interest extraction module is removed. Although position encoding is still introduced in the model, the temporal relationship is not captured, that is, the importance of historical behaviors at different positions does not show differences. The reason is that the user interest module will pool the interaction vectors between all historical behaviors and target ads when outputting the user's historical behavior vector. During the pooling operation, the temporal information of different historical behaviors is actually disrupted. At this time, only the length information of the historical sequence is retained, and the position encoding does not interact with the historical behavior at the corresponding position. Therefore, in the TIN model, the attention module is indispensable.

[0191] exist Fig. 10D The interaction between the historical behaviors and the target ads in the representation part is removed. Although for most historical behaviors, the model captures the trend that the importance decreases as the position number increases, the decay trend of its importance with respect to the position is not obvious. This shows that after removing the interaction between the historical behaviors and the target ads, the model's expressive power decreases.

[0192] Therefore, it is believed that the TIN model has good interpretability, and each component is indispensable for improving the expressiveness of the model.

[0193] like Fig.11 As shown, it is a logical schematic diagram of an information delivery method in an embodiment of the present application. For the target user Xiao Ming, there are historical behavior features 1-5 corresponding to the historical delivery information 1-5, and candidate content features 1-3 corresponding to the candidate delivery content 1-3. Taking the candidate delivery content 1 as an example, the following process is used to predict the click rate of the candidate delivery content 1:

[0194] In the temporal interest module, dot products are performed on candidate content feature 1 of candidate delivery content 1 with historical behavior features 1-5 to obtain temporal attention weights 1-5, dot products are performed on candidate content feature 1 with historical behavior features 1-5 to obtain temporal content features 1-5, temporal attention weights are multiplied by corresponding temporal content features to obtain information interaction features 1-5, information interaction features 1-5 and other features (such as user age, gender, hobbies, etc.) are concatenated to obtain information cross features, and the information cross features are input into a multi-layer perceptron to obtain a click-through rate of candidate delivery content 1 of 0.7. Correspondingly, based on the above process, a click-through rate of candidate delivery content 2 is obtained of 0.8, and a click-through rate of candidate delivery content 3 is obtained of 0.1. The preset delivery condition is to deliver the content with the highest click-through rate, and candidate delivery content 2 is delivered to user Xiao Ming.

[0195] Based on the same inventive concept, the present application also provides an information delivery device. Fig.12As shown, it is a schematic diagram of the structure of the information delivery device 1200, which may include:

[0196] The acquisition unit 1201 is used to respond to the information delivery request for the target object and respectively acquire the historical behavior characteristics of each of the multiple historical delivery information for the target object and the candidate content characteristics of each of the candidate delivery information; wherein the historical behavior characteristics at least include: the attribute information of the corresponding historical delivery information, the historical delivery time, and the historical feedback behavior of the target object to the historical delivery information; the candidate content characteristics at least include: the attribute information of the corresponding candidate delivery information and the expected delivery time;

[0197] The cross unit 1202 is used to perform at least two types of feature cross processing on each candidate content feature and each historical behavior feature, respectively, to obtain information cross features of each candidate delivery information;

[0198] Prediction unit 1203, used for obtaining prediction feedback information of the target object for each candidate delivery information based on the cross-features of each information;

[0199] The screening unit 1204 is used to screen target delivery information that meets preset delivery conditions from various candidate delivery information based on the obtained prediction feedback information, and deliver the target delivery information to the target object.

[0200] In an embodiment of the present application, during the information delivery process, the candidate content features of each candidate delivery information are respectively subjected to at least two types of feature cross-processing with the historical behavior features of multiple historical delivery information of the target object to obtain each information cross-feature. Through different types of deep-level feature cross-processing, the expression ability of the obtained information cross-feature on the behavioral preferences of the target object can be effectively improved. Moreover, since the candidate content features include the estimated delivery time of the candidate delivery information, and the historical behavior features include the historical delivery time of the historical delivery information, the obtained information cross-feature fully captures the temporal relationship of the behavioral preferences of the target object in different time periods. Furthermore, based on the information cross-feature, the obtained predictive feedback information can more accurately predict the current behavioral preferences of the target object, and the obtained predictive feedback information is more accurate. Finally, based on the obtained predictive feedback information, the selected target delivery information is more accurate, thereby improving the accuracy of information delivery.

[0201] Optionally, the cross-connect unit 1202 is specifically configured to:

[0202] For each candidate content feature, perform the following operations:

[0203] A candidate content feature is cross-processed with each historical behavior feature to obtain the attention weight of each time series and the content feature of each time series;

[0204] Multiply each temporal attention weight by the corresponding temporal content feature to obtain each information interaction feature; wherein the temporal attention weight and the corresponding temporal content feature correspond to the same historical behavior feature;

[0205] The information interaction features are spliced ​​together to obtain the corresponding information cross features.

[0206] Optionally, the cross-connect unit 1202 is specifically configured to:

[0207] Perform dot product of a candidate content feature with each historical behavior feature to obtain the attention weight of each time series;

[0208] A candidate content feature is multiplied by each historical behavior feature to obtain each time series content feature.

[0209] Optionally, the acquiring unit 1201 is specifically configured to:

[0210] The historical behavior features and candidate content features are obtained in the following ways:

[0211] Encode the attribute information and the historical feedback behavior of a historical delivery information and the attribute information of a candidate delivery information respectively to obtain a historical encoding vector and a content encoding vector;

[0212] Based on the difference between the historical delivery time of a historical delivery information and the estimated delivery time of a candidate delivery information, respectively encode the historical delivery time and the estimated delivery time to obtain a historical time sequence encoding vector and a content time sequence encoding vector;

[0213] Based on the historical coding vector and the historical time series coding vector, the historical behavior feature is obtained, and based on the content coding vector and the content time series coding vector, the candidate content feature is obtained.

[0214] Optionally, the acquiring unit 1201 is specifically configured to:

[0215] Add the historical coding vector and the historical time series coding vector to obtain the historical behavior characteristics;

[0216] The content encoding vector and the content timing encoding vector are added together to obtain the candidate content features.

[0217] Optionally, the prediction unit 1203 is specifically configured to:

[0218] Each cross feature is input into the estimation network of the target estimation model to obtain the output prediction feedback information.

[0219] Optionally, the prediction unit 1203 is specifically configured to:

[0220] Acquire an initial training sample set, wherein each initial training sample in the initial training sample set at least includes: a sample historical behavior feature of each of a plurality of sample historical delivery information of a sample object, and a sample content feature of each of the sample delivery information;

[0221] Based on the initial training sample set, the initial estimation model is iteratively trained to obtain the target estimation model.

[0222] Optionally, the prediction unit 1203 is specifically configured to:

[0223] During one training iteration, the following operations are performed:

[0224] Based on the feature extraction network of the initial estimation model, the content features of each sample in the selected initial training samples are cross-processed with the historical behavior features of each sample to obtain the sample cross-features of each sample delivery information;

[0225] Input the cross-features of each sample into the estimation network for information estimation, and obtain the sample feedback information of each sample delivery information;

[0226] The target loss function is constructed based on the feedback information of each sample, and the parameters of the initial estimation model are adjusted based on the target loss function.

[0227] Optionally, the predicted feedback information includes at least one of the following: click rate and conversion rate.

[0228] For the convenience of description, the above parts are divided into modules (or units) according to their functions and described separately. Of course, when implementing this application, the functions of each module (or unit) can be implemented in the same or multiple software or hardware.

[0229] Those skilled in the art will appreciate that various aspects of the present application may be implemented as a system, method or program product. Therefore, various aspects of the present application may be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to as "circuit", "module" or "system" herein.

[0230] Based on the same inventive concept as the above method embodiment, an electronic device is also provided in the embodiment of the present application. In one embodiment, the electronic device may be a server, such as Figure 1 In this embodiment, the structure of the electronic device can be as follows: Fig.13 As shown, it includes a memory 1301 , a communication module 1303 and one or more processors 1302 .

[0231] The memory 1301 is used to store computer programs executed by the processor 1302. The memory 1301 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and programs required for running the instant messaging function, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0232] The memory 1301 may be a volatile memory, such as a random-access memory (RAM); the memory 1301 may also be a non-volatile memory, such as a read-only memory, a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD); or the memory 1301 may be any other medium that can be used to carry or store a desired computer program in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 1301 may be a combination of the above memories.

[0233] The processor 1302 may include one or more central processing units (CPU) or a digital processing unit, etc. The processor 1302 is used to implement the above information delivery method when calling the computer program stored in the memory 1301 .

[0234] The communication module 1303 is used to communicate with terminal devices and other servers.

[0235] The specific connection medium between the memory 1301, the communication module 1303 and the processor 1302 is not limited in the embodiment of the present application. Fig.13 In the embodiment, the memory 1301 and the processor 1302 are connected via a bus 1304. The bus 1304 is connected to the processor 1302 via a bus 1304. Fig.13 The connections between the other components are only for illustration and are not intended to be limiting. The bus 1304 can be divided into an address bus, a data bus, a control bus, etc. For ease of description, Fig.13 The diagram shows that only one thick line is used, but this does not mean that there is only one bus or only one type of bus.

[0236] The memory 1301 stores a computer storage medium, and the computer storage medium stores computer executable instructions, and the computer executable instructions are used to implement the information delivery method of the embodiment of the present application. The processor 1302 is used to execute the above-mentioned information delivery method, such as Figure 2 shown.

[0237] In another embodiment, the electronic device may also be other electronic devices, such as Figure 1 In this embodiment, the structure of the electronic device can be as follows: Fig.14 As shown, it includes: a communication component 1410, a memory 1420, a display unit 1430, a camera 1440, a sensor 1450, an audio circuit 1460, a Bluetooth module 1470, a processor 1480 and other components.

[0238] The communication component 1410 is used to communicate with the server. In some embodiments, a circuit wireless fidelity (Wireless Fidelity, WiFi) module may be included. The WiFi module belongs to a short-range wireless transmission technology. The electronic device can help the user to send and receive information through the WiFi module.

[0239] The memory 1420 can be used to store software programs and data. The processor 1480 executes various functions and data processing of the terminal device by running the software programs or data stored in the memory 1420. The memory 1420 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. The memory 1420 stores an operating system that enables the terminal device to run. In the present application, the memory 1420 can store an operating system and various application programs, and can also store a computer program for executing the information delivery method of the embodiment of the present application.

[0240] The display unit 1430 can also be used to display information input by the user or information provided to the user and a graphical user interface (GUI) of various menus of the terminal device. Specifically, the display unit 1430 may include a display screen 1432 disposed on the front of the terminal device. The display screen 1432 may be configured in the form of a liquid crystal display, a light emitting diode, etc. The display unit 1430 may be used to display the information delivery user interface in the embodiment of the present application, etc.

[0241] The display unit 1430 can also be used to receive input digital or character information and generate signal input related to user settings and function control of the terminal device. Specifically, the display unit 1430 may include a touch screen 1431 arranged on the front of the terminal device, which can collect user touch operations on or near it, such as clicking a button, dragging a scroll box, etc.

[0242] The touch screen 1431 can be covered on the display screen 1432, or the touch screen 1431 and the display screen 1432 can be integrated to realize the input and output functions of the terminal device, and the integrated display screen can be referred to as a touch display screen. In this application, the display unit 1430 can display the application and the corresponding operation steps.

[0243] The camera 1440 can be used to capture still images, and users can post comments on the images taken by the camera 1440 through the application. The camera 1440 can be one or more. The object generates an optical image through the lens and projects it onto the photosensitive element. The photosensitive element can be a charge coupled device (CCD) or a complementary metal-oxide-semiconductor (CMOS) phototransistor. The photosensitive element converts the light signal into an electrical signal, and then passes the electrical signal to the processor 1480 to convert it into a digital image signal.

[0244] The terminal device may further include at least one sensor 1450, such as an acceleration sensor 1451, a distance sensor 1452, a fingerprint sensor 1453, and a temperature sensor 1454. The terminal device may also be configured with other sensors such as a gyroscope, a barometer, a hygrometer, a thermometer, an infrared sensor, a light sensor, and a motion sensor.

[0245] The audio circuit 1460, the speaker 1461, and the microphone 1462 can provide an audio interface between the user and the terminal device. The audio circuit 1460 can transmit the electrical signal converted from the received audio data to the speaker 1461, which is converted into a sound signal for output. The terminal device can also be equipped with a volume button for adjusting the volume of the sound signal. On the other hand, the microphone 1462 converts the collected sound signal into an electrical signal, which is received by the audio circuit 1460 and converted into audio data, and then the audio data is output to the communication component 1410 to be sent to, for example, another terminal device, or the audio data is output to the memory 1420 for further processing.

[0246] The Bluetooth module 1470 is used to exchange information with other Bluetooth devices having Bluetooth modules through the Bluetooth protocol. For example, the terminal device can establish a Bluetooth connection with a wearable electronic device (such as a smart watch) that also has a Bluetooth module through the Bluetooth module 1470 to exchange data.

[0247] The processor 1480 is the control center of the terminal device. It uses various interfaces and lines to connect various parts of the entire terminal. It executes various functions of the terminal device and processes data by running or executing software programs stored in the memory 1420 and calling data stored in the memory 1420. In some embodiments, the processor 1480 may include one or more processing units; the processor 1480 may also integrate an application processor and a baseband processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the baseband processor mainly processes wireless communications. It is understandable that the above-mentioned baseband processor may not be integrated into the processor 1480. In the present application, the processor 1480 can run the operating system, application programs, user interface display and touch response, as well as the information delivery method of the embodiment of the present application. In addition, the processor 1480 is coupled to the display unit 1430.

[0248] In some possible implementations, various aspects of the information delivery method provided by the present application may also be implemented in the form of a program product, which includes a computer program. When the program product is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of the information delivery method according to various exemplary embodiments of the present application described above in this specification. For example, the electronic device may execute the following steps: Figure 2 Follow the steps shown in .

[0249] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The 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 readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable 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.

[0250] The program product of the embodiment of the present application may adopt a portable compact disk read-only memory (CD-ROM) and include a computer program, and can be run on an electronic device. However, the program product of the present application is not limited thereto, and in this document, a readable storage medium may be any tangible medium containing or storing a program, which can be used by or in combination with a command execution system, apparatus, or device.

[0251] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, wherein a readable computer program is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with a command execution system, apparatus, or device.

[0252] The computer program embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0253] Computer programs for performing the operations of the present application may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The computer program may be executed entirely on the user electronic device, partially on the user device, as a separate software package, partially on the user electronic device and partially on a remote electronic device, or entirely on a remote electronic device or server. In the case of a remote electronic device, the remote electronic device may be connected to the user electronic device via any type of network including a local area network (LAN) or a wide area network (WAN), or may be connected to an external electronic device (e.g., via the Internet using an Internet service provider).

[0254] It should be noted that, although several units or subunits of the device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided into multiple units to be embodied.

[0255] In addition, although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0256] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain a computer-usable computer program.

[0257] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program commands. These computer program commands can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the commands executed by the processor of the computer or other programmable data processing device generate commands for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0258] These computer program commands may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the commands stored in the computer readable memory produce an article of manufacture comprising a command device, the command device implementing the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0259] These computer program commands can also be loaded onto a computer or other programmable data processing device so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the commands executed on the computer or other programmable device provide the instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0260] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0261] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for delivering information, characterized in that: The method comprises: In response to an information delivery request for a target object, respectively obtain historical behavior features of each of a plurality of historical delivery information for the target object, and candidate content features of each of the candidate delivery information; wherein the historical behavior features at least include: attribute information of the corresponding historical delivery information, historical delivery time, and historical feedback behavior of the target object to the historical delivery information; the candidate content features at least include: attribute information of the corresponding candidate delivery information and an estimated delivery time; Perform at least two types of feature cross processing on each candidate content feature and each historical behavior feature to obtain information cross features of each candidate delivery information; Based on the cross-features of each information, obtaining the predicted feedback information of each candidate delivery information by the target object; Based on the obtained prediction feedback information, target delivery information that meets the preset delivery conditions is screened out from the candidate delivery information, and the target delivery information is delivered to the target object.

2. The method according to claim 1, characterized in that The step of performing at least two types of feature cross processing on each candidate content feature and each historical behavior feature to obtain information cross features of each candidate delivery information includes: For each candidate content feature, the following operations are performed respectively: Perform two types of feature cross processing on a candidate content feature and each of the historical behavior features to obtain each time series attention weight and each time series content feature; Multiplying each temporal attention weight by the corresponding temporal content feature to obtain each information interaction feature; wherein the temporal attention weight and the corresponding temporal content feature correspond to the same historical behavior feature; The information interaction features are concatenated to obtain corresponding information cross features.

3. The method according to claim 2, characterized in that The method of performing two types of feature cross processing on a candidate content feature and each of the historical behavior features to obtain each time series attention weight and each time series content feature includes: Perform dot product of the candidate content feature and each of the historical behavior features to obtain each of the temporal attention weights; The one candidate content feature is multiplied by each of the historical behavior features to obtain each of the temporal content features.

4. The method according to claim 1, characterized in that The historical behavior features and the candidate content features are obtained in the following manner: Encode the attribute information and the historical feedback behavior of a historical delivery information and the attribute information of a candidate delivery information respectively to obtain a historical encoding vector and a content encoding vector; Based on the difference between the historical delivery time of the one historical delivery information and the expected delivery time of the one candidate delivery information, respectively encode the historical delivery time and the expected delivery time to obtain a historical time sequence encoding vector and a content time sequence encoding vector; The historical behavior feature is obtained based on the historical coding vector and the historical time sequence coding vector, and the candidate content feature is obtained based on the content coding vector and the content time sequence coding vector.

5. The method according to claim 4, characterized in that The obtaining of the historical behavior feature based on the historical coding vector and the historical time sequence coding vector, and the obtaining of the candidate content feature based on the content coding vector and the content time sequence coding vector, comprises: Adding the historical coding vector and the historical time series coding vector to obtain the historical behavior feature; The content coding vector and the content timing coding vector are added together to obtain the candidate content feature.

6. The method according to any one of claims 1 to 5, characterized in that: The obtaining, based on the cross-features of the information, the predicted feedback information of the target object for each of the candidate delivery information includes: The information cross features are respectively input into the estimation network of the target estimation model to obtain output prediction feedback information.

7. The method according to claim 6, characterized in that The target estimation model is trained by the following method: Acquire an initial training sample set, wherein each initial training sample in the initial training sample set at least includes: a sample historical behavior feature of each of a plurality of sample historical delivery information of a sample object, and a sample content feature of each of the sample delivery information; Based on the initial training sample set, the initial estimation model is iteratively trained to obtain the target estimation model.

8. The method according to claim 7, characterized in that During one training iteration, the following operations are performed: Based on the feature extraction network of the initial estimation model, the content features of each sample in the selected initial training samples are subjected to feature cross processing with the historical behavior features of each sample, so as to obtain the sample cross features of each sample delivery information; Inputting the cross features of each sample into the estimation network to perform information estimation, and obtaining sample feedback information of each sample delivery information; A target loss function is constructed based on feedback information of each sample, and parameters of the initial estimation model are adjusted based on the target loss function.

9. The method according to any one of claims 1 to 5, characterized in that: The predicted feedback information includes at least one of the following: click rate and conversion rate.

10. An information delivery device, characterized in that: include: an acquisition unit, configured to respectively acquire, in response to an information delivery request for a target object, a historical behavior feature of each of a plurality of historical delivery information for the target object, and a candidate content feature of each of the candidate delivery information; wherein the historical behavior feature at least includes: attribute information of the corresponding historical delivery information, historical delivery time, and historical feedback behavior of the target object to the historical delivery information; and the candidate content feature at least includes: attribute information of the corresponding candidate delivery information and an estimated delivery time; A cross unit, used to perform at least two types of feature cross processing on each candidate content feature and each historical behavior feature, respectively, to obtain information cross features of each candidate delivery information; A prediction unit, configured to obtain prediction feedback information of the target object for each of the candidate delivery information based on cross-features of each information; The screening unit is used to screen out target delivery information that meets the preset delivery conditions from the candidate delivery information based on the obtained prediction feedback information, and deliver the target delivery information to the target object.

11. An electronic device, characterized in that: It comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any one of the methods of claims 1 to 9.

12. A computer-readable storage medium, characterized in that: It comprises a computer program. When the computer program is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of any method described in claims 1 to 9.

13. A computer program product, characterized in that The method comprises a computer program stored in a computer-readable storage medium; when a processor of an electronic device reads the computer program from the computer-readable storage medium, the processor executes the computer program, so that the electronic device executes the steps of any one of the methods described in claims 1 to 9.

Images