Delivery processing method, related device and medium
Through multi-task modeling and expert network prediction technology, the problem of low matching in delivery processing is solved, high-precision matching between the delivery content and the object is achieved, and the delivery effect is improved.
Patent Information
- Application Number
- CN202310591204.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-23
- Publication Date
- 2024-11-26
AI Technical Summary
In the delivery process, the matching degree between the delivery content and the object is not high, resulting in low delivery accuracy and the inability to effectively predict the multi-task completion status of the object's treatment of the delivery content.
Using multi-task modeling method, by obtaining multiple delivery basic features, vectorized and inputting them into the corresponding expert network, using gated nodes to generate task association features, and predicting the task completion probability through multi-layer network models to improve delivery content Match degree with the object.
It improves the accuracy and matching of delivery processing, ensures that the delivered content more accurately matches the target object's interests, and improves the relevance and user experience of the delivered content.
Smart Images

Figure CN119025740A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of big data, and in particular to a delivery processing method, related devices and media. Background Art
[0002] At present, in the process of delivering content to be delivered (such as published videos, works, etc.) to an object, the object features and the features of the content to be delivered are generally quantized and input into various expert networks together, and different implicit task representations are extracted using the expert network. The output results of each expert network are weighted through the gated nodes to effectively balance the sharing and mutual exclusion between multiple tasks (such as likes, comments, etc.), and the multi-task modeling method is used to predict the object's multi-task completion of the delivered content (whether it will like, comment, etc.), and based on this, find the object that may need it for the content to be delivered. However, the delivery processing method is not very accurate, and the objects found for the content to be delivered are often not interested in the content to be delivered. Summary of the invention
[0003] The embodiments of the present disclosure provide a delivery processing method, related devices and media, which can improve the accuracy of delivery processing and enhance the matching degree between the content to be delivered and the object to which it is delivered.
[0004] According to one aspect of the present disclosure, a delivery processing method is provided, including:
[0005] Acquire a plurality of delivery basic features, where the delivery basic features are used to predict task completion probabilities of a target object with respect to a plurality of tasks for the delivered content;
[0006] Quantizing the plurality of delivery basic feature vectors to obtain a plurality of first feature vectors;
[0007] Inputting each of the first feature vectors into a first expert network corresponding to the first feature vector, wherein the first expert network includes a plurality of first dedicated models corresponding to a plurality of the tasks and a plurality of first gating nodes, and the first gating nodes generate task-associated features corresponding to the tasks and the first feature vectors based on outputs of the first dedicated models;
[0008] generating a second feature vector of the task based on a plurality of the task-associated features corresponding to the task and a plurality of the first feature vectors, and inputting the second feature vector into a second expert network corresponding to the task to obtain a third feature vector corresponding to the task;
[0009] The third feature vectors corresponding to the multiple tasks are input into the prediction models corresponding to the multiple tasks to obtain the predicted task completion probabilities of the target object for the to-be-delivered content with respect to the multiple tasks.
[0010] According to one aspect of the present disclosure, there is provided a delivery processing device, comprising:
[0011] A first acquisition unit is used to acquire a plurality of delivery basic features, wherein the delivery basic features are used to predict task completion probabilities of a target object with respect to a plurality of tasks for the delivered content;
[0012] A vectorization unit, used for vectorizing the plurality of delivery basic features to obtain a plurality of first feature vectors;
[0013] a first input unit, configured to input each of the first feature vectors into a first expert network corresponding to the first feature vector, wherein the first expert network includes a plurality of first dedicated models corresponding to a plurality of the tasks, and a plurality of first gating nodes, and the first gating nodes generate task-associated features corresponding to the tasks and the first feature vectors based on outputs of the first dedicated models;
[0014] A first generating unit is used to generate a second feature vector of the task based on the task and a plurality of the task-associated features corresponding to the plurality of the first feature vectors, and input the second feature vector into a second expert network corresponding to the task to obtain a third feature vector corresponding to the task;
[0015] The second input unit is used to input the third feature vectors corresponding to the multiple tasks into the prediction model corresponding to the multiple tasks to obtain the predicted task completion probability of the target object for the content to be delivered with respect to the multiple tasks.
[0016] Optionally, the second expert network includes a second dedicated model corresponding to the task, and a second gating node corresponding to the task;
[0017] The first generating unit is specifically used for:
[0018] Inputting the second feature vector into a second dedicated model corresponding to the task to obtain an output of the second expert model;
[0019] The third feature vector corresponding to the task is generated by the second gating node based on the output of the second dedicated model.
[0020] Optionally, the first expert network further includes a first sharing model, and the second expert network further includes a second sharing model;
[0021] The first input unit is specifically configured to: if the first gating node corresponds to one of the first dedicated models, the first gating node performs a weighted sum of the output of the first shared model and the output of the first dedicated model corresponding to the first gating node to obtain the task-associated feature corresponding to the task and the first feature vector; if the first gating node corresponds to the first shared model, the first gating node performs a weighted sum of the output of the first shared model and the outputs of each of the first dedicated models to obtain a first shared feature corresponding to the first feature vector;
[0022] The first generation unit is specifically used to: generate a first shared feature vector based on multiple first shared features corresponding to multiple first feature vectors, and input it into the second shared model; through the second gating node, perform a weighted sum of the output of the second shared model and the output of the second dedicated model corresponding to the second gating node to obtain the third feature vector corresponding to the task.
[0023] Optionally, the first input unit is specifically used for:
[0024] Based on the first feature vector, obtaining first weights of the first shared model and each of the first dedicated models;
[0025] Using the first shared model and the first weight of the first dedicated model corresponding to the first gating node, a weighted sum is performed on an output of the first shared model and an output of the first dedicated model corresponding to the first gating node to obtain the task-associated feature corresponding to the task and the first feature vector;
[0026] The first input unit is specifically used to: use the first shared model and the first weights of each of the first dedicated models to perform a weighted sum on the output of the first shared model and the output of each of the first dedicated models to obtain a first shared feature corresponding to the first feature vector.
[0027] Optionally, the first input unit is specifically used for:
[0028] Obtaining the first shared model and each first dedicated model's first matrix, wherein the number of columns of the first matrix is the total number of models of the first dedicated model and the first shared model, and the number of rows of the first matrix is the dimension of the first eigenvector;
[0029] performing a first multiplication operation on the first matrix and the transpose of the first eigenvector;
[0030] A normalized exponent process is performed on a result of the first multiplication operation to obtain the first weight.
[0031] Optionally, the first generating unit is specifically configured to:
[0032] Cascading a plurality of the first feature vectors to obtain a first cascade feature vector;
[0033] Based on the first cascade feature vector, obtaining second weights of the second shared model and the second dedicated model corresponding to the second gating node;
[0034] Using the second weights of the second shared model and the second dedicated model corresponding to the second gating node, the output of the second shared model and the output of the second dedicated model corresponding to the second gating node are weighted summed to obtain the third feature vector corresponding to the task.
[0035] Optionally, the first generating unit is specifically configured to:
[0036] Obtaining the second shared model and the second matrix of each of the second dedicated models, wherein the number of columns of the second matrix is the total number of models of the second dedicated model and the second shared model, and the number of rows of the second matrix is the dimension of the first cascaded feature vector;
[0037] performing a second multiplication operation on the second matrix and the transpose of the first concatenated eigenvector;
[0038] A normalized exponent process is performed on the operation of the second multiplication operation to obtain the second weight.
[0039] Optionally, the first acquiring unit is specifically configured to:
[0040] Acquire object features of the target object;
[0041] Acquire scene features of the current scene where the target object is located;
[0042] Filtering out a filtered content set from the to-be-filtered content set, and taking the features of each filtered content in the filtered content set as features of the to-be-delivered content in turn;
[0043] The object feature, the scene feature, and the feature of the content to be delivered are used as a plurality of delivery basic features.
[0044] Optionally, the first acquiring unit is specifically configured to:
[0045] Obtaining a first label of the target object;
[0046] Obtain a second tag of the content to be filtered in the set of content to be filtered;
[0047] Based on the matching degree between the first tag and the second tag, a filtered content set is filtered out from the to-be-filtered content set.
[0048] Optionally, the first acquiring unit is specifically configured to:
[0049] Based on the matching degree, selecting matching contents in which the first tag and the second tag match from the to-be-screened content set;
[0050] Obtaining multiple delivery indicators for each of the matching contents;
[0051] Based on the multiple delivery indicators, filtered contents are filtered out from the matching contents to form the filtered content set.
[0052] Optionally, the vectorization unit is specifically used for:
[0053] If the delivery basic feature is a character feature, input the character feature into an embedding layer to obtain the first feature vector corresponding to the character feature;
[0054] If the delivery basic feature is a numerical feature, numerical mapping is performed on the numerical feature to obtain the first feature vector corresponding to the numerical feature;
[0055] If the delivery basic feature is a variable-length sequence feature, the variable-length sequence feature is pooled to obtain the first feature vector corresponding to the variable-length sequence feature.
[0056] Optionally, the delivery processing device further includes:
[0057] A second acquisition unit, used to acquire the focus task of the target object;
[0058] A third acquisition unit, configured to acquire a task completion probability for the task of interest from the task completion probabilities for the plurality of tasks;
[0059] A sorting unit, configured to sort the to-be-delivered content based on the task completion probability of the concerned task;
[0060] A delivery unit is used to deliver the content to be delivered to the target object according to the ranking.
[0061] Optionally, the sorting unit is specifically used to:
[0062] If there are multiple tasks of interest, determining an overall task completion probability based on the task completion probabilities of the multiple tasks of interest;
[0063] The to-be-delivered contents are sorted based on the overall task completion probability.
[0064] Optionally, the plurality of delivery basic features are selected from a training sample in a training sample set, the training sample having a plurality of training labels for the plurality of tasks, the training labels indicating whether the task is completed for the training sample;
[0065] The first expert network, the second expert network, and the prediction model are jointly trained through the following process:
[0066] For a single task, determining an error subfunction of the task based on the training labels and the task completion probability of the task;
[0067] Based on the error sub-function, determining the error function of the training sample;
[0068] Based on the error function, the first expert network, the second expert network, and the prediction model are jointly trained.
[0069] Optionally, inputting each of the first feature vectors into a first expert network corresponding to the first feature vector comprises:
[0070] Obtaining task weights of a plurality of the tasks;
[0071] Based on the task weights, a weighted sum is performed on the error sub-functions of a plurality of the tasks to obtain the error function.
[0072] Optionally, the first input unit is specifically used for:
[0073] Inputting each of the first feature vectors into a plurality of the first expert networks corresponding to the first feature vector, to obtain a plurality of candidate task-associated features corresponding to the task and the first feature vector;
[0074] Based on a predetermined criterion, a task-associated feature corresponding to the task and the first feature vector is determined according to a plurality of candidate task-associated features.
[0075] According to one aspect of the present disclosure, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-mentioned delivery processing method when executing the computer program.
[0076] According to one aspect of the present disclosure, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the delivery processing method as described above is implemented.
[0077] According to one aspect of the present disclosure, a computer program product is provided. The computer program product includes a computer program. The computer program is read and executed by a processor of a computer device, so that the computer device executes the delivery processing method as described above.
[0078] In the disclosed embodiment, a multi-task modeling approach is adopted not only in the overall aspect of determining the task completion probability of the object to be delivered content based on the delivery basic features, but also in the aspect of obtaining input features. It allows each delivery basic feature to correspond to a first expert network, and each first expert network contains multiple first special models and multiple first gating nodes corresponding to multiple tasks. In this way, the output of each first gating node corresponds to both a single task and a single first feature vector converted from a single delivery basic feature. The outputs of multiple first gating nodes corresponding to the same task are input to the second expert network corresponding to the task, so that the input features of the second expert network are task-specific input features, and are no longer shared input feature vectors, thereby fully considering the role of multi-task modeling in feature construction, and improving the overall delivery accuracy of the constructed delivery model, the accuracy of delivery processing, and the matching degree between the content to be delivered and the object to be delivered.
[0079] Other features and advantages of the present disclosure will be described in the following description, and partly become apparent from the description, or understood by practicing the present disclosure. The purpose and other advantages of the present disclosure can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0080] The accompanying drawings are used to provide further understanding of the technical solution of the present disclosure and constitute a part of the specification. Together with the embodiments of the present disclosure, they are used to explain the technical solution of the present disclosure and do not constitute a limitation on the technical solution of the present disclosure.
[0081] Figure 1 is a system architecture diagram of the delivery processing method of the embodiment of the present disclosure;
[0082] Figure 2A -B is a schematic diagram of an interface of an embodiment of the present disclosure applied to a video delivery scenario;
[0083] Figure 3 is a flow chart of a delivery processing method according to an embodiment of the present disclosure;
[0084] Figure 4A -B is two overall implementation structure diagrams of the delivery processing method of the embodiment of the present disclosure;
[0085] Figure 5 yes Figure 3A flow chart of obtaining the basic features of the delivery in step 310;
[0086] Figure 6 yes Figure 5 Get a schematic diagram of the basic features of the delivery;
[0087] Figure 7 yes Figure 5 A flowchart of obtaining the filtered content set in step 530;
[0088] Figure 8 yes Figure 7 Obtain a schematic diagram of the filtered content set;
[0089] Fig. 9 yes Figure 7 A flowchart of obtaining the filtered content set in step 730;
[0090] Fig.10 yes Fig. 9 Obtain a schematic diagram of the filtered content set;
[0091] Fig.11A -C is a schematic diagram of generating the first feature vector for different types of delivery basic features;
[0092] Fig.12 yes Figure 3 A flow chart of the first gating node generating task association features in step 330;
[0093] Fig.13 is Fig.12 A flow chart based on
[0094] Fig.14 yes Fig.13 A flowchart of obtaining a first weight in step 1211;
[0095] Fig.15 yes Figure 3 A flow chart of generating a task association vector in step 330;
[0096] Fig.16 yes Fig.15 A schematic diagram of obtaining a task association vector;
[0097] Fig.17A -B is a schematic diagram of two methods of generating the second feature vector based on the task association vector;
[0098] Fig.18 yes Figure 3 A flow chart of generating a third eigenvector in step 340;
[0099] Fig.19 exist Fig.18A flow chart of further generating a third eigenvector based on step 1820;
[0100] Fig. 20 is Fig.19 A flow chart of further generating a third eigenvector based on step 1920;
[0101] Fig.21 yes Fig. 20 A flowchart of obtaining the second weight in step 2020;
[0102] Fig. 22 is Figure 3 Then add a flow chart of the process of determining the order of delivery of the content to be delivered;
[0103] Fig.23 yes Fig. 22 A schematic diagram for determining the order in which the content to be delivered is delivered;
[0104] Fig.24 yes Fig. 22 A flowchart of sorting the content to be delivered in step 2230;
[0105] Fig.25 is a flow chart for jointly training the first expert network, the second expert network, and the prediction model;
[0106] Fig.26 It is an implementation detail diagram of the delivery processing method of the embodiment of the present disclosure.
[0107] Fig. 27 is a module diagram of a delivery processing device according to an embodiment of the present disclosure;
[0108] Fig.28 According to the embodiment of the present disclosure Figure 3 The terminal structure diagram of the delivery processing method shown;
[0109] Fig.29 According to the embodiment of the present disclosure Figure 3 The server structure diagram of the delivery processing method shown. DETAILED DESCRIPTION
[0110] In order to make the purpose, technical solution and advantages of the present disclosure more clear, the present disclosure is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present disclosure and are not used to limit the present disclosure.
[0111] Before further describing the embodiments of the present disclosure in detail, the nouns and terms involved in the embodiments of the present disclosure are described. The nouns and terms involved in the embodiments of the present disclosure are subject to the following interpretations:
[0112] Embedding layer: A word encoding technology specifically used in the field of natural language processing, the purpose is to reduce the waste of resources. It uses low-dimensional vectors to represent the encoding, which can express the relationship between words through neural network optimization.
[0113] Deep neural network: It is a multi-layer unsupervised neural network. Deep neural network uses the output features of the previous layer as the input of the next layer for feature learning. After layer-by-layer feature mapping, the features of the existing spatial samples are mapped to another feature space, so as to learn to have better feature expression for the existing input. Deep neural network has multiple nonlinear mapping feature transformations and can fit highly complex functions.
[0114] Gradient descent algorithm: A method for minimizing the loss function. It calculates the gradient of the loss function at the current position and updates the parameters in the direction of gradient descent to find the optimal solution that minimizes the value of the loss function.
[0115] System architecture and scenario description of the application of the embodiments of the present disclosure
[0116] Figure 1 1 is a system architecture diagram used in the delivery processing method according to an embodiment of the present disclosure, and includes a target terminal 110, the Internet 120, and a server 130, etc.
[0117] The object terminal 110 is a device used to deliver content so that the object can view the delivered content. It includes various forms such as desktop computers, laptops, PDAs (personal digital assistants), mobile phones, car terminals, home theater terminals, and dedicated terminals. In addition, it can be a single device or a collection of multiple devices. For example, multiple devices are connected through a local area network and work together using a common display device to form a terminal. The terminal can also communicate with the Internet 120 in a wired or wireless manner to exchange data.
[0118] The server 130 refers to a computer system that can provide content delivery services to the target terminal 110. Compared with the target terminal 110, the server 130 has higher requirements in terms of stability, security, performance, etc. The server 130 can be a high-performance computer in a network platform, a cluster of multiple high-performance computers, a part of a high-performance computer (such as a virtual machine), a combination of parts of multiple high-performance computers (such as virtual machines), etc. The server 130 can also communicate with the Internet 120 in a wired or wireless manner to exchange data.
[0119] The embodiments of the present disclosure can be applied in various scenarios, such as Figure 2A -B shows a scenario of viewing the delivered video content in an instant short message application, etc.
[0120] like Figure 2A The interface of the instant short message application in the object terminal 110 is shown in FIG. The instant short message application can be used to communicate with contacts instantly. After the "Short Message" option in the function bar below is triggered, a list of short messages with multiple contacts is displayed. The instant short message application can also be used to watch videos. After the object triggers the "Video" option in the function bar, the interface of the instant short message application is as follows: Figure 2B shown.
[0121] exist Figure 2B In the instant short message application, the video interface is opened and the video content is played. The interface also displays the number of likes, favorites, and comments on the current video content before the current time. The subject can also like, favorite, and comment on the current video content in the interface. If the subject wants to watch other videos, the next video can be switched through specific operations (such as sliding the screen up). The video content delivered to the instant short message application for the subject to view can be determined by the server 130, that is, the server 130 predicts the video content that the subject may be interested in and delivers the video content to the instant short message application. Therefore, it is very important to correctly judge the subject's preferences and accurately deliver video content to the subject according to the preferences. The disclosed embodiment can improve the matching degree between the delivered video content and the subject's preferences.
[0122] General description of the disclosed embodiments
[0123] According to an embodiment of the present disclosure, a delivery processing method is provided.
[0124] The delivery processing method is the process of delivering content that the target object is interested in to the object terminal 110 of the target object. The target object is the object that wants to view the content to be delivered. The embodiment of the present disclosure can be applied to the delivery of different types of content, and the content to be delivered can be pictures, videos, articles, applications and other types. If the target object opens the video recommendation software, the delivery processing method of the embodiment of the present disclosure is used to predict the video content that the target object may be interested in, and deliver the video content to the object terminal 110 of the target object through the video recommendation software. When determining the delivery content, the completion status of the target object for different tasks to be delivered must also be considered. For example, whether the target object will like the content to be delivered, whether the target object will collect the content to be delivered, etc. The delivery order of the content to be delivered is determined based on the completion status of the target object for the content to be delivered.
[0125] The delivery processing method of the embodiment of the present disclosure is executed on the server 130. After the execution is completed, the obtained delivery content is transmitted to the object terminal 110 through the Internet 120, and the object terminal 110 displays the delivery content to the target object.
[0126] At present, in one method, the basic features of delivery are quantized and input into the expert network corresponding to the task, the output results of each expert network are weighted through the gated node, and then the multi-task completion of the object on the content to be delivered is predicted through multi-task modeling, and the content to be delivered to the target object is determined accordingly. The content to be delivered determined by this method may not be the content that the target object is interested in, and the accuracy of the delivery processing result is low. The embodiment of the present disclosure provides a delivery processing method that also performs multi-task modeling on the input features to improve the overall delivery accuracy, the precision of the delivery processing, and the matching degree between the content to be delivered and the delivered object.
[0127] like Figure 3 As shown, according to one embodiment of the present disclosure, the delivery processing method includes:
[0128] Step 310: Obtain multiple basic delivery features;
[0129] Step 320, quantizing the multiple delivery basic feature vectors to obtain multiple first feature vectors;
[0130] Step 330: input each first feature vector into a first expert network corresponding to the first feature vector, wherein the first expert network includes a plurality of first dedicated models corresponding to a plurality of tasks and a plurality of first gating nodes, wherein the first gating nodes generate task-associated features corresponding to the task and the first feature vector based on outputs of the first dedicated models;
[0131] Step 340: Generate a second feature vector for the task based on the task and the plurality of task-associated features corresponding to the plurality of first feature vectors, and input the second feature vector into a second expert network corresponding to the task to obtain a third feature vector corresponding to the task;
[0132] Step 350: Input the third feature vectors corresponding to the multiple tasks into the prediction models corresponding to the multiple tasks to obtain the predicted task completion probabilities of the target object for the content to be delivered with respect to the multiple tasks.
[0133] The above steps 310 - 350 are described in general below.
[0134] Figure 4A What is shown is the overall structural diagram of the delivery processing method of the embodiment of the present disclosure.
[0135] Figure 4AThe bottom layer is the basic delivery features in step 310. There are multiple basic delivery features, which are features used to predict the completion of different tasks performed by the target object on the content to be delivered. Tasks refer to matters related to the content to be delivered, which are used to weigh the target object's interest in the content to be delivered from multiple aspects. For example, a task is that the target object likes the content to be delivered, or the target object collects the content to be delivered, etc.
[0136] Step 320 is executed on the delivery basic feature to perform feature vectorization to obtain a first feature vector (not shown in the figure). Each delivery basic feature corresponds to a first feature vector.
[0137] Step 330 is performed on each first feature vector and inputted into the corresponding first expert network. Each first feature vector corresponds to a first expert network. The first expert network includes multiple first special models and multiple first gating nodes corresponding to multiple tasks. That is to say, in each expert network, a first special model and a first gating node corresponding to task A, and a first special model and a first gating node corresponding to task B are included. The first special models corresponding to different tasks are used to perform multi-task modeling on the features, and the first gating nodes corresponding to different tasks are used to perform high-order processing on the output of the first special model corresponding to the task, so as to obtain different feature vectors obtained by the first feature vector corresponding to different tasks. The first feature vector is inputted into different first special models corresponding to task A and task B in the corresponding first expert network. The output result of the first special model corresponding to task A is inputted into the first gating node corresponding to task A, and the first gating node outputs the task-associated features corresponding to task A and the first feature vector. The output result of the first special model corresponding to task B is inputted into the first gating node corresponding to task B, and the first gating node outputs the task-associated features corresponding to task B and the first feature vector (not shown in the figure).
[0138] Each first expert network will obtain different feature vectors generated by the first feature vector corresponding to different tasks. Execute step 340, perform feature cascade on the task-related features generated by the first expert networks corresponding to different first feature vectors for the same task, and generate a second feature vector corresponding to the task (not shown in the figure). Feature cascade is used to combine task-related vectors from different first expert networks into one vector. The second feature vector is input into the second expert network, and the second expert network includes multiple second special models and multiple second gating nodes corresponding to multiple tasks. That is to say, in the second expert network, a second special model and a second gating node corresponding to task A are included, and a second special model and a second gating node corresponding to task B are also included. The second special model is used to model the input features specific to the task, and the second gating node is used to perform high-order processing on the output of the second special model corresponding to the task to obtain a feature vector that can be used to predict the task result. For example, the task-related features corresponding to task A output by different first expert networks corresponding to different first feature vectors are cascaded to generate a second feature vector corresponding to task A. The second feature vector corresponding to task A is input into the second special model corresponding to task A in the second expert network, and then the third feature vector of task A is generated according to the output of the second special model through the second gating node corresponding to task A. The third feature vector corresponding to task B is obtained by the task-related features corresponding to task B output by different first expert networks corresponding to different first feature vectors, which is the same as the process of generating the third feature vector corresponding to task A, and will not be repeated here.
[0139] After obtaining the third feature vector corresponding to the task, execute step 350 to input the third feature vector into the prediction model corresponding to the task. The prediction model can obtain the task completion probability of the target object performing the corresponding task on the content to be delivered based on the third feature vector. For example, the third feature vector corresponding to task A is input into the prediction model corresponding to task A, and the prediction model inputs the task completion probability of the target object performing task A on the content to be delivered. The process of obtaining the task completion probability of the target object performing task B on the content to be delivered is the same as the above-mentioned process of obtaining the task completion probability of performing task A, and will not be repeated here.
[0140] Figure 4B It is an overall structural diagram of a delivery processing method according to another embodiment of the present disclosure.
[0141] Figure 4B and Figure 4A The difference is that in Figure 4BIn the figure, the first expert network also includes a first shared model, and the first shared model corresponds to a first gating node; the second expert network also includes a second shared model. The first shared model is used to model the first feature vector according to the connection between different tasks. In addition to being input into the first special models corresponding to different tasks, the first feature vector is also input into the first shared model. The first gating node corresponding to the task will perform high-order processing on the output of the first special model corresponding to the task and the output of the first shared model to obtain a task association vector corresponding to the first feature vector and the task. The first gating node corresponding to the first shared model will obtain the first shared feature corresponding to the first shared model based on the output of the first shared model and the output of all other first special models (not shown in the figure).
[0142] The first shared features corresponding to the first shared models in all the first expert networks are cascaded to obtain a first shared feature vector (not shown in the figure). The first shared feature vector is input into the second shared model of the second expert network. The second shared model is used to model the first shared feature vector. The second gating node corresponding to the task performs high-order processing on the output of the second dedicated model corresponding to the task and the output of the second shared model to obtain a third feature vector corresponding to the task. It should be noted that in Figure 4B In the example, the second shared model is incorporated into the second expert network for task A, while the second expert network for task B is not shown. In fact, the second shared model is included in the second expert network for each task. For example, in the second expert network for task A, the second gating node generates the third feature vector for task A based on the output of the second shared model and the output of the second dedicated model for task A; in the second expert network for task B, the second gating node generates the third feature vector for task B based on the output of the second shared model and the output of the second dedicated model for task B.
[0143] Figure 4B The embodiments shown will be described in detail in the following content, and only a brief description is given here. Figure 4A The differences between the embodiments shown.
[0144] The above steps 310 - 350 are described in detail below.
[0145] Detailed description of step 310
[0146] In step 310, the delivery basic features are used to predict the target object's task completion probability for multiple tasks of the delivered content. Which content to be delivered is used as the delivered content is determined based on the task completion probability. Therefore, the delivery basic features include multiple aspects that can affect the target object's preference for the delivered content.
[0147] In one embodiment, Figure 5 As shown, multiple basic delivery features are obtained, including:
[0148] Step 510: Obtain object features of the target object;
[0149] Step 520: Obtain scene features of the current scene where the target object is located;
[0150] Step 530: Filter out a filtered content set from the to-be-filtered content set, and use the features of each filtered content in the filtered content set in turn as features of the to-be-delivered content;
[0151] Step 540: Use the object features, scene features, and features of the content to be delivered as multiple delivery basic features.
[0152] In step 510, the object feature refers to a feature related to the target object. The object feature can be a feature of the target object itself, such as the industry, education level, etc. of the target object. For the same feature, if the feature value is different, the target object may have different preferences for the delivered content. For example, the target objects in different industries may prefer different delivery content. Taking video content as an example, the target in the news industry may pay more attention to videos related to hot events, and the target in the catering industry may like videos related to health and wellness.
[0153] Note that when obtaining the target object's object characteristics, the target object's consent must be obtained in advance. In addition, the collection, use and processing of these object characteristics will comply with relevant laws, regulations and standards. When obtaining the target object's consent, the target object's separate permission or consent can be obtained through pop-ups or by jumping to a confirmation page.
[0154] The object characteristics may also be the object characteristics of the object group (such as a family or work group) to which the target object belongs. This is because, in some cases, the object's preference for delivered content is greatly affected by the object group to which it belongs. For example, the object itself likes to watch music-related video content, but because other members of the object's family group like to watch game-related video content, the object may also be interested in game-related video content. Therefore, in one embodiment, obtaining the object characteristics of the target object also includes obtaining the object characteristics of the object group to which the target object belongs. Considering the object characteristics of the object group to which the target object belongs can improve the accuracy of content delivery. Similar to the above, when obtaining the object characteristics of the object group (such as a family or work group) to which the target object belongs, the consent of the object group members must also be sought in advance, so it will not be repeated.
[0155] In one embodiment, the object characteristics of the target object can be obtained through registration information. The registration information here can be the registration information of the target object on the object terminal 110, or the registration information in the software for delivering the delivery content, etc. When the target object uses the object terminal 110 for the first time, the object terminal 110 may require the target object to register, and registration requires filling in information such as industry as registration information. In this way, the corresponding object characteristics can be obtained from the object terminal 110 used by the target object. When the target object uses the software for delivering the content to be delivered for the first time, it is also necessary to register, and the object may be required to fill in information such as industry as registration information. Therefore, the object characteristics can also be obtained through the registration information in the software. The advantages of obtaining the object characteristics through registration information are: convenient and fast, and high acquisition efficiency. However, when obtaining the registration information of the target object, as mentioned above, the consent of the target object must also be sought in advance, and in order to save space, it is not repeated.
[0156] In another embodiment, the object characteristics of the target object can be obtained through the historical viewing log of the target object within a predetermined time period before the current time point. The historical viewing log refers to the relevant records of the target object viewing the delivery content within the predetermined time period before the current time point. The predetermined time period can be one hour, one day, one week, one month, etc. The predetermined time period can be determined according to the frequency of the target object viewing the delivery content. If the target object views the delivery content frequently, the current object characteristics of the target object can be obtained through the viewing records within a shorter time period; if the target object views the delivery content less frequently, it is necessary to use the viewing records within a longer time period to ensure that there is enough data to obtain the object characteristics. The predetermined time period can also be determined according to the type of delivery content. For example, if the delivery content is video content, a longer time period can be used as the predetermined time period to obtain the long-term preference of the target object; if the delivery content is news content, since news is time-sensitive, a shorter time period needs to be used as the predetermined time period. To obtain the historical viewing log of the target object, the consent of the target object must also be sought in advance.
[0157] The content recorded in the viewing log may include: viewing content, viewing time, content type, whether it is watched completely, whether it is liked, whether it is commented, etc. For example, Table 1 shows the historical viewing log of the target object within one day before the current time point.
[0158]
[0159] Table 1
[0160] According to the historical viewing logs of the target object within one day before the current time point shown in Table 1, the object features of the target object can be obtained through the historical viewing logs.
[0161] In one embodiment, obtaining the object features of the target object through the historical viewing log includes: inputting the historical viewing log into the object feature extraction model, and generating the object features by the object feature extraction model. The object feature extraction model can predict the object features of the target object based on the data in the historical viewing log. For example, according to the historical viewing log in Table 1, the object feature extraction model predicts the object features of the target object including: favorite music, current hot topics, etc. The advantage of this embodiment is that the object features of the target object can be obtained more accurately and comprehensively through the object feature extraction model, thereby improving the accuracy of obtaining the object features of the target object.
[0162] The advantage of the embodiment of obtaining object characteristics through the historical viewing log of the target object within a predetermined time period before the current time point is that the target object's interest level in the content that has been watched can be accurately obtained through the historical viewing log, and the target object's preferences can be predicted based on the interest level, thereby improving the accuracy of obtaining the target object's object characteristics.
[0163] In step 520, the scene feature refers to the feature of the current scene in which the target object is viewing the delivered content. The same target object may also want to view different delivered content in different scenes. For example, in a work scene, the target object may want to view work-related delivered content more. If the delivered content is of interest to the target object, the target object may perform a collection task on the delivered content so that it can be viewed at any time when needed; when at home, the target object may want to view the delivered content related to his or her own preferences more. The target object is more likely to perform a task of liking the delivered content of interest. Therefore, the scene feature can be used to determine the task completion probability of the target object performing different tasks on the delivered content. The scene features include: the characteristics of the geographical area where the target object is currently located, the characteristics of the current time period, the characteristics of the current input day type (weekday, weekend, or holiday), etc.
[0164] When obtaining the scene features of the target object, the target object's consent must be obtained in advance, which is similar to the above and will not be repeated here.
[0165] The characteristics of the current geographical area refer to the characteristics of the geographical area where the target object is currently located. This geographical area can be an administrative geographical area, such as City A, City B, etc., or it can be a geographical area divided by entities on the map, such as the XX shopping mall area, the XX hospital area, etc. For administrative geographical areas, the delivery content that the target object is interested in has an impact. For example, if the target object is in City A, the delivery content can be tourist attractions or food culture related to City A. For geographical areas divided by entities, the delivery content that the target object is interested in also has an impact. For example, if the target object is in the XX shopping mall area, the delivery content can be activity information or special projects in the mall; if the target object is in the XX hospital area, the delivery content can be content related to health and wellness.
[0166] The current geographical area feature can be obtained from the positioning information of the object terminal 110 of the target object. The object terminal 110 of the target object has a positioning device such as GPS and Beidou, and the positioning device can continuously obtain the positioning information of the object terminal 110. The geographical area where the target object is currently located can be determined based on the positioning information as the current input geographical area feature. For example, according to the GPS positioning information obtained by the positioning device, it is mapped onto an electronic map and it is found that it belongs to China, then the current input geographical area feature "China" is obtained.
[0167] The current time period feature refers to the time period within a day when the target object views the delivered content, for example, 9:00-11:00, 11:00-13:00, 13:00-18:00, 18:00-22:00, etc. Since the target object has different activities in different time periods of the day, the purpose of viewing the recommended content may also be different. Therefore, the delivered content will be affected by the current time period feature. For example, from 9:00 to 11:00, the target object may be working, so the target object is likely to want to view work-related content when viewing the delivered content; from 18:00 to 22:00, the target object is often resting, and at this time, it is likely that the target object wants to view content related to his or her interests.
[0168] The current input day type feature refers to the feature of the type of day when the target object views the delivery content. It is divided into weekdays, weekends, and holidays. The type of delivery content that the target object wants to view may be different on different types of days. For example, on weekdays (Monday to Friday), the target object is often at work, and the delivery content that the target object wants to view may be work-related; on weekends (Saturday and Sunday), when the target object does not need to work, the delivery content that the target object wants to view may be related to his or her hobbies; on holidays, the target object may want to view delivery content related to holiday culture.
[0169] The current time period feature and the current input day type feature can be obtained from the system time of the target object's object terminal 110. The system time generally includes information of year, month, day, day of the week, hour, minute, and second. The current input time period feature can be determined from the information of hour, minute, and second. For example, if the current time is 13:52:26, it is determined that it belongs to 13:00-18:00, that is, the current input time period feature is 13:00-18:00. If the current date is Tuesday, May 9, 2023, it is determined that the current input day type feature is a working day.
[0170] In step 530, a filtered content set is selected from the content set to be filtered, and the features of each filtered content in the filtered content set are used as the features of the content to be delivered in turn. The content set to be filtered refers to a content set that has not been filtered and contains various types of content. The content set to be filtered contains a lot of content that the target object is not interested in. Therefore, the content in the content set to be filtered needs to be filtered to form a filtered content set. The filtered content set contains multiple content to be delivered that the target object may be interested in.
[0171] If you want to predict the target object's task completion probability for multiple tasks of the content to be delivered, you need to take out each content to be delivered in the filtered content set in turn and obtain the features of the content to be delivered. The features of the content to be delivered are used to represent the characteristics of multiple aspects of the content to be delivered, such as content type, content length, content release time, content keywords, etc.
[0172] In step 540, the features of the content to be delivered are combined with the object features obtained in step 510 and the scene features obtained in step 520 to serve as multiple delivery basic features. Figure 6As shown in the schematic diagram of the basic features of delivery, the object features include: object feature A, object feature B, and object feature C; the scene features include: scene feature A, scene feature B, and scene feature C; the filtered content set selected from the to-be-screened content set includes: to-be-delivered content A and to-be-delivered content B, wherein the to-be-delivered content A includes to-be-delivered content feature A1 and to-be-delivered content feature A2, and the to-be-delivered content B includes to-be-delivered content feature B1 and to-be-delivered content feature B2. The to-be-delivered content features of the to-be-delivered content in the filtered content set are sequentially used together with the object features and the scene features as the basic features of delivery. The delivery basic feature A includes object feature A, object feature B, object feature C, scene feature A, scene feature B, scene feature C, to-be-delivered content feature A1, and to-be-delivered content feature A2, which are used to predict the target object's task completion probability for multiple tasks of the to-be-delivered content A. The basic delivery feature B includes object feature A, object feature B, object feature C, scene feature A, scene feature B, scene feature C, feature B1 of the content to be delivered, and feature B2 of the content to be delivered, which are used to predict the task completion probability of the target object for multiple tasks of the content B to be delivered.
[0173] In one embodiment, Figure 7 As shown, in step 530, filtering out a filtered content set from the to-be-filtered content set includes:
[0174] Step 710: Obtain a first tag of the target object;
[0175] Step 720: Obtain a second tag of the content to be filtered in the set of content to be filtered;
[0176] Step 730: Filter out a filtered content set from the to-be-filtered content set based on the matching degree between the first tag and the second tag.
[0177] In step 710, the first tag indicates the type of content that the target object is interested in.
[0178] In one embodiment, the first tag of the target object may be obtained through the historical viewing log. The historical viewing log has been described in detail in the above embodiment and will not be described again here.
[0179] Note that when obtaining the first tag of the target object, the target object's consent must be obtained in advance. In addition, the collection, use and processing of the first tag will comply with relevant laws, regulations and standards. When obtaining the target object's consent, the target object's separate permission or separate consent can be obtained through a pop-up window or by jumping to a confirmation page.
[0180] Obtaining the first label of the target object through the historical viewing log includes: determining the target object's preference for the delivered content through task statistics of the watched content of each content type in the historical viewing log; and using the preference as the first label of the target object. Since the historical viewing log records the completion status of the target object's tasks for each watched content (for example, whether it has been watched in full, whether it has been liked, whether it has been commented on, etc.), it is possible to infer whether the target object is interested in the watched content through the historical viewing log. For example, in the historical viewing log shown in Table 1, the target object watched the entire delivered content A and liked the content, so it can be inferred that the target object is interested in the delivered content A. However, the target object did not perform any operation on the delivered content B, so it can be inferred that the target object is not interested in the delivered content B. For example, Table 2 shows the task statistics for the watched content of each content type in the historical viewing log shown in Table 1.
[0181] Content Type Views Number of likes Number of comments music 15 9 5 Movie 4 1 0 Current Affairs 12 6 10 study 8 2 1 …… …… …… ……
[0182] Table 2
[0183] As shown in Table 2, the target object has watched 15 complete times, liked 9 times, and commented 5 times on the content of type "music".
[0184] Note that when obtaining the viewing content of each content type in the historical viewing log, the consent of the target object must be obtained.
[0185] In one embodiment, by counting the task completion of each content type in the historical viewing log, the first label of the target object is determined, including: calculating the task completion average of the content type, ranking the task completion average of all content types, and determining the first label of the target object. For example, according to the task statistics of the viewing content of each content type in Table 2, the task completion average of the music type content is (15+9+5) / 3=9.67; the task completion average of the movie type content is (4+1+0) / 3=1.67. After determining the task completion average of each content type, the content types are ranked in order from high to low according to the task completion average, and the first predetermined number of content types in the ranking are selected as the first label of the target object. For example, after calculating the task completion average of each content type in Table 2, the content types are ranked according to the task completion average, and the top 3 content types in the ranking are selected as the first label of the target object. According to the ranking of the task completion average of each content type in Table 2, it can be obtained that the first label of the target object for the delivery content is music, current affairs, and learning. The advantage of using the average result to determine the first label of the target object is that it fairly reflects the impact of each task on the target object's preference and improves the fairness of determining the first label.
[0186] In another embodiment, the first label of the target object is determined by counting the tasks of the viewing content of each content type in the historical viewing log, including: setting a weight for each task completion number; and determining the first label of the target object based on the weighted average ranking of the task completion of all content types. The reason for setting weights for different task completion numbers is that different task completion situations indicate that the target object has different levels of interest in the delivery content. For example, the target object may be more interested in the delivery content that is liked than in the delivery content that is fully watched; and the target object may be more interested in the delivery content that is commented on than in the delivery content that is liked. Therefore, the target object's interest in different types of delivery content can be represented by setting weights. For example, the weight of "full viewing times" is 0.2; the weight of "like times" is 0.3; and the weight of "comment times" is 0.5. The weighted average of task completion of content types is calculated based on the weights and the task completion times. For example, in Table 2, for music-type content, the weighted average value of task completion is (15*0.2+9*0.3+5*0.5) / 1=8.2; for movie-type content, the weighted average value of task completion is (4*0.2+1*0.3+0*0.5) / 1=1.1. After determining the weighted average value of task completion for each content type, the content types are ranked in descending order according to the weighted average value of task completion, and the first predetermined number of content types in the ranking are selected as the first label of the target object. For example, the top three content types in Table 2 are selected as the first label of the target object, so the target object's preferred types for the delivered content are finally obtained as current affairs, music, and learning. The advantage of using the weighted average result to determine the target object's preferred type for the delivered content is that different weights can be set for different task completion times according to the actual situation, which improves the flexibility of determining the first label.
[0187] The advantage of determining the first label of the target object by counting the tasks of viewing content of each content type in the historical viewing log is that the preference of the target object for the delivered content can be quickly obtained, thereby improving the efficiency of obtaining the first label of the target object through the historical viewing log.
[0188] In step 720, the second tag indicates the type of content to be screened.
[0189] In one embodiment, obtaining the second label of the content to be screened includes: inputting the content to be screened into a label prediction model; and the label prediction model outputting the second label of the content to be screened. For example, inputting a content to be screened into the label prediction model, the second label of the content to be screened is obtained as: study, campus, reading.
[0190] When obtaining the second tag of the content to be screened, the consent of the publisher of the content to be screened must be obtained.
[0191] The advantage of outputting the second label through the label prediction model is that the accuracy of obtaining the second label is improved.
[0192] In step 730, the matching degree between the first tag and the second tag represents the matching degree between the to-be-screened content and the target object, and the to-be-screened content whose matching degree meets the predetermined condition is added to the screened content set.
[0193] In one embodiment, the matching degree between the first label and the second label is the ratio of the overlapping labels of the second label and the first label to the total number of first labels. Figure 8 As shown, the first labels of the target object include: "student", "music", "popular", and "funny", and the second labels of the content A to be screened include: "music", "popular", and "folk songs". The overlapping labels of the first label and the second label are "music" and "popular", and the overlapping labels account for 50% of all first labels. Therefore, the matching degree between the content A to be screened and the target object is 50%. The overlapping label of the second label of the content B to be screened and the first label is "student", and the overlapping labels account for 25% of all first labels. Therefore, the matching degree between the content B to be screened and the target object is 25%. Similarly, the matching degree between the content C to be screened and the target object is 50%; the matching degree between the content D to be screened and the target object is 75%; and the matching degree between the content E to be screened and the target object is 0. The advantage of determining the matching degree by the proportion of overlapping labels in all first labels is that the matching degree between the content to be screened and the target object can be quickly determined by overlapping labels, thereby improving the efficiency of obtaining the filtered content set.
[0194] The content to be screened that meets the predetermined conditions is added to the filtered content set. In one embodiment, the predetermined condition is to add the content to be screened that matches the target object greater than or equal to the preset matching threshold to the filtered content set. For example, the preset matching threshold is 50%. Figure 8 The matching degree between the to-be-screened content and the target object is shown, and the to-be-screened content with a matching degree greater than or equal to 50% is added to the screened content set. Therefore, the screened content set includes the to-be-screened content A, the to-be-screened content C, and the to-be-screened content D. In this embodiment, the advantage of determining the screened content set based on the matching degree threshold is that it ensures that the to-be-delivered content in the screened content set is content with a high matching degree with the target object, thereby improving the accuracy of determining the delivered content from the screened content set.
[0195] In another embodiment, adding the to-be-screened contents whose matching degree meets the predetermined conditions to the set of screened contents includes: ranking the matching degree of the to-be-screened contents; setting a fixed number, and adding the first fixed number of the to-be-screened contents to the set of screened contents according to the order of the matching degree ranking from high to low. For example, if the fixed number is set to 10, the to-be-screened contents with the top 10 matching degrees with the target object are selected from the set of to-be-screened contents to be added to the set of screened contents. When the matching degree of the first label of the target object and the second label of the to-be-screened contents are both low, if the to-be-screened contents are screened according to the matching degree threshold, the number of contents in the set of screened contents may not meet the requirements for delivery to the target object. Therefore, through this embodiment, it can be ensured that the number of contents to be delivered in the set of screened contents meets the requirements for delivery to the target object. By analyzing the reaction of the target object to the contents in the set of screened contents, a more accurate first label of the target object is obtained, so as to improve the matching degree of the target object and the to-be-screened contents in subsequent screening, thereby obtaining contents with a higher matching degree. Therefore, in this embodiment, the advantage of determining the filtered content set based on the matching ranking is that it provides data support for obtaining a more accurate first label of the target object, thereby gradually improving the accuracy of determining the delivery content of the content set to be filtered.
[0196] In another embodiment, Fig. 9 As shown, based on the matching degree between the first tag and the second tag, a filtered content set is filtered out from the to-be-filtered content set, and further includes:
[0197] Step 910: based on the matching degree, select the matching content having the first tag and the second tag matching from the to-be-screened content set;
[0198] Step 920: Obtain multiple delivery indicators for each matching content;
[0199] Step 930: Based on multiple delivery indicators, filtered contents are filtered out from the matching contents to form a filtered content set.
[0200] In step 910, the process of obtaining matching content based on the matching degree between the first tag and the second tag is the same as the process of filtering the content to be filtered according to the predetermined conditions in step 730 in the above embodiment, and will not be repeated here.
[0201] In step 920, the delivery index of the matched content refers to an index used to evaluate the delivery effect of the content. The delivery effect refers to the feedback of the object viewing the content on the delivery of the content, such as whether the content has been watched to the end, whether the content has been liked, etc. The delivery index is to count the delivery effect of the content into specific values in order to evaluate the content. The delivery index can be the number of likes, comments, favorites, playback volume, completion rate, and recommendation volume of the content.
[0202] When obtaining multiple delivery indicators for each matching content, the consent of the recipients of the matching content must be obtained.
[0203] For example Fig.10 As shown, the obtained matching contents include matching contents A, matching contents C, and matching contents D. The delivery indicators of each matching content are obtained respectively, and the delivery indicators of matching content A include: number of likes (250), number of comments (1650), number of playbacks (1235), completion rate (72.5%), etc. Similarly, the delivery indicators of matching contents B and C can be obtained.
[0204] In step 930, the selected content is selected from the matching content according to the delivery index. Since the delivery index is a numerical representation of the delivery effect of the content, the delivery effect obtained after the selected content determined based on the delivery index is delivered to the target object may be better. Fig.10 As shown, according to the delivery indicators of the three matching contents, the final filtered content set includes matching content C and matching content D.
[0205] In one embodiment, based on multiple delivery indicators, filtered content is filtered out from the matching content to form a filtered content set, including: determining a sub-score of each delivery indicator; determining a matching content delivery score based on the sub-scores of all delivery indicators; and filtering out the filtered content from the matching content based on the matching content delivery score.
[0206] In one embodiment, the sub-score of each delivery indicator is determined by a delivery indicator and sub-score comparison table. Each delivery indicator has a corresponding delivery indicator and sub-score comparison table. The score of the delivery indicator can be determined by the comparison table.
[0207] For example, here is an example of a table comparing the number of likes / comments and sub-scores:
[0208] Likes / Comments Likes / Comments Sub-score More than 10,000 times 100 5000-10000 times 80 1000-5000 times 60 500-1000 times 40 200-500 times 20 Less than 200 times 10
[0209] Table 3
[0210] For example, in Fig.10 In the example, the number of likes for the matching content A is 250. By looking up Table 3, the corresponding sub-score of the number of likes is 20 points; the number of comments is 1650. By looking up Table 3, the corresponding sub-score of the number of comments is 60 points.
[0211] The following is an example of a comparison table of playback indicators and sub-scores:
[0212] Views Play Quantum Fraction More than 500,000 times 100 100000-500000 times 80 50000-100000 times 60 10000-50000 times 40 5000-10000 times 30 1000-5000 times 20 Less than 1000 times 10
[0213] Table 4
[0214] For example, in Fig.10In the figure, the playback volume of the matching content A is 3235 times. By looking up Table 4, the corresponding playback quantum score is 20 points; the playback volume of the matching content B is 456810 times. By looking up Table 4, the corresponding playback volume score is 80 points.
[0215] Here is an example of a table comparing the completion rate metric and sub-scores:
[0216] Completion rate Completion Rate Sub-Score 80%-100% 100 60%-80% 80 40%-60% 60 20%-40% 40 10%-20% 20 0-10% 10
[0217] Table 5
[0218] For example, in Fig.10 In the example, the completion rate of the matching content A is 72.5%. By looking up Table 5, the corresponding completion rate sub-score is 80 points; the completion rate of the matching content B is 46.6%. By looking up Table 5, the corresponding completion rate sub-score is 60 points.
[0219] The advantages of using a comparison table to determine the sub-scores of the delivery indicator are: easy to find and low processing overhead.
[0220] After the sub-scores of all delivery indicators are obtained, the matching content delivery score of the matching content is determined according to the sub-scores.
[0221] In one embodiment, the matching content delivery score may be the average of the delivery indicator sub-scores. Fig.10 In the example, the likes sub-score of matching content A is 20, the comments sub-score is 60, the play sub-score is 20, and the completion rate sub-score is 80. Then the matching content delivery score of matching content A is (20+60+20+80) / 4=45. The advantage of using the average to calculate the matching content delivery score is that it equally reflects the impact of different delivery indicators on matching content.
[0222] In another embodiment, the matching content delivery score can be a weighted average of the delivery indicator sub-scores. In this embodiment, it is necessary to set a weight for each delivery indicator sub-score. Based on the above example, the weight of the likes sub-score is 0.6, the weight of the comments sub-score is 0.6, the weight of the play quantum score is 0.4, and the weight of the completion rate sub-score is 0.7. Then the matching content delivery score of the matching content is (20*0.6+60*0.6+20*0.4+80*0.7) / (0.6+0.6+0.4+0.7)=48.7. The advantage of using a weighted average to calculate the matching content delivery score is that different delivery indicators have different effects on the matching content. By setting different weights for different delivery indicator sub-scores, the flexibility of screening the screened content can be improved.
[0223] After determining the matching content delivery scores of all matching contents, the filtered contents are filtered according to the matching content delivery scores and the predetermined conditions. The predetermined conditions for filtering may be:
[0224] Set a delivery score threshold, and use the matching content with a delivery score greater than the delivery score threshold as the content to be selected;
[0225] Alternatively, the matching contents are sorted from high to low according to the matching content delivery scores, and the first predetermined number of matching contents are used as the contents to be screened.
[0226] The method of filtering the filtered content according to the matching content delivery score is similar to the method of filtering the deleted content according to the matching degree of the first tag and the second tag in the aforementioned embodiment, and will not be repeated here.
[0227] The advantage of determining the filtered content based on the matching content delivery score of each matching content is that the calculation speed is fast, which improves the efficiency of determining the filtered content.
[0228] The advantage of the embodiment of steps 910-930 is that the to-be-screened content is further screened based on the delivery index, so that the content in the screened content set is more likely to be of interest to the target object, further improving the accuracy of delivery.
[0229] The advantage of the embodiment of steps 710-730 is that the delivery content is preliminarily screened according to the label of the target object, ensuring that the content to be delivered for multi-task prediction in the subsequent process is highly matched with the target object, saving computing resources and improving the matching degree between the content to be delivered and the target object.
[0230] The embodiment of steps 510-540 combines the object characteristics of the target object, the scene characteristics of the scene, and the characteristics of the content to be delivered to form the basic delivery characteristics, so that the constructed basic delivery characteristics cover all aspects that affect the probability of the target object completing the task of the content to be delivered, thereby improving the comprehensiveness of obtaining the basic delivery characteristics.
[0231] Detailed description of step 320
[0232] Since the delivery basic features may include multiple types of features, such as character type or numeric type, it is necessary to vectorize the delivery basic features into a form that can be learned by the machine.
[0233] The basic features for delivery may be character-based features, such as industry and education level in object features, geographic area features in scene features, and content type in features of content to be delivered.
[0234] If the basic feature is a character feature, the character feature can be input into the embedding layer to obtain the first feature vector corresponding to the character feature.
[0235] Since character features are sequences of characters, each character is considered a discrete symbol that cannot be recognized by machines. The role of the embedding layer is to convert discrete characters into vectors that can be recognized by machines.
[0236] Input the character feature into the embedding layer, and the embedding layer will output the corresponding unique first feature vector for the character feature. For example, Fig.11A As shown, after the character feature "student" is input into the embedding layer, the first feature vector is [1,0,1,0,...,1,1,0]; after the character feature "Shenzhen" is input into the embedding layer, the first feature vector is [0,0,1,0,...,0,1,0]; after the character feature "popular" is input into the embedding layer, the first feature vector is [1,1,0,1,...,1,0,0].
[0237] The delivery basic feature may also be a numerical feature, for example, the content delivery amount in the feature of the content to be delivered.
[0238] If the basic feature is a numerical feature, the numerical feature is numerically mapped to obtain a first feature vector corresponding to the numerical feature.
[0239] The process of value mapping can be to convert the value into its corresponding binary form. Fig. 11B The numerical mapping conversion shown in the figure. The first eigenvector obtained by the numerical feature "27" after numerical conversion is 00011011; the first eigenvector obtained by the numerical feature "560" after numerical conversion is 1000110000; the first eigenvector obtained by the numerical feature "99" after numerical conversion is 01100011.
[0240] The basic delivery feature can also be an indefinite length sequence feature. An indefinite length sequence feature refers to a string feature consisting of a large number of characters of an uncertain number. For example, the object signature in the object feature is an object feature that is filled in and formed by the object itself.
[0241] If the basic feature is a variable-length sequence feature, the variable-length sequence feature is pooled to obtain a first feature vector corresponding to the variable-length sequence feature.
[0242] Pooling is the process of reducing the dimensionality of a high-dimensional vector. The specific process of pooling is to integrate several features in a preset neighborhood of a high-dimensional vector generated based on indefinite length sequence features into a new feature. Fig. 11CThe pooling process shown is shown in the figure. The indefinite length sequence feature is the object signature in the object feature. The object signature is "Reading makes me progress,...". According to the indefinite length sequence feature, the corresponding high-dimensional feature vector can be obtained, and the high-dimensional feature vector is [0,1,1,0,0,1,0,0,0,1,1,1,0,0,0,1,0,1,0,0,0,1,1,1]. The value of the preset neighborhood is set to 3, that is, every 3 feature quantities in the high-dimensional feature vector are taken as a neighborhood, and the 3 feature quantities are integrated into 1 feature quantity. The integration method can be to take the average value of the feature quantity in the neighborhood as the feature quantity of the neighborhood. For example, in the above high-dimensional feature vector, the first neighborhood is taken as [0,1,1], and the average value of the feature quantity in this neighborhood is 0.67. Therefore, the feature value of this neighborhood is 0.67. The second neighborhood is [0,0,1], and the average value of the feature quantity in this neighborhood is 0.33. Therefore, the feature value of this neighborhood is 0.33. The third neighborhood is [0,0,0], and the average value of the feature quantity in this neighborhood is 0. Therefore, the feature value of this neighborhood is 0. By analogy, the feature quantities of the subsequent neighborhoods are 1, 0, 0.67, 0, and 1. Therefore, the first feature vector obtained after pooling of the indefinite length sequence feature is [0.67, 0.33, 0, 1, 0, 0.67, 0, 1].
[0243] Integrating several features in a preset neighborhood into a new feature can also be done by taking the maximum value of the feature quantity in the neighborhood as the feature value of a neighborhood. For example, in the above high-dimensional feature vector, the value of the preset neighborhood is also set to 3, and the maximum value in the neighborhood is taken as the feature value of the neighborhood. Then, the first feature vector obtained after the indefinite length sequence feature is pooled is [1,1,0,1,0,1,0,1].
[0244] The advantage of using different vectorization methods for different types of delivery basic features is that the features can be vectorized according to the characteristics of different types of delivery basic features, thereby ensuring the accuracy of the feature vectorization results.
[0245] Detailed description of step 330
[0246] In step 330, each first feature vector is input into a first expert network corresponding to the first feature vector. The first expert network is a network for multi-objective modeling of feature vectors. Each first feature vector has a corresponding first expert network, so each first feature vector will obtain a corresponding output for a different task.
[0247] The first expert network includes multiple first dedicated models and multiple first gating nodes. Each task corresponds to a first dedicated model and a first gating node. The first dedicated model is used to perform task modeling for features corresponding to the task; the first gating node is used to generate high-order features corresponding to the task based on the output of the first dedicated model. The first dedicated model and the first gating node are both established based on the characteristics of the task and are not applicable to other tasks. For example, a task is to predict whether an object will like a post. The first dedicated model and the first gating node are generated based on this task and cannot be used to predict other tasks, such as predicting whether an object will comment.
[0248] In one embodiment, the first expert network also includes a first shared model. The first shared model is a general model that serves multiple tasks and does not correspond to any specific task. The first shared model is generated based on all tasks. There may be connections between different tasks that can influence each other. For example, if the target object collects a content, it means that the target object recognizes the content and is therefore likely to like the content. However, the first dedicated model only serves the corresponding task and does not consider other tasks. Therefore, the first shared model can be used in the first expert network to add correlation factors between tasks when generating task-related features corresponding to the task, and to balance the relationship between the first dedicated models corresponding to different tasks.
[0249] like Fig.12 As shown, the first gating node generates a task-related feature corresponding to the task and the first feature vector based on the output of the first dedicated model, including:
[0250] Step 1210: If the first gating node corresponds to a first dedicated model, the first gating node performs a weighted sum of an output of the first shared model and an output of the first dedicated model corresponding to the first gating node to obtain a task-associated feature corresponding to the task and the first feature vector;
[0251] Step 1220: If the first gating node corresponds to the first shared model, the first gating node performs a weighted sum of the output of the first shared model and the outputs of each first dedicated model to obtain a first shared feature corresponding to the first feature vector.
[0252] In step 1210, when the first gating node outputs the task-associated features corresponding to the task, it is necessary to use the output of the first dedicated model and the output of the first shared model corresponding to the task at the same time. The output of the first dedicated model is the result of modeling the first feature vector for the task; the first shared model is the result of generalized modeling of the first feature vector for multiple tasks, so that the generated task-associated features can be associated with the features of other tasks. The first gating node generates the task-associated features corresponding to the task by calculating the weighted sum of the output of the first shared model and the output of the first dedicated model. The purpose of using the weighted sum to generate the task-associated features is that the effects of specialized modeling of the first feature vector for the task and generalized modeling of the first feature vector for multiple tasks on the generated task-associated features may be different.
[0253] Therefore, in one embodiment, if Fig.13 As shown, the output of the first shared model and the output of the first dedicated model corresponding to the first gating node are weighted and obtained to obtain the task-related features corresponding to the task and the first feature vector, including:
[0254] Step 1211: Based on the first feature vector, obtain the first weight of the first shared model and each first dedicated model;
[0255] Step 1212, using the first shared model and the first weight of the first dedicated model corresponding to the first gating node, perform weighted sum on the output of the first shared model and the output of the first dedicated model corresponding to the first gating node to obtain task-associated features corresponding to the task and the first feature vector.
[0256] Based on the operation on the first feature vector, the first weights of the first shared model and each first dedicated model can be obtained.
[0257] In one embodiment, Fig.14 As shown, based on the first feature vector, obtaining the first shared model and the first weights of each first dedicated model includes:
[0258] Step 1410: Obtain the first shared model and the first matrix of each first dedicated model;
[0259] Step 1420, performing a first multiplication operation on the transpose of the first matrix and the first eigenvector;
[0260] Step 1430: Perform normalized exponential processing on the result of the first multiplication operation to obtain a first weight.
[0261] The values in the first matrix set for the first shared model and the different first dedicated models are different, but the number of columns and rows of the first matrix for different models are the same. The number of columns of the first matrix is the total number of models of the first dedicated model and the first shared model, and the number of rows of the first matrix is the dimension of the first eigenvector. For example, if the number of first shared models is 1 and the number of first dedicated models is 3, then the number of columns of the first matrix is 4; if the dimension of the first eigenvector is 3, then the number of rows of the first matrix is 3.
[0262] The first multiplication operation is performed on the transpose of the first matrix and the first eigenvector, which is equivalent to multiplying a matrix by a column vector. The result is a column vector with the dimension of the first eigenvector. For example, the first matrix is The first eigenvector is [0,1,1], so the calculation process of the first multiplication operation is shown in Formula 1:
[0263]
[0264] After obtaining the result of the first multiplication operation, the result is normalized exponentially to obtain the first weight. The purpose of normalized exponential processing is to limit the result of the first multiplication operation to a specific range so that the values of the first weights of different models are within the same range and comparable. The normalized exponential processing uses the softmax function. The softmax function inputs the result of the first multiplication operation and outputs a Therefore, the process of calculating the first weight of the first shared model is shown in Formula 2:
[0265]
[0266] In formula 2, k refers to the serial number of the basic features. Assuming there are 20 basic features, and the age feature of the object is the sixth of the 20 basic features, then the k value of the object feature is 6. Therefore, x k Refers to the first feature vector corresponding to the kth basic feature of the delivery. s refers to the first shared model or the sequence number of all first shared models in the first expert network. For example, in a first expert network, there are 2 first shared models, then the s value of the first first shared model is 1, and the s value of the second first shared model is 2. Therefore, Refers to the first matrix of the sth first shared model in the first expert network corresponding to the first feature vector of the kth delivery basic feature. Refers to the first weight of the sth first shared model in the first expert network corresponding to the first feature vector of the kth delivery basic feature.
[0267] The process of calculating the first weight of the first dedicated model is shown in Formula 3:
[0268]
[0269] The difference between Formula 3 and Formula 2 is that t in Formula 3 represents the sequence number of the task corresponding to the first dedicated model among all tasks. For example, there are 3 tasks, each of which corresponds to a first dedicated model, and the sequence number of the task corresponding to one of the first dedicated models is 2, then the value of t is 2. Therefore, Refers to the first matrix of the t-th first dedicated model in the first expert network corresponding to the first feature vector of the k-th basic feature, Refers to the first weight of the tth first dedicated model in the first expert network corresponding to the first feature vector of the kth delivery basic feature.
[0270] This embodiment obtains the first weights of each first shared model and the first dedicated model through normalized index processing, so that the first weights of each model are within the same numerical range, which can more effectively reflect the impact of the outputs of different models on the acquisition of task-related features, thereby improving the accuracy of the acquired task-related features.
[0271] In step 1212, the output of the first shared model and the output of the first dedicated model corresponding to the first gating node are weighted and added using the first shared model and the first weight of the first dedicated model corresponding to the first gating node. The specific calculation process is shown in Formula 4:
[0272]
[0273] In formula 4, m s represents the number of the first shared model. Therefore, It refers to the high-order output of the first feature vector of the kth basic feature delivered by the sth first shared model in the first expert network. The dimension of the high-order output vector is therefore, It refers to the first weight of the sth first shared model multiplied by the output of the first shared model, with a dimension of Refers to
[0274] The first special model corresponding to the first gating node in the first expert network corresponding to the first feature vector of the k basic features of the delivery has a high-order output of the first feature vector. The dimension of the high-order output vector is also therefore, Refers to the first weight of the first dedicated model corresponding to the first gating node multiplied by the output of the first dedicated model, with a dimension of M s The output of the first shared model, the output of the first dedicated model, and the first weight corresponding to the model are weighted and calculated to obtain the task-related feature output by the first gating node. The dimension is
[0275] The advantage of the embodiment of steps 1211-1212 is that the task association vector is obtained by using different first weights of the first shared model and the first dedicated model, which reflects the influence of different models on generating the task association vector and improves the accuracy of obtaining the task association vector.
[0276] In step 1220, since the first shared model is a general model serving multiple tasks, if the first gating node corresponds to the first shared model, the first shared feature output by the first gating node needs to be generated in combination with the outputs of all first dedicated models. Therefore, the vector generated by the first gating node corresponding to the first shared model needs to be generated in combination with the output of the first shared model and the output of all first dedicated models.
[0277] In one embodiment, Fig.13 As shown, the output of the first shared model and the output of each first dedicated model are weighted and summed to obtain a first shared feature corresponding to the first feature vector, including:
[0278] Step 1221: Using the first shared model and the first weights of each first dedicated model, perform a weighted sum on the output of the first shared model and the output of each first dedicated model to obtain a first shared feature corresponding to the first feature vector.
[0279] Similar to steps 1211 - 1212 , the first gating node generates a first shared feature by calculating a weighted sum of the output of the first shared model and the outputs of each first dedicated model using the first weight.
[0280] The method for calculating the first weight of the first shared model is shown in Formula 2; the method for calculating the first weight of the first dedicated model is shown in Formula 3, which will not be repeated here.
[0281] After obtaining the first weights of the first shared model and each first dedicated model, the output of the first shared model and the output of each first dedicated model are weighted and summed using the first shared model and the first weights of each first dedicated model. The specific calculation process is shown in Formula 5:
[0282]
[0283] In formula 5, m p represents the number of first specialized models in the first expert network. Therefore, Refers to the output of the first dedicated model corresponding to the t-th task in the expert network corresponding to the first feature vector of the k-th basic feature, with a dimension of It refers to the first weight of the first dedicated model corresponding to the t-th task multiplied by the output of the first dedicated model, with a dimension of M s The output of the first shared model, m p The output of the first dedicated model is weighted and calculated with the first weight corresponding to the model to obtain the first shared feature of the first gating node output corresponding to the first shared model and the first feature vector.
[0284] The advantages of step 1221 are similar to those of the embodiments of steps 1211-1212, and can improve the accuracy of obtaining the first shared feature, which will not be repeated here.
[0285] In the embodiment of steps 1210-1220, if the categories of the models corresponding to the first gating nodes are different, the output results are different, and the calculation methods for obtaining the output results are also different. Therefore, the advantage of this embodiment is that different calculation methods are used for the first gating nodes corresponding to different tasks, so that the obtained calculation results are more in line with the requirements of multi-objective modeling of the first feature vector, which is conducive to obtaining more accurate multi-task prediction results and improving the accuracy of determining the delivery content.
[0286] The advantage of the embodiment of adding the first shared model to the first expert network is that the possible connections between different tasks can be considered in the first expert network, and the relationship between the first dedicated models corresponding to different tasks can be balanced, so that the first expert network can obtain more accurate modeling results when performing multi-task feature modeling, thereby improving the accuracy of multi-task prediction results.
[0287] In another embodiment, Fig.15 As shown, each first feature vector is input into a first expert network corresponding to the first feature vector, including:
[0288] Step 1510: input each first feature vector into a plurality of first expert networks corresponding to the first feature vector, and obtain a plurality of candidate task-associated features corresponding to the task and the first feature vector;
[0289] Step 1520: Based on a predetermined criterion and in accordance with a plurality of candidate task-associated features, determine a task-associated feature corresponding to the task and the first feature vector.
[0290] In this embodiment, each first feature vector corresponds to multiple first expert networks. The purpose of setting multiple first expert networks is that if a first expert network is used to model the first feature vector for a task, the obtained task-related features only depend on this first expert network. Even if the task-related features obtained by this first expert network are inaccurate, they must be used as the modeling result of the first feature vector for the task. Therefore, in this embodiment, multiple corresponding first expert networks are set for the first feature vector. The first feature vector is output to multiple first expert networks, and each first expert network will generate candidate task-related features for the task and the first feature vector.
[0291] For example, in Fig.16 In , the first feature vector corresponds to three first expert networks, namely, the first expert network A, the first expert network B, and the first expert network C. For a task, the first expert network A generates a candidate task-related feature A for the first feature vector; the first expert network B generates a candidate task-related feature B for the first feature vector; and the first expert network C generates a candidate task-related feature C for the first feature vector.
[0292] The process of each first expert network generating candidate task-related features is the same as the process of the first expert network generating task-related features in the aforementioned embodiment, and will not be repeated here.
[0293] After each first expert network generates candidate task-related features for the first feature vector for the task, the task-related features are determined based on the candidate task-related features according to a predetermined criterion. The predetermined criterion is the basis for determining the task-related features.
[0294] Detailed description of step 340
[0295] The first expert network corresponding to the first feature vector of each basic feature of the delivery generates multiple task-related features for multiple tasks and the first feature vector. Fig.17A In the first expert network A corresponding to the first feature vector A output, the first gating node A1 in the first expert network A will generate a task-related feature for the first feature vector A and task 1; the first gating node A2 will generate a task-related feature for the first feature vector A and task 2; the first gating node A3 will generate a task-related feature for the first feature vector A and task 3. Similarly, the first expert network B will generate task-related features for task 1, task 2, and task 3 for the first feature vector B; the first expert network C will generate task-related features for task 1, task 2, and task 3 for the first feature vector C.
[0296] At present, there are multiple task-related features related to the task. In order to perform the subsequent multi-task prediction process, it is necessary to generate a second feature vector corresponding to the task based on multiple task-related features generated by different first expert networks for the same task and the corresponding first feature vector.
[0297] In one embodiment, generating a second feature vector for a task based on a plurality of task-related features corresponding to the task and a plurality of first feature vectors includes: cascading the plurality of task-related features to generate the second feature vector.
[0298] Cascading multiple task-related features refers to the process of concatenating the task-related features in the order of the first expert network that generated the task-related features. Fig.17A In the example, the task-related features generated by the first expert network A for the first feature vector A and task 1 are [0, 1.35, 1.5, 0.58, 0]; the task-related features generated by the first expert network B for the first feature vector B and task 1 are [1.23, 1, 1.97, 2, 0.27]; the task-related features generated by the first expert network C for the first feature vector C and task 1 are [0.56, 1.39, 2.21, 0.75, 1]. The second feature vector generated for task 1 is cascaded to obtain: [0, 1.35, 1.5, 0.58, 0, 1.23, 1, 1.97, 2, 0.27, 0.56, 1.39, 2.21, 0.75, 1].
[0299] The advantage of generating the second feature vector by cascading multiple task-related features is that the second feature vector can be obtained by connecting the features end to end in series, thereby improving the efficiency of generating the second feature vector.
[0300] In another embodiment, generating a second feature vector of the task based on multiple task-related features corresponding to the task and multiple first feature vectors includes: arranging and combining feature values at the same position in the multiple task-related features to generate a second feature vector corresponding to the task.
[0301] Arranging and combining feature values at the same position in multiple task-related features means taking feature values at the same position in multiple task-related features and arranging the feature values in the order of the first expert network that generates the task-related features. When the feature value arrangement of one position is completed, the feature value of the next position is arranged in the same way after the arrangement result of the previous position.
[0302] For example, in Fig. 17BIn the example, the first expert network A generates the first feature vector A and the first task-related feature of task 1 with 0; the first expert network B generates the first feature vector B and the first task-related feature of task 1 with 1.23; the first expert network C generates the first feature vector C and the first task-related feature of task 1 with 0.56; the first positions of the three first expert networks are arranged as [0, 1.23, 0.56]. Similarly, the second positions of the three first expert networks are 1.35, 1, and 1.39 respectively. After arranging the three second positions in the arrangement results of the first positions, we get [0, 1.23, 0.56, 1.35, 1, 1.39]. By analogy, the second feature vectors corresponding to the tasks generated according to the three task-related features are [0, 1.23, 0.56, 1.35, 1, 1.39, 1.5, 1.97, 2.21, 0.58, 2, 0.75, 0, 0.27, 1].
[0303] The feature values at the same position in different task-related features may represent the same type of features. Therefore, the advantage of generating the second feature vector by permuting and combining the feature values at the same position in multiple task-related features is that the feature values at the same position are placed together, which facilitates the identification of the same type of features in the second expert network and improves the accuracy of multi-task modeling of the second feature vector.
[0304] After the second feature vector is generated, the second feature vector is input into a second expert network corresponding to the task to obtain a third feature vector corresponding to the task. The second expert network is used to model the second feature vector for the task.
[0305] In one embodiment, the second expert network includes a second dedicated model corresponding to the task and a second gating node corresponding to the task. Fig.18 As shown, the second feature vector is input into the second expert network corresponding to the task to obtain a third feature vector corresponding to the task, including:
[0306] Step 1810: input the second feature vector into a second dedicated model corresponding to the task to obtain an output of the second expert model;
[0307] Step 1820: Generate a third feature vector corresponding to the task based on the output of the second dedicated model through the second gating node.
[0308] Since the second expert network corresponds to the task, the second expert network includes a second dedicated model and a second gating node corresponding to the task. The second dedicated model is used to model the second feature vector for the task; the second gating node is used to generate high-order features corresponding to the task based on the output of the second dedicated model. The second dedicated model and the second gating node are only applicable to the corresponding task and are not affected by other tasks.
[0309] In one embodiment, the second expert network further includes a second sharing model. Fig.19 As shown, generating a third feature vector corresponding to the task based on the output of the second dedicated model through the second gating node includes:
[0310] Step 1910: Generate a first shared feature vector based on a plurality of first shared features corresponding to a plurality of first feature vectors, and input the first shared feature vector into a second shared model;
[0311] Step 1920: Through the second gating node, perform a weighted sum on the output of the second shared model and the output of the second dedicated model corresponding to the second gating node to obtain a third feature vector corresponding to the task.
[0312] The second shared model is similar to the first shared model. It is a general model that serves multiple tasks and does not correspond to any specific task. Therefore, the second expert network corresponding to each task contains the same second shared model. The second shared model is used to add correlation factors between tasks when generating the third feature vector corresponding to the task.
[0313] In step 1910, the input of the second shared model is generated based on the first shared features generated by the multiple first expert networks in the aforementioned embodiment. In the aforementioned embodiment, each first expert network contains a first shared model, and the first gating node corresponding to the first shared model generates a first shared feature according to the output of the first shared model. The first shared features output by the multiple first expert networks are combined to generate a first shared feature vector. Based on the multiple first shared features, the first shared feature vector can be generated by cascading the multiple first shared features, or by arranging and combining the feature values at the same position in the multiple first shared features. The detailed process of generating the first shared feature vector is similar to the process of generating the second feature vector based on the task-related features in the aforementioned embodiment, and will not be repeated here.
[0314] In step 1920, the output of the second shared model and the output of the second dedicated model corresponding to the second gating node are input into the second gating node, and the third eigenvector corresponding to the task is obtained by the second gating node. The output of the second shared model and the output of the second dedicated model corresponding to the second gating node have different effects on generating the third eigenvector. Therefore, the third eigenvector corresponding to the task is obtained by a weighted sum method.
[0315] In one embodiment, Fig. 20 As shown, through the second gating node, the output of the second shared model and the output of the second dedicated model corresponding to the second gating node are weighted and obtained to obtain a third feature vector corresponding to the task, including:
[0316] Step 210: cascade multiple first feature vectors to obtain a first cascade feature vector;
[0317] Step 2020: Based on the first cascade feature vector, obtain the second weights of the second shared model and the second dedicated model corresponding to the second gating node;
[0318] Step 2030: Using the second weights of the second shared model and the second dedicated model corresponding to the second gating node, perform weighted sum on the output of the second shared model and the output of the second dedicated model corresponding to the second gating node to obtain a third feature vector corresponding to the task.
[0319] In step 210, a plurality of first feature vectors are cascaded. The specific process is the same as the process of generating a second feature vector by cascading a plurality of task-related features in the aforementioned embodiment, and will not be described in detail here.
[0320] The purpose of obtaining the first cascade feature vector is that in a subsequent process, the second weights of the second shared model and the second dedicated model need to be obtained based on the first cascade feature vector.
[0321] In step 2020, the first cascade feature vector is operated to obtain the second weights of the second shared model and the second dedicated model corresponding to the second gating node.
[0322] In one embodiment, Fig.21 As shown, based on the first cascade feature vector, obtaining the second weights of the second shared model and the second dedicated model corresponding to the second gating node, respectively, includes:
[0323] Step 2110: Obtain the second shared model and the second matrix of each second dedicated model;
[0324] Step 2120, performing a second multiplication operation on the second matrix and the transpose of the first cascaded eigenvector;
[0325] Step 2130: Perform normalized exponent processing on the operation of the second multiplication operation to obtain a second weight.
[0326] The second matrix set for the second shared model and each second dedicated model is different, but the number of columns and rows of the second matrices of different models are the same. The number of columns of the second matrix is the total number of models of the second dedicated model and the second shared model in the second expert network, and the number of rows of the second matrix is the dimension of the first cascade feature vector. For example, if the number of second shared models is 2 and the number of second dedicated models is 1, then the number of columns of the second matrix is 3; if the dimension of the first cascade feature vector is 5, then the number of rows of the second matrix is 5.
[0327] The second multiplication operation is performed on the transpose of the second matrix and the first concatenated eigenvector, which is equivalent to multiplying a matrix by a column vector, and the result is a column vector with the dimension of the first concatenated eigenvector. The specific process is similar to the first multiplication operation in the above embodiment, and will not be repeated here.
[0328] After obtaining the result of the second multiplication operation, the result is subjected to normalized exponential processing to obtain the second weight. The process of obtaining the second weight of the second shared model and the second dedicated model is shown in Formula 6:
[0329] g j (X) = softmax(W j ·X T ) (Formula 6).
[0330] In Formula 6, j represents the sequence number of the second shared model or the second dedicated model in the second expert network. X refers to the first cascade feature vector. s Refers to the second matrix of the second shared model or the second dedicated model. s (X) refers to the second weight of the second shared model.
[0331] The process of obtaining the second weight in this embodiment is similar to the process of obtaining the first weight in the above embodiment, and will not be repeated here. The advantage of this embodiment is that the second weight more accurately reflects the influence of different models on generating the third eigenvector, thereby improving the accuracy of obtaining the third eigenvector.
[0332] In step 2030, the output of the second shared model and the output of the second dedicated model corresponding to the second gating node are weighted and added using the second weights of the second shared model and the second dedicated model corresponding to the second gating node. The specific calculation process is shown in Formula 7:
[0333]
[0334] In Formula 7, n srefers to the number of the second shared models in the second expert network, n p Refers to the number of second specialized models in the second expert network. j (X) refers to the second weight of the model with sequence number j in the second expert network. E is a matrix storing the outputs of the second shared model and the second dedicated model in the second expert network.
[0335] The matrix E can be expressed as: In the matrix E, s indicates the second shared model and t indicates the second dedicated model; refers to the first shared eigenvector, refers to the second eigenvector; refers to the output of the second shared model after the first shared feature vector is input into the second shared model; Refers to the output of the second special model after the second eigenvector is input into the second special model. Therefore, there are n s +n p vectors, each vector represents a second shared model or a second dedicated model in the second expert network. j represents the index value of the vector corresponding to the second shared model or the second dedicated model in the matrix E.
[0336] For example, a second shared model has a sequence number of 0 in the second expert network, based on the second matrix W 0 With X T , obtain the second weight g of the second shared model 0 (X). In the matrix E, the vector with index 0 represents the output of the second shared model based on the first shared feature vector, that is, E 0 . 0 (X) and E 0 The next second shared model in the second expert network has a sequence number of 1, based on the second matrix W 1 With X T , obtain the second weight g of the second shared model 1 (X). In the matrix E, the vector with index 1 represents the output of the second shared model based on the first shared feature vector, that is, E 1 . 1 (X) and E 1 The second dedicated model is numbered 2 in the second expert network, based on the second matrix W 2 With X T , obtain the second weight g of the second dedicated model 2(X). In the matrix E, the vector with index 2 represents the output of the second dedicated model based on the second eigenvector, that is, E 2 . 2 (X) and E 2 Multiply them together to obtain the weight value of the second dedicated model when the second gating node is used to generate the third eigenvector. Therefore, the third eigenvector output by the second gating node is g 0 (X)*E 0 +g 1 (X)*E 1 +g 2 (X)*E 2 .
[0337] In the embodiment of steps 2010-2030, the second weight of the second shared model and the second dedicated model corresponding to the second gating node is obtained based on the first feature vector. That is, the second weight is adapted to the characteristics of the original feature vector and is not affected by the multi-task modeling process of the feature. Therefore, the advantage of this embodiment is that the obtained second weight is more accurate.
[0338] The advantage of the embodiment of steps 1910-1920 is that a second shared model is introduced into the second expert network, the correlation between different tasks is taken into account in the process of calculating the third eigenvector, and the third eigenvector is calculated using the second weights that are different between the second shared model and the second dedicated model, which reflects the influence of different models on the generation of the third feature and improves the accuracy of the obtained third eigenvector.
[0339] The advantage of the embodiment of steps 1810-1820 is that a high-order representation of the second eigenvector is generated by the second expert model, and then the third eigenvector is generated by the second gating node. The second expert model and the second gating node have clear division of labor in the second expert network, and the data of each step can be quickly obtained, thereby improving the efficiency of obtaining the third eigenvector.
[0340] Detailed description of step 350
[0341] In step 350, the third feature vector is input into the prediction model corresponding to the task, and the prediction model outputs the predicted target object's task completion probability for the content to be delivered and the corresponding task.
[0342] For example, the task corresponding to the second expert network is: whether the target object will like the content to be delivered. Then, the third feature vector output by the second gated node in the second expert network is input into the prediction model corresponding to the task, and the prediction model will output the task completion probability for this task. For example, the task completion probability is 0.85, which means that the target object has an 85% chance of liking the content to be delivered.
[0343] The prediction model corresponding to each task will output the task completion probability of the target object for the corresponding task.
[0344] The process of the prediction model obtaining the task completion probability of the corresponding task is shown in Formula 8:
[0345]
[0346] In formula 8, represents the third feature vector output by the second gating node in the second expert network; p t represents the task completion probability obtained by the prediction model; θ t Indicates the prediction model corresponding to the task, which can be a deep neural network model.
[0347] The process of determining the content to be delivered to the target audience
[0348] After obtaining the task completion probability of the target object for each task, the content to be delivered to the target object is determined based on the task completion probabilities of different tasks.
[0349] In one embodiment, Fig. 22 As shown, after the third feature vectors corresponding to the multiple tasks are input into the prediction models corresponding to the multiple tasks to obtain the predicted task completion probabilities of the target objects for the content to be delivered with respect to the multiple tasks, the delivery processing method further includes:
[0350] Step 2210: Obtain the focus tasks of the target object;
[0351] Step 2220: Obtain the task completion probability of the task of interest from the task completion probabilities of the plurality of tasks;
[0352] Step 2230: sort the content to be delivered based on the task completion probability of the concerned task;
[0353] Step 2240: Deliver the content to be delivered to the target object according to the ranking.
[0354] The target object's attention task refers to the task that can represent the target object's interest in the delivered content among multiple tasks. For example, multiple tasks include: whether to like, whether to comment, and whether to forward. If the target object likes to comment on the content of interest when viewing the delivered content, then the target object's attention task is "whether to comment".
[0355] In one embodiment, obtaining the focus tasks of the target object includes: obtaining the historical viewing logs of the target object within a predetermined time period before the current time; and determining the focus tasks of the target object based on the historical viewing logs.
[0356] When obtaining the target object's attention task, the target object's consent must be obtained in advance. This is similar to the previous content, so I will not repeat it.
[0357] In the above embodiment, the object characteristics of the target object are obtained through the historical viewing log. The historical viewing log refers to the relevant records of the delivery content viewed by the target object in a predetermined time period before the current time point, including the task execution status for each delivery content. The historical viewing log has been introduced in detail in the above embodiment and will not be repeated here. When obtaining the historical viewing log of the target object, the consent of the target object must be obtained in advance.
[0358] In one embodiment, determining the focus task of the target object based on the historical viewing log includes: counting the number of executions of each task in the historical viewing log; and taking the task with the highest number of executions as the focus task of the target object.
[0359] Take the historical viewing log in Table 1 as an example, in which the tasks performed on the delivered content include: whether to watch the whole thing, whether to like it, and whether to comment. Get the number of times each task is executed in the historical viewing log, that is, the number of times the value corresponding to each task is "yes". In Table 1, the number of times "whether to watch the whole thing" is 4 times, the number of times "whether to like it" is 2 times, and the number of times "whether to comment" is 2 times. Therefore, it can be inferred from Table 1 that the operation that the target object may perform on the content of interest is "watch the whole thing". Therefore, the target object's focus task is "whether to watch the whole thing".
[0360] In this embodiment, the focus task is determined by counting the number of times the target object executes each task, which ensures that the focus task is the action most frequently executed by the target object on the delivered content, thereby improving the accuracy of obtaining the focus task.
[0361] The advantage of determining the attention task of the target object based on the historical viewing log is that the attention task is determined by analyzing the short-term actions of the target object, which improves the matching degree between the attention task and the target object, thereby improving the accuracy of obtaining the attention task.
[0362] After step 350, the prediction model corresponding to each task will output the task completion probability of the target object for the task. The task completion probability of the target object's focus task is determined from the task completion probabilities of the multiple tasks. Fig.23 As shown, for the content A to be delivered, the content B to be delivered, and the content C to be delivered, the task completion probabilities of task A, task B, task C, and task D are obtained through the prediction models corresponding to multiple tasks. The task of interest of the target object is task B, so the task completion probabilities of task B of the content A to be delivered, the content B to be delivered, and the content C to be delivered are obtained.
[0363] After obtaining the task completion probability of each task to be delivered, the task completion probability is sorted. Fig.23 In the figure, the task completion probability of task B of content A to be delivered is 0.56, the task completion probability of task B of content B to be delivered is 0.37, and the task completion probability of task B of content C to be delivered is 0.73. Then, the three contents to be delivered are sorted in descending order according to the task completion probability of task B, and the sorting result is: content C to be delivered, content A to be delivered, and content B to be delivered.
[0364] The order in which the content to be delivered is sorted by the probability of completing the task is the order in which the content to be delivered is delivered to the target object. In the above example, the content C to be delivered is delivered to the target object first, because the target object has the highest probability of executing task B on the content C to be delivered, which means that the target object is most likely to be interested in the content C among all the content to be delivered.
[0365] The advantage of the embodiment of steps 2210-2240 is that the delivery order of the content to be delivered is determined by obtaining the focus tasks of the target object, so that the priority delivered content to be delivered is more likely to be of interest to the target object, thereby improving the matching degree between the delivered content and the target object.
[0366] In another embodiment, Fig.24 As shown, based on the task completion probability of the concerned task, the content to be delivered is sorted, including:
[0367] Step 2410: If there are multiple tasks of interest, determine the overall task completion probability based on the task completion probabilities of the multiple tasks of interest;
[0368] Step 2420: Sort the content to be delivered based on the overall task completion probability.
[0369] The target object may perform multiple tasks for the content that it is interested in. For example, if the target object is interested in a delivery content, it may like the delivery content and comment on it.
[0370] If there are multiple tasks of interest, then the task completion probability of each task of interest is obtained, and then the overall task completion probability of the content to be delivered is determined based on the task completion probability of each task of interest.
[0371] For example, the task completion probabilities of multiple focus tasks for multiple contents to be delivered are shown in Table 6:
[0372] Content to be delivered Focus on Task A Focus on Task B Content to be delivered A 0.45 0.56 Content to be delivered B 0.39 0.78 Content to be delivered C 0.69 0.44
[0373] Table 6
[0374] As shown in Table 6, the target object's attention tasks include attention content A and attention content B.
[0375] The overall task completion probability can be determined by calculating the average of the task completion probabilities of multiple tasks of interest. For example, according to Table 1, the overall task completion probability of content A to be delivered is (0.45+0.56) / 2=0.505. The overall task completion probability of content B to be delivered is (0.39+0.78) / 2=0.585. The overall task completion probability of content C to be delivered is (0.69+0.44) / 2=0.565. Therefore, based on the overall task completion probability, the result of sorting the content to be delivered is: content B to be delivered, content C to be delivered, and content A to be delivered.
[0376] The advantage of calculating the overall task completion probability by averaging is that the task completion probabilities of different focus tasks have the same impact on determining the overall task completion probability, which improves the fairness of determining the overall task completion probability.
[0377] The overall task completion probability can also be determined by calculating the weighted average of the task completion probabilities of multiple tasks of interest. In this embodiment, weights must first be set for different tasks of interest. For example, in the above example, the weight set for task A is 0.6, and the weight set for task B is 0.4. Then, the overall task completion probability of content A to be delivered is 0.45*0.6+0.56*0.4=0.494; the overall task completion probability of content B to be delivered is 0.39*0.6+0.78*0.4=0.546; the overall task completion probability of content C to be delivered is 0.69*0.6+0.44*0.4=0.59. Therefore, based on the overall task completion probability, the result of sorting the content to be delivered is: content C to be delivered, content B to be delivered, and content A to be delivered.
[0378] The advantage of calculating the overall task completion probability by weighted average is that the weights of the task completion probabilities of different focus tasks can be adjusted according to the actual application situation, which improves the flexibility of calculating the overall task completion probability.
[0379] The advantage of the embodiment of steps 2410-2420 is that the delivery order of the content to be delivered is determined by obtaining multiple focus tasks of the target object, so that the final delivery order obtained does not only depend on the task completion probability of one task, thereby improving the accuracy of determining the delivery order of the content to be delivered.
[0380] Training process of the first expert network, the second expert network, and the prediction model in the disclosed embodiment
[0381] In order to train the first expert network, the second expert network, and the prediction model in the embodiment of the present disclosure, a training sample set is constructed. The training sample set includes multiple training samples. Each training sample contains multiple basic features for delivery and multiple training labels for multiple tasks. The training label indicates whether the task is completed for the training sample. For example, the multiple training labels in a training sample are {Task A: 1, Task B; 0, Task C: 0}, where the training label 1 represents that the task is completed, and the training label 0 represents that the task is not completed. That is to say, in this example, Task A is completed, and Task B and Task C are not completed.
[0382] The first expert network, the second expert network, and the prediction model are trained based on the training sample set.
[0383] In one embodiment, Fig.25 As shown, the first expert network, the second expert network, and the prediction model are jointly trained through the following process:
[0384] Step 2510: for a single task, based on the training labels and the task completion probability of the task, determine the error subfunction of the task;
[0385] Step 2520: Determine the error function of the training sample based on the error sub-function;
[0386] Step 2530: Based on the error function, jointly train the first expert network, the second expert network, and the prediction model.
[0387] In this embodiment, first, basic delivery features in a training sample are obtained, and steps 310 - 350 are performed in sequence, and finally task completion probabilities of multiple tasks are obtained by the prediction model.
[0388] In step 2510, the error subfunction is calculated for a single task among the multiple tasks obtained by the prediction model. The error subfunction refers to the distance between the task completion probability and the task label for a single task. The closer the distance, the smaller the error and the better the prediction effect. First, determine the task completion probability of a single task; second, obtain the training label for the task in the training sample. If the training label indicates that the task is completed, then the training label is 1, otherwise, the training label is 0; finally, based on the task completion probability of a single task and the training label of the task, determine the error subfunction of the task.
[0389] The calculation process of determining the error sub-function is shown in Formula 9:
[0390] L t =-q t ·log(p t )-(1-q t)·log(1-p t ) (Formula 9).
[0391] In Formula 9, q t represents the training label (0 or 1), p t represents the task completion probability of the task. For example, if the training label of task A is 1 and the task completion probability is 0.5, then the error subfunction L of the task is t =-1·log(0.5)=0.3.
[0392] In step 2520, after determining the error subfunction of a single task, the error function of the entire training sample is determined based on the error subfunctions of all tasks. The error function refers to the gap between the task completion probability of all tasks and the corresponding training labels. The smaller the gap, the better the prediction result of the model.
[0393] In one embodiment, determining the error function of the training sample based on the error sub-function includes: summing the error sub-functions of multiple tasks to obtain the error function.
[0394] In this embodiment, the calculation process of determining the error function is shown in Formula 10:
[0395]
[0396] In Formula 10, T represents the number of tasks, L t Represents the error subfunction of a single task. For example, the training sample includes three tasks, among which the error subfunction of task A is 0.3; the error subfunction of task B is 0.45; and the error subfunction of task C is 0.27. The error function of the training sample is 0.3+0.45+0.27=1.02.
[0397] The advantage of determining the error function by summing up the error sub-functions is that the error sub-function corresponding to each task contributes equally to determining the error function of the training sample, thereby improving the fairness of multiple tasks when training the model.
[0398] In another embodiment, determining the error function of the training sample based on the error sub-function includes:
[0399] Get the task weights of multiple tasks;
[0400] Based on the task weights, the error sub-functions of multiple tasks are weighted summed to obtain the error function.
[0401] In this embodiment, the calculation process of determining the error function is shown in Formula 11:
[0402]
[0403] In Formula 11, T represents the number of tasks, W t represents the task weight of a single task, L t Represents the error subfunction of a single task. The task weights of multiple tasks can be determined according to the actual application scenario. For example, the training sample includes three tasks, the task weight of task A is 0.5, the task weight of task B is 0.3, and the task weight of task C is 0.2. The error subfunction of task A is 0.3; the error subfunction of task B is 0.45; and the error subfunction of task C is 0.27. Then, the error function of the training sample is 0.5*0.3+0.3*0.45+0.2*0.27=0.339.
[0404] The advantage of weighting the error sub-functions and determining the error function is that different weights can be set for different tasks according to the actual application scenarios, thereby improving the flexibility of the training model.
[0405] In step 2530, after the error function is determined, the first expert network, the second expert network, and the prediction model are trained based on the error function. The purpose of training is to reduce the error function, and a reduced error function represents a more accurate prediction result. Therefore, the training process is to update the parameters in the first expert network, the second expert network, and the prediction model so that the updated model will reduce the error function generated in the end.
[0406] The first expert network, the second expert network, and the prediction model can be trained based on the error function using a gradient descent method so that the parameters of the first expert network, the second expert network, and the prediction model can be updated in the direction of the gradient descent of the error function.
[0407] The advantage of the embodiment of steps 2510-2530 is that the error function of the entire sample is determined by the error sub-function of a single task, so that the prediction result of each task has an impact on the training model, thereby improving the comprehensiveness and accuracy of the training model.
[0408] Detailed diagram of the implementation of the delivery processing method of the present disclosure embodiment
[0409] Refer to the following Fig.26 , the implementation details of the delivery processing method of the embodiment of the present disclosure are explained in detail and by way of example.
[0410] In step 2610, object features of the target object and scene features of the current scene where the target object is located are obtained.
[0411] In step 2611, a filtered content set is selected from the to-be-filtered content set, and the features of each filtered content in the filtered content set are used in turn as features of the to-be-delivered content.
[0412] In step 2612, object features, scene features, and features of the content to be delivered are used as multiple delivery basic features.
[0413] In step 2613, the multiple delivery basic feature vectors are vectorized to obtain multiple first feature vectors.
[0414] In step 2614, each first feature vector is input into a first expert network corresponding to the first feature vector, wherein the first expert network includes a first shared model, a plurality of first dedicated models corresponding to a plurality of tasks, and a first gating node.
[0415] In step 2615, the first shared model and the first weights of each first dedicated model are obtained.
[0416] In step 2616, if the first gating node corresponds to the first dedicated model, then the first shared model and the first weight of the first dedicated model corresponding to the first gating node are used to perform a weighted sum on the output of the first shared model and the output of the first dedicated model corresponding to the first gating node to obtain the target association feature corresponding to the task and the first feature vector.
[0417] In step 2617, if the first gating node corresponds to the first shared model, then the output of the first shared model and the output of each first dedicated model are weighted summed using the first shared model and the first weights of each first dedicated model to obtain the first shared feature corresponding to the first feature vector.
[0418] In step 2618, a second feature vector of the target is generated based on the plurality of target-associated features corresponding to the task and the plurality of first feature vectors.
[0419] In step 2619, a first shared feature vector is generated based on the multiple first shared features corresponding to the multiple first feature vectors.
[0420] In step 2620, the second feature vector is input into a second dedicated model in a second expert network corresponding to the task.
[0421] In step 2621, the first shared feature vector is input into the second shared model in the second expert network.
[0422] In step 2622, the second weights of the second shared model and the second dedicated model corresponding to the second gating node are obtained.
[0423] In step 2623, the output of the second shared model and the output of the second dedicated model corresponding to the second gating node are weighted summed using the second weights of the second shared model and the second dedicated model corresponding to the second gating node to obtain a third feature vector corresponding to the task.
[0424] In step 2624, the third feature vectors corresponding to the multiple tasks are input into the prediction model corresponding to the multiple targets to obtain the predicted task completion probabilities of the target objects for the content to be delivered with respect to the multiple tasks.
[0425] In step 2625, the contents to be delivered are sorted based on the task completion probability, and are delivered to the target object according to the sorting.
[0426] Description of the apparatus and device of the present disclosure
[0427] It is to be understood that, although the steps in the above-mentioned flowcharts are sequentially displayed according to the characterization of arrows, these steps are not necessarily executed in sequence according to the order of arrow characterization. Unless there is a clear description in the present embodiment, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the above-mentioned flowcharts can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the steps or stages in other steps.
[0428] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to the characteristics of the task content, such as task content attribute information or attribute information sets, the permission or consent of the task content will be obtained first, and the collection, use and processing of these data will comply with relevant laws, regulations and standards. In addition, when the embodiment of the present application needs to obtain task content attribute information, it will obtain a separate permission or separate consent for the task content through a pop-up window or jump to a confirmation page. After clearly obtaining a separate permission or separate consent for the task content, the necessary task content-related data used to enable the normal operation of the embodiment of the present application will be obtained.
[0429] Fig. 27 The structure diagram of the delivery processing device 2700 provided in the embodiment of the present disclosure. The delivery processing device 2700 includes:
[0430] The first acquisition unit 2710 is used to acquire a plurality of delivery basic features, where the delivery basic features are used to predict the task completion probability of a target object with respect to a plurality of tasks for the delivered content;
[0431] A vectorization unit 2720, configured to vectorize the plurality of delivery basic features to obtain a plurality of first feature vectors;
[0432] A first input unit 2730 is used to input each first feature vector into a first expert network corresponding to the first feature vector, wherein the first expert network includes a plurality of first dedicated models corresponding to a plurality of tasks and a plurality of first gating nodes, and the first gating node generates a task-associated feature corresponding to the task and the first feature vector based on an output of the first dedicated model;
[0433] A first generating unit 2740 is used to generate a second feature vector of the task based on the multiple task-associated features corresponding to the task and the multiple first feature vectors, and input the second feature vector into a second expert network corresponding to the task to obtain a third feature vector corresponding to the task;
[0434] The second input unit 2750 is used to input the third feature vectors corresponding to the multiple tasks into the prediction models corresponding to the multiple tasks to obtain the predicted task completion probabilities of the target object for the content to be delivered with respect to the multiple tasks.
[0435] Optionally, the second expert network includes a second dedicated model corresponding to the task, and a second gating node corresponding to the task;
[0436] The first generating unit 2740 is specifically used for:
[0437] Inputting the second feature vector into a second dedicated model corresponding to the task to obtain an output of the second expert model;
[0438] A third feature vector corresponding to the task is generated based on the output of the second dedicated model through the second gating node.
[0439] Optionally, the first expert network further includes a first sharing model, and the second expert network further includes a second sharing model;
[0440] The first input unit 2730 is specifically used for: if the first gating node corresponds to a first dedicated model, the first gating node performs a weighted sum of the output of the first shared model and the output of the first dedicated model corresponding to the first gating node to obtain a task-associated feature corresponding to the task and the first feature vector; if the first gating node corresponds to the first shared model, the first gating node performs a weighted sum of the output of the first shared model and the outputs of each first dedicated model to obtain a first shared feature corresponding to the first feature vector;
[0441] The first generation unit 2740 is specifically used to: generate a first shared feature vector based on multiple first shared features corresponding to multiple first feature vectors, and input it into the second shared model; through the second gating node, perform a weighted sum on the output of the second shared model and the output of the second dedicated model corresponding to the second gating node to obtain a third feature vector corresponding to the task.
[0442] Optionally, the first input unit 2730 is specifically used for:
[0443] Based on the first feature vector, obtaining first weights of the first shared model and each first dedicated model;
[0444] Using the first shared model and the first weight of the first dedicated model corresponding to the first gating node, a weighted sum is performed on the output of the first shared model and the output of the first dedicated model corresponding to the first gating node to obtain a task-associated feature corresponding to the task and the first feature vector;
[0445] The first input unit 2730 is specifically used to: use the first shared model and the first weights of each first dedicated model to perform weighted sum on the output of the first shared model and the output of each first dedicated model to obtain a first shared feature corresponding to the first feature vector.
[0446] Optionally, the first input unit 2730 is specifically used for:
[0447] Obtaining the first shared model and the first matrices of each of the first dedicated models, wherein the number of columns of the first matrix is the total number of models of the first dedicated model and the first shared model, and the number of rows of the first matrix is the dimension of the first eigenvector;
[0448] performing a first multiplication operation on the transpose of the first matrix and the first eigenvector;
[0449] A normalized exponential process is performed on the result of the first multiplication operation to obtain a first weight.
[0450] Optionally, the first generating unit 2740 is specifically configured to:
[0451] Cascading multiple first eigenvectors to obtain a first cascaded eigenvector;
[0452] Based on the first cascade feature vector, obtaining second weights of the second shared model and the second dedicated model corresponding to the second gating node;
[0453] Using the second weights of the second shared model and the second dedicated model corresponding to the second gating node, the output of the second shared model and the output of the second dedicated model corresponding to the second gating node are weighted summed to obtain a third feature vector corresponding to the task.
[0454] Optionally, the first generating unit 2740 is specifically configured to:
[0455] Obtaining the second shared model and the second matrix of each second dedicated model, wherein the number of columns of the second matrix is the total number of models of the second dedicated model and the second shared model, and the number of rows of the second matrix is the dimension of the first cascaded feature vector;
[0456] performing a second multiplication operation on the second matrix and the transpose of the first concatenated eigenvector;
[0457] The operation of the second multiplication operation is subjected to normalized exponent processing to obtain a second weight.
[0458] Optionally, the first acquiring unit 2710 is specifically configured to:
[0459] Obtain object features of a target object;
[0460] Obtain scene features of the current scene where the target object is located;
[0461] Filter out a filtered content set from the to-be-filtered content set, and use the features of each filtered content in the filtered content set in turn as features of the to-be-delivered content;
[0462] The object features, scene features, and features of the content to be delivered are used as multiple delivery basic features.
[0463] Optionally, the first acquiring unit 2710 is specifically configured to:
[0464] Get the first label of the target object;
[0465] Obtain the second label of the content to be filtered in the content set to be filtered;
[0466] Based on the matching degree between the first tag and the second tag, a filtered content set is filtered out from the to-be-filtered content set.
[0467] Optionally, the first acquiring unit 2710 is specifically configured to:
[0468] Based on the matching degree, the matching content whose first label matches the second label is selected from the content set to be selected;
[0469] Get multiple delivery metrics for each matching content;
[0470] Based on multiple delivery indicators, filtered contents are selected from the matching contents to form a filtered content set.
[0471] Optionally, the vectorization unit 2720 is specifically used for:
[0472] If the basic feature is a character feature, the character feature is input into the embedding layer to obtain the first feature vector corresponding to the character feature;
[0473] If the basic feature is a numerical feature, the numerical feature is numerically mapped to obtain the first feature vector corresponding to the numerical feature;
[0474] If the basic feature is a variable-length sequence feature, the variable-length sequence feature is pooled to obtain a first feature vector corresponding to the variable-length sequence feature.
[0475] Optionally, the delivery processing device further includes:
[0476] A second acquisition unit (not shown), configured to acquire the attention task of the target object;
[0477] A third acquisition unit (not shown), configured to acquire a task completion probability for a focused task from among the task completion probabilities for a plurality of tasks;
[0478] A sorting unit (not shown), configured to sort the content to be delivered based on the task completion probability of the concerned task;
[0479] The delivery unit (not shown) is used to deliver the content to be delivered to the target object according to the ranking.
[0480] Optionally, the sorting unit (not shown) is specifically used to:
[0481] If there are multiple tasks of interest, the overall task completion probability is determined based on the task completion probabilities of the multiple tasks of interest;
[0482] Based on the overall task completion probability, the content to be delivered is sorted.
[0483] Optionally, the multiple delivery basic features are selected from a training sample in the training sample set, the training sample has multiple training labels for multiple tasks, and the training label indicates whether the task is completed for the training sample;
[0484] The first expert network, the second expert network, and the prediction model are jointly trained through the following process:
[0485] For a single task, determine the error subfunction of the task based on the training labels and the task completion probability of the task;
[0486] Based on the error sub-function, determining the error function of the training sample;
[0487] Based on the error function, the first expert network, the second expert network, and the prediction model are jointly trained.
[0488] Optionally, inputting each first feature vector into a first expert network corresponding to the first feature vector comprises:
[0489] Get the task weights of multiple tasks;
[0490] Based on the task weights, the error sub-functions of multiple tasks are weighted and summed to obtain the error function.
[0491] Optionally, the first input unit 2730 is specifically used for:
[0492] Inputting each first feature vector into a plurality of first expert networks corresponding to the first feature vector to obtain a plurality of candidate task-associated features corresponding to the task and the first feature vector;
[0493] Based on a predetermined criterion, a task-associated feature corresponding to the task and the first feature vector is determined according to a plurality of candidate task-associated features.
[0494] Reference Fig.28 , Fig.28 The structural block diagram of the terminal part for implementing the delivery processing method of the embodiment of the present disclosure is as follows: the terminal includes: a radio frequency (RF) circuit 2810, a memory 2815, an input unit 2830, a display unit 2840, a sensor 2850, an audio circuit 2860, a wireless fidelity (WiFi) module 2870, a processor 2880, and a power supply 2890. Those skilled in the art can understand that Fig.28 The terminal structure shown does not constitute a limitation on the mobile phone or computer, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.
[0495] The RF circuit 2810 may be used for receiving and sending signals during information transmission or calls. In particular, after receiving the downlink information from the base station, it is sent to the processor 2880 for processing; in addition, the designed uplink data is sent to the base station.
[0496] The memory 2815 may be used to store software programs and modules. The processor 2880 executes various functional applications and data processing of the content terminal by running the software programs and modules stored in the memory 2815 .
[0497] The input unit 2830 may be used to receive input digital or character information and generate key signal input related to the setting and function control of the content terminal. Specifically, the input unit 2830 may include a touch panel 2831 and other input devices 2832.
[0498] The display unit 2840 may be used to display input information or provided information and various menus of the content terminal. The display unit 2840 may include a display panel 2841.
[0499] The audio circuit 2860, the speaker 2861, and the microphone 2862 may provide an audio interface.
[0500] In this embodiment, the processor 2880 included in the terminal can execute the delivery processing method of the previous embodiment.
[0501] The terminals of the embodiments of the present disclosure include but are not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle terminals, aircraft, etc. The embodiments of the present invention can be applied to various scenarios, including but not limited to content recommendation, data screening, etc.
[0502] Fig.26 A block diagram of the structure of part of the server 130 for implementing the delivery processing method of the embodiment of the present disclosure. The server 130 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 2922 (for example, one or more processors) and memory 2932, and one or more storage media 2630 (for example, one or more mass storage devices) storing application programs 2942 or data 2944. Among them, the memory 2932 and the storage medium 2930 can be temporary storage or permanent storage. The program stored in the storage medium 2930 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the server 130. Furthermore, the central processing unit 2922 can be configured to communicate with the storage medium 2930 and execute a series of instruction operations in the storage medium 2930 on the server 130.
[0503] The server 130 may also include one or more power supplies 2926, one or more wired or wireless network interfaces 2950, one or more input and output interfaces 2958, and / or one or more operating systems 2941, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.
[0504] The central processor 2922 in the server 130 can be used to execute the delivery processing method of the embodiment of the present disclosure.
[0505] The embodiments of the present disclosure also provide a computer-readable storage medium, which is used to store program codes, and the program codes are used to execute the delivery processing methods of the aforementioned embodiments.
[0506] The embodiment of the present disclosure also provides a computer program product, which includes a computer program. A processor of a computer device reads and executes the computer program, so that the computer device performs the above-mentioned delivery process.
[0507] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present disclosure and the above-mentioned drawings are used to distinguish similar contents, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can, for example, be implemented in an order other than those illustrated or described herein. In addition, the terms "comprises" and "comprising" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0508] It should be understood that in the present disclosure, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated content, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the associated content before and after is in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.
[0509] It should be understood that in the description of the embodiments of the present disclosure, the meaning of multiple (or multiple items) is more than two, greater than, less than, exceed, etc. are understood to not include the number, and above, below, within, etc. are understood to include the number.
[0510] In the several embodiments provided in the present disclosure, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.
[0511] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0512] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0513] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present disclosure is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server 130, or network device, etc.) to perform all or part of the steps of the various embodiments of the present disclosure. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store program codes.
[0514] It should also be understood that the various implementations provided in the embodiments of the present disclosure can be combined arbitrarily to achieve different technical effects.
[0515] The above is a specific description of the implementation methods of the present disclosure, but the present disclosure is not limited to the above implementation methods. Technical personnel familiar with the art can also make various equivalent modifications or substitutions without violating the spirit of the present disclosure. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present disclosure.
Claims
1. A delivery processing method, It is characterized in that include: Acquire a plurality of delivery basic features, where the delivery basic features are used to predict task completion probabilities of a target object with respect to a plurality of tasks for the delivered content; Quantizing the plurality of delivery basic feature vectors to obtain a plurality of first feature vectors; Inputting each of the first feature vectors into a first expert network corresponding to the first feature vector, wherein the first expert network includes a plurality of first dedicated models corresponding to a plurality of the tasks and a plurality of first gating nodes, and the first gating nodes generate task-associated features corresponding to the tasks and the first feature vectors based on outputs of the first dedicated models; generating a second feature vector of the task based on a plurality of the task-associated features corresponding to the task and a plurality of the first feature vectors, and inputting the second feature vector into a second expert network corresponding to the task to obtain a third feature vector corresponding to the task; The third feature vectors corresponding to the multiple tasks are input into the prediction models corresponding to the multiple tasks to obtain the predicted task completion probabilities of the target object for the to-be-delivered content with respect to the multiple tasks.
2. The delivery processing method according to claim 1, It is characterized in that The second expert network includes a second dedicated model corresponding to the task, and a second gating node corresponding to the task; The step of inputting the second feature vector into a second expert network corresponding to the task to obtain a third feature vector corresponding to the task includes: Inputting the second feature vector into a second dedicated model corresponding to the task to obtain an output of the second expert model; The third feature vector corresponding to the task is generated by the second gating node based on the output of the second dedicated model.
3. The delivery processing method according to claim 2, It is characterized in that The first expert network further includes a first sharing model, and the second expert network further includes a second sharing model; The first gating node generates a task-associated feature corresponding to the task and the first feature vector based on the output of the first dedicated model, including: if the first gating node corresponds to one of the first dedicated models, the first gating node performs a weighted sum of the output of the first shared model and the output of the first dedicated model corresponding to the first gating node to obtain the task-associated feature corresponding to the task and the first feature vector; if the first gating node corresponds to the first shared model, the first gating node performs a weighted sum of the output of the first shared model and the outputs of each of the first dedicated models to obtain a first shared feature corresponding to the first feature vector; The generating of the third feature vector corresponding to the task based on the output of the second dedicated model through the second gating node includes: generating a first shared feature vector based on multiple first shared features corresponding to multiple first feature vectors, and inputting the first shared feature vector into the second shared model; and performing a weighted sum of the output of the second shared model and the output of the second dedicated model corresponding to the second gating node through the second gating node to obtain the third feature vector corresponding to the task.
4. The delivery processing method according to claim 3, It is characterized in that The step of performing a weighted sum operation on the output of the first shared model and the output of the first dedicated model corresponding to the first gating node to obtain the task-associated feature corresponding to the task and the first feature vector includes: Based on the first feature vector, obtaining first weights of the first shared model and each of the first dedicated models; Using the first shared model and the first weight of the first dedicated model corresponding to the first gating node, a weighted sum is performed on an output of the first shared model and an output of the first dedicated model corresponding to the first gating node to obtain the task-associated feature corresponding to the task and the first feature vector; The step of taking a weighted sum of the output of the first shared model and the output of each of the first dedicated models to obtain a first shared feature corresponding to the first feature vector includes: using the first shared model and the first weights of each of the first dedicated models to take a weighted sum of the output of the first shared model and the output of each of the first dedicated models to obtain a first shared feature corresponding to the first feature vector.
5. The delivery processing method according to claim 4, It is characterized in that The acquiring, based on the first feature vector, first weights of the first shared model and each of the first dedicated models includes: Obtaining the first shared model and each first dedicated model's first matrix, wherein the number of columns of the first matrix is the total number of models of the first dedicated model and the first shared model, and the number of rows of the first matrix is the dimension of the first eigenvector; performing a first multiplication operation on the first matrix and the transpose of the first eigenvector; A normalized exponent process is performed on a result of the first multiplication operation to obtain the first weight.
6. The delivery processing method according to claim 3, It is characterized in that The step of performing a weighted sum operation on the output of the second shared model and the output of the second dedicated model corresponding to the second gating node through the second gating node to obtain the third feature vector corresponding to the task includes: Cascading a plurality of the first feature vectors to obtain a first cascade feature vector; Based on the first cascade feature vector, obtaining second weights of the second shared model and the second dedicated model corresponding to the second gating node; Using the second weights of the second shared model and the second dedicated model corresponding to the second gating node, the output of the second shared model and the output of the second dedicated model corresponding to the second gating node are weighted summed to obtain the third feature vector corresponding to the task.
7. The delivery processing method according to claim 6, It is characterized in that The acquiring, based on the first cascade feature vector, second weights of the second shared model and the second dedicated model corresponding to the second gating node, respectively, includes: Obtaining the second shared model and the second matrix of each of the second dedicated models, wherein the number of columns of the second matrix is the total number of models of the second dedicated model and the second shared model, and the number of rows of the second matrix is the dimension of the first cascaded feature vector; performing a second multiplication operation on the second matrix and the transpose of the first concatenated eigenvector; A normalized exponent process is performed on the operation of the second multiplication operation to obtain the second weight.
8. The delivery processing method according to claim 1, It is characterized in that The obtaining of multiple basic delivery features includes: Acquire object features of the target object; Acquire scene features of the current scene where the target object is located; Filtering out a filtered content set from the to-be-filtered content set, and taking the features of each filtered content in the filtered content set as features of the to-be-delivered content in turn; The object feature, the scene feature, and the feature of the content to be delivered are used as a plurality of delivery basic features.
9. The delivery processing method according to claim 8, It is characterized in that The step of filtering out a filtered content set from the to-be-filtered content set includes: Obtaining a first label of the target object; Obtain a second tag of the content to be filtered in the set of content to be filtered; Based on the matching degree between the first tag and the second tag, a filtered content set is filtered out from the to-be-filtered content set.
10. The delivery processing method according to claim 9, It is characterized in that The filtering out a filtered content set from the to-be-filtered content set based on the matching degree between the first tag and the second tag includes: Based on the matching degree, selecting matching contents in which the first tag and the second tag match from the to-be-screened content set; Obtaining multiple delivery indicators for each of the matching contents; Based on the multiple delivery indicators, filtered contents are filtered out from the matching contents to form the filtered content set.
11. The delivery processing method according to claim 1, It is characterized in that The step of vectorizing the plurality of delivery basic features to obtain a plurality of first feature vectors includes: If the delivery basic feature is a character feature, input the character feature into an embedding layer to obtain the first feature vector corresponding to the character feature; If the delivery basic feature is a numerical feature, numerical mapping is performed on the numerical feature to obtain the first feature vector corresponding to the numerical feature; If the delivery basic feature is a variable-length sequence feature, the variable-length sequence feature is pooled to obtain the first feature vector corresponding to the variable-length sequence feature.
12. The delivery processing method according to claim 1, It is characterized in that After inputting the third feature vectors corresponding to the plurality of tasks into the prediction models corresponding to the plurality of tasks to obtain the predicted task completion probabilities of the target objects for the to-be-delivered content with respect to the plurality of tasks, the delivery processing method further includes: Obtaining the focus task of the target object; Acquire the task completion probability of the task of interest from the task completion probabilities of the plurality of tasks; sorting the to-be-delivered contents based on the task completion probabilities of the concerned tasks; The content to be delivered is delivered to the target object according to the order.
13. The delivery processing method according to claim 12, It is characterized in that The sorting of the to-be-delivered contents based on the task completion probability of the concerned task includes: If there are multiple tasks of interest, determining an overall task completion probability based on the task completion probabilities of the multiple tasks of interest; The to-be-delivered contents are sorted based on the overall task completion probability.
14. The delivery processing method according to claim 1, It is characterized in that The plurality of delivery basic features are selected from a training sample in a training sample set, the training sample having a plurality of training labels for the plurality of tasks, the training labels indicating whether the tasks are completed for the training sample; The first expert network, the second expert network, and the prediction model are jointly trained through the following process: For a single task, determining an error subfunction of the task based on the training labels and the task completion probability of the task; Based on the error sub-function, determining the error function of the training sample; Based on the error function, the first expert network, the second expert network, and the prediction model are jointly trained.
15. The delivery processing method according to claim 14, It is characterized in that The step of determining the error function of the training sample based on the error sub-function includes: Obtaining task weights of a plurality of the tasks; Based on the task weights, a weighted sum is performed on the error sub-functions of a plurality of the tasks to obtain the error function.
16. The delivery processing method according to claim 1, It is characterized in that The step of inputting each of the first feature vectors into a first expert network corresponding to the first feature vector comprises: Inputting each of the first feature vectors into a plurality of the first expert networks corresponding to the first feature vector, to obtain a plurality of candidate task-associated features corresponding to the task and the first feature vector; Based on a predetermined criterion, a task-associated feature corresponding to the task and the first feature vector is determined according to a plurality of candidate task-associated features.
17. A delivery processing device, It is characterized in that include: A first acquisition unit is used to acquire a plurality of delivery basic features, wherein the delivery basic features are used to predict task completion probabilities of a target object with respect to a plurality of tasks for the delivered content; A vectorization unit, used for vectorizing the plurality of delivery basic features to obtain a plurality of first feature vectors; a first input unit, configured to input each of the first feature vectors into a first expert network corresponding to the first feature vector, wherein the first expert network includes a plurality of first dedicated models corresponding to a plurality of the tasks, and a plurality of first gating nodes, and the first gating nodes generate task-associated features corresponding to the tasks and the first feature vectors based on outputs of the first dedicated models; A first generating unit is used to generate a second feature vector of the task based on the task and a plurality of the task-associated features corresponding to the plurality of the first feature vectors, and input the second feature vector into a second expert network corresponding to the task to obtain a third feature vector corresponding to the task; The second input unit is used to input the third feature vectors corresponding to the multiple tasks into the prediction model corresponding to the multiple tasks to obtain the predicted task completion probability of the target object for the content to be delivered with respect to the multiple tasks.
18. An electronic device comprising a memory and a processor, wherein the memory stores a computer program. It is characterized in that When the processor executes the computer program, the delivery processing method according to any one of claims 1 to 16 is implemented.
19. A computer-readable storage medium storing a computer program. It is characterized in that When the computer program is executed by a processor, the delivery processing method according to any one of claims 1 to 16 is implemented.
20. A computer program product, comprising a computer program, wherein the computer program is read and executed by a processor of a computer device, so that the computer device executes the delivery processing method according to any one of claims 1 to 16.
