Click rate prediction method and device

Abstract

The invention discloses a click rate prediction method and device, and relates to the technical field of computers. A specific embodiment of the method comprises the steps of determining attention matrixes corresponding to sequences in a video attribute information sequence, a spatial information sequence and a time information sequence of historical video click behaviors of a user, and fusing the attention matrixes corresponding to the sequences to obtain a fused attention matrix; generating a historical video behavior sequence based on the fusion attention matrix; generating target historical video behavior information associated with the to-be-predicted video based on the historical video behavior sequence and the attribute information of the to-be-predicted video; and generating the click rate of the to-be-predicted video based on the target historical video behavior information, the attribute information of the to-be-predicted video, the current time information and the current space information. According to the embodiment, the accuracy of the click rate of the generated to-be-predicted video is effectively improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a click rate prediction method and device. Background Art

[0002] The most important step in personalized e-commerce short video recommendations is to estimate the probability that users are willing to watch a specific short video. Accurate click-through rate prediction can not only provide users with content that they like and improve user experience, but also help content producers reach interested viewers to gain commercial benefits.

[0003] Existing CTR estimation methods mainly include the following two directions:

[0004] 1) Feature interaction, that is, time- and space-related features are treated equally with other features and sent to the model for feature interaction. This method cannot capture spatiotemporal information well, resulting in poor accuracy of predicted click-through rate.

[0005] 2) Behavior sequence modeling, which uses the user's historical behavior to capture the user's hidden interests. This method only considers the relative order of historical behavior and ignores the impact of spatiotemporal information on capturing user interests, resulting in poor accuracy of predicted click-through rate. Summary of the invention

[0006] Embodiments of the present application provide a click rate prediction method, apparatus, device, and storage medium.

[0007] According to the first aspect, an embodiment of the present application provides a click-through rate prediction method, which includes: determining the attention matrix corresponding to each sequence in the video attribute information sequence, spatial information sequence and time information sequence of the user's historical video click behavior, and fusing the attention matrices corresponding to each sequence to obtain a fused attention matrix, and generating a historical video behavior sequence based on the fused attention matrix; generating target historical video behavior information associated with the video to be predicted based on the historical video behavior sequence and the attribute information of the video to be predicted; generating the click-through rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, the current time information and spatial information.

[0008] According to the second aspect, an embodiment of the present application provides a click-through rate prediction device, which includes: a fusion module, an acquisition module, a generation module and a prediction module, wherein the fusion module is configured to determine the attention matrix corresponding to each sequence in the video attribute information sequence, the spatial information sequence and the time information sequence of the user's historical video click behavior, and fuse the attention matrices corresponding to each sequence to obtain a fused attention matrix; the acquisition module is configured to generate a historical video behavior sequence based on the fused attention matrix; the generation module is configured to generate target historical video behavior information associated with the video to be predicted based on the historical video behavior sequence and the attribute information of the video to be predicted; the prediction module is configured to generate the click-through rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, the current time information and the spatial information.

[0009] According to the third aspect, an embodiment of the present application provides an electronic device, which includes one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation method in the first aspect.

[0010] According to a fourth aspect, an embodiment of the present application provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any implementation manner in the first aspect.

[0011] According to a fifth aspect, an embodiment of the present disclosure provides a computer program product, including a computer program, which implements the method described in any embodiment of the first aspect when executed by a processor.

[0012] The present application determines the attention matrix corresponding to each sequence in the video attribute information sequence, spatial information sequence and time information sequence of the user's historical video click behavior, and fuses the attention matrices corresponding to each sequence to obtain a fused attention matrix. Based on the fused attention matrix, a historical video behavior sequence is generated, and the spatiotemporal information of the user's historical video click behavior is moved from the input layer to the attention layer, thereby realizing the decoupling of time information, spatial information and video attribute information, and enhancing the modeling capability of spatiotemporal information fusion. Furthermore, based on the historical video behavior sequence and the attribute information of the video to be predicted, target historical video behavior information associated with the video to be predicted is generated; based on the target historical video behavior information, the attribute information of the video to be predicted, the current time information and spatial information, the click-through rate of the video to be predicted is generated, which effectively captures the user's interest preference for videos in different time and space, thereby effectively improving the accuracy of the click-through rate of the generated video to be predicted.

[0013] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0015] Figure 2 is a flow chart of an embodiment of a click rate prediction method according to the present application;

[0016] Figure 3 is a flow chart of an application scenario of the click rate prediction method according to the present application;

[0017] Figure 4 is a flow chart of another embodiment of the click rate prediction method according to the present application;

[0018] Figure 5 is a schematic diagram of an embodiment of a click rate prediction device according to the present application;

[0019] Figure 6 It is a structural diagram of a computer system suitable for implementing a server of an embodiment of the present application. DETAILED DESCRIPTION

[0020] The following is a description of exemplary embodiments of the present application in conjunction with the accompanying drawings, including various details of the embodiments of the present application to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0021] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0022] Figure 1 An exemplary system architecture 100 is shown to which an embodiment of the click rate prediction method of the present application can be applied.

[0023] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0024] The terminal devices 101, 102, 103 interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications, such as short video applications, communication applications, etc., may be installed on the terminal devices 101, 102, 103.

[0025] Terminal devices 101, 102, 103 can be hardware or software. When terminal devices 101, 102, 103 are hardware, they can be various electronic devices with display screens, including but not limited to mobile phones and laptops. When terminal devices 101, 102, 103 are software, they can be installed in the electronic devices listed above. It can be implemented as multiple software or software modules (for example, to provide click-through rate prediction services), or it can be implemented as a single software or software module. No specific limitation is made here.

[0026] Server 105 can be a server that provides various services. For example, in an e-commerce short video recommendation scenario, for a short video to be recommended, that is, a video to be predicted, the attention matrix corresponding to each sequence in the video attribute information sequence, spatial information sequence, and time information sequence of the user's historical video click behavior can be determined, and the attention matrices corresponding to each sequence are fused to obtain a fused attention matrix. Based on the fused attention matrix, a historical video behavior sequence is generated; based on the historical video behavior sequence and the attribute information of the video to be predicted, target historical video behavior information associated with the video to be predicted is generated; based on the target historical video behavior information, the attribute information of the video to be predicted, the current time information and spatial information, the click rate of the video to be predicted is generated.

[0027] It should be noted that the server 105 can be hardware or software. When the server 105 is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or it can be implemented as a single server. When the server is software, it can be implemented as multiple software or software modules (for example, for providing click rate prediction services), or it can be implemented as a single software or software module. No specific limitation is made here.

[0028] It should be noted that the click rate prediction method provided in the embodiments of the present disclosure can be executed by the server 105, or by the terminal devices 101, 102, 103, or by the server 105 and the terminal devices 101, 102, 103 cooperating with each other. Accordingly, the various parts (such as various units, sub-units, modules, sub-modules) included in the click rate prediction device can be all set in the server 105, or all set in the terminal devices 101, 102, 103, or respectively set in the server 105 and the terminal devices 101, 102, 103.

[0029] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0030] Figure 2 The process 200 of an embodiment of the click rate prediction method that can be applied to the present application is shown. In this embodiment, the click rate prediction method includes the following steps:

[0031] Step 201, determine the attention matrix corresponding to each sequence in the video attribute information sequence, spatial information sequence and time information sequence of the user's historical video click behavior, and fuse the attention matrices corresponding to each sequence to obtain a fused attention matrix.

[0032] In this embodiment, the execution subject (such as Figure 1 The server 105 or terminal devices 101, 102, 103 shown in the figure can first use a coding network to convert each sequence in the video attribute information sequence of the user's historical video click behavior, the spatial information sequence of the user's historical video click behavior, and the time information sequence of the user's historical video click behavior into corresponding embedded vectors; and then use a network based on the attention mechanism to determine the attention matrix corresponding to the embedded vectors corresponding to each sequence.

[0033] Furthermore, the execution subject fuses the attention matrices corresponding to each sequence to obtain a fused attention matrix.

[0034] Here, the network based on the attention mechanism can be a self-attention network or a multi-head attention network, and this application does not limit this.

[0035] Among them, the video attribute information sequence of the user's historical video click behavior can be obtained by sorting the video attribute information of the user's historical clicked videos in chronological order; the time information sequence can be obtained by arranging the click time information of the user's historical clicked videos in chronological order; the space information sequence can be obtained by sorting the click space information of the user's historical clicked videos in chronological order.

[0036] Here, video attribute information may include video identification, video type, video length, video cover image vector, etc.; click spatial information may include spatial information of a single scale, or may include spatial information of multiple different scales, that is, the spatial information sequence may include spatial information subsequences of at least two different spatial scales; click time information may include time information of a single scale, or may include time information of multiple different scales, that is, the time information sequence may include time information subsequences of at least two different time scales.

[0037] In addition, it should be pointed out that for the time information sequence, the execution subject can use Time2vec, that is, the vector representation of the time series features, to encode the time information sequence into the corresponding embedded vector.

[0038] Among them, Time2vec can capture the periodic and non-periodic information of time information sequences of different scales.

[0039] Specifically, for Time2vec, given a scalar time τ, if we want to capture the time change information of the week, τ is a value from 1 to 7. The embedding of τ is defined as t2v(τ), assuming the embedding dimension is 2k+1, and the definition formula is as follows:

[0040]

[0041] t2v(τ)[i] represents the i-th element of t2v(τ), w i , is a learnable parameter, i=0 captures non-periodic information. Φ(τ) captures periodic information and is defined as follows:

[0042]

[0043] At the same time, Φ(τ) can also model the influence of the time interval between two pairs through the mechanism of kernel function, as follows:

[0044] ψ(t1-t2)=k(k1,k2)=Φ(t1)·Φ(t2)

[0045] In some optional methods, the attention matrix corresponding to each sequence in the video attribute information sequence, spatial information sequence and time information sequence of the user's historical video click behavior is determined, including: using a network based on the attention mechanism to determine the attention matrix corresponding to the video attribute information sequence of the user's historical video click behavior, each time information subsequence in at least two time information subsequences of different scales, and each spatial information subsequence in at least two spatial information subsequences of different scales.

[0046] In this implementation, the spatial information sequence includes at least two spatial information subsequences of different spatial scales, for example, a location information subsequence, a city information subsequence, a province information subsequence, etc., and the time information sequence includes at least two time information subsequences of different time scales, for example, a timestamp information subsequence, an hour information subsequence, a week information subsequence, etc.

[0047] The execution entity may adopt a network based on the attention mechanism to determine the attention matrix corresponding to the video attribute information sequence of the user's historical video click behavior, each time information subsequence in at least two time information subsequences of different scales, and each spatial information subsequence in at least two spatial information subsequences of different scales.

[0048] Specifically, the video attribute information sequence of the user's historical video click behavior is Using attention-based network to determine video attribute information sequence The corresponding attention matrix It can be expressed by the following formula:

[0049]

[0050] The time information sequence includes: timestamp time information subsequence Hour time information subsequence and the weekday time information subsequence The time-stamped temporal information subsequence is determined by the network based on the attention mechanism The corresponding attention matrix Hour time information subsequence The corresponding attention matrix and the weekday time information subsequence The corresponding attention matrix It can be expressed by the following formula:

[0051]

[0052]

[0053]

[0054] The spatial information sequence includes: position spatial information subsequence City space information subsequence and space-saving information subsequence The position spatial information subsequence is determined by the network based on the attention mechanism The corresponding attention matrix City space information subsequence The corresponding attention matrix and space-saving information subsequence The corresponding attention matrix It can be expressed by the following formula:

[0055]

[0056]

[0057]

[0058] Furthermore, the execution entity can fuse the attention matrix corresponding to the video attribute information sequence, the attention matrix corresponding to at least two spatial information subsequences of different spatial scales, and the attention matrix corresponding to at least two temporal information subsequences of different temporal scales through a fusion network to obtain a fused attention matrix.

[0059] Among them, the fusion network can be a neural network used for matrix fusion, such as FNN (Fully Connected Neural Network, feedforward neural network), RNN (Recurrent Neural Networks, recurrent neural network), etc., which is not limited in this application.

[0060] Specifically, the video attribute information sequence is fused into Timestamp time information subsequence Hour time information subsequence Week time information subsequence Position space information subsequence City space information subsequence and space-saving information subsequence The corresponding attention matrices are fused to obtain the att of the fused attention matrix i ,att i It can be expressed by the following formula:

[0061]

[0062] Furthermore, the execution entity may adopt a network based on an attention mechanism to generate a historical video behavior sequence based on a fused attention matrix.

[0063] Specifically, for a network based on the attention mechanism with only one head, the historical video behavior sequence It can be expressed by the following formula:

[0064]

[0065]

[0066] This implementation method determines the video attribute information sequence of the user's historical video click behavior, the attention matrices corresponding to each time information subsequence in at least two time information subsequences of different scales, and each spatial information subsequence in at least two spatial information subsequences of different scales by adopting a network based on the attention mechanism, and generates a historical video behavior sequence based on the fused attention matrix, thereby capturing the user's interest cycle from different scales and the user's location preference from different location granularities, improving the effectiveness of extracting spatiotemporal information for the user's historical video behavior information, and thereby effectively improving the effectiveness of the generated historical video behavior sequence.

[0067] In some optional methods, the attention matrix corresponding to each sequence in the video attribute information sequence, spatial information sequence and time information sequence of the user's historical video click behavior is determined, including: using a network based on a multi-head attention mechanism to determine the attention matrix corresponding to each sequence in the video attribute information sequence, spatial information sequence and time information sequence of the user's historical video click behavior.

[0068] In this implementation, the execution entity may first use an encoding network to convert each sequence of the video attribute information sequence of the user's historical video click behavior, the spatial information sequence of the user's historical video click behavior, and the time information sequence of the user's historical video click behavior into a corresponding embedded vector; and then use a network based on a multi-head attention mechanism to determine the attention matrix corresponding to the embedded vector corresponding to each sequence, and generate a historical video behavior sequence based on the fused attention matrix.

[0069] Specifically, a network based on the multi-head attention mechanism can have multiple heads, for example, h heads, and the fusion attention matrix corresponding to each head is att i , the output of each head is head i , head i It can be expressed by the following formula:

[0070]

[0071] Historical video behavior sequence It can be expressed by the following formula:

[0072]

[0073] This implementation method determines the attention matrix corresponding to each sequence in the video attribute information sequence, spatial information sequence and temporal information sequence of the user's historical video click behavior by adopting a network based on the multi-head attention mechanism, and generates a historical video behavior sequence based on the fused attention matrix. That is, the multi-head attention network is used to process multiple attention distributions in parallel, so that the generated historical video behavior sequence can capture key features of multiple different dimensions, effectively improving the effectiveness of generating historical video behavior sequences.

[0074] Step 202: Generate a historical video behavior sequence based on the fused attention matrix.

[0075] In this embodiment, the execution entity can directly generate a historical video behavior sequence based on the fused attention matrix, or it can first generate an initial video behavior sequence based on the fused attention matrix; and then generate a historical video behavior sequence based on the initial video behavior sequence, the time interval sequence and the spatial interval sequence. This application does not limit this.

[0076] In some optional methods, a historical video behavior sequence is generated based on a fused attention matrix, including: generating an initial video behavior sequence based on the fused attention matrix; generating a historical video behavior sequence based on the initial video behavior sequence, a time interval sequence, and a spatial interval sequence.

[0077] In this implementation, the execution entity may first use a network based on an attention mechanism to generate an initial video behavior sequence according to a fused attention matrix, and then generate a historical video behavior sequence based on the initial video behavior sequence, the time interval sequence, and the spatial interval sequence.

[0078] Among them, the time interval sequence is composed of the time interval between the time information sequence of the user's historical video click behavior and the current time information, and the space interval sequence is composed of the space information sequence of the user's historical video click behavior and the space interval between the current space information.

[0079] Here, the execution subject can directly obtain the historical video behavior sequence based on the initial video behavior sequence, the time interval sequence and the spatial interval sequence. For example, the initial video behavior sequence, the time interval sequence and the spatial interval sequence are added to obtain the historical video behavior sequence; the historical video behavior sequence can also be obtained based on the initial video behavior sequence and the corresponding first weight, the time interval sequence and the corresponding second weight, and the spatial interval sequence and the corresponding third weight. This application does not limit this.

[0080] Among them, the first weight, the second weight and the third weight can be set according to experience, actual needs and specific application scenarios.

[0081] Specifically, the initial video behavior sequence is The time interval sequence is The spaced sequence is Historical video behavior sequence It can be expressed by the following formula:

[0082]

[0083] This implementation method generates an initial video behavior sequence based on a fused attention matrix; generates a historical video behavior sequence based on the initial video behavior sequence, the time interval sequence and the spatial interval sequence, that is, extracts video interests related to the current time information and the current spatial information in the initial video behavior sequence through time difference and spatial difference to strengthen the time information and spatial information.

[0084] Step 203: Generate target historical video behavior information associated with the video to be predicted based on the historical video behavior sequence and the attribute information of the video to be predicted.

[0085] In this embodiment, after the execution subject generates the historical video behavior sequence, it can determine the target historical video behavior information associated with the video to be predicted according to the historical video behavior sequence, the attribute information of the video to be predicted and the preset historical video behavior sequence, the attribute information of the video to be predicted and the mapping relationship table of the target historical video behavior information; it can also use a network based on an attention mechanism, for example, a TVA (Target Video Attention) network, to process the historical video behavior sequence and the attribute information of the video to be predicted, and generate the target historical video behavior information associated with the video to be predicted, that is, to extract the video interest associated with the video to be predicted from the historical video behavior sequence.

[0086] Specifically, the target historical video behavior information I v It can be expressed by the following formula:

[0087]

[0088] in, is the historical video behavior sequence, a v (·) is a feedforward neural network for calculating attention weights. The activation function of the last layer is softmax, so

[0089] Step 204, based on the target historical video behavior information, the attribute information of the video to be predicted, the current time information and the space information, the click rate of the video to be predicted is generated.

[0090] In this embodiment, after obtaining the target historical video behavior information, the execution entity can generate the click rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, the current time information and the current space information, and a preset target historical video behavior information, the attribute information of the video to be predicted, the current time information and the space information and click-through rate mapping relationship table; it can also generate the click rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, the current time information and the space information, and a preset click-through rate prediction network, and the present application does not limit this.

[0091] Among them, the click-through rate prediction network is used to characterize the target historical video behavior information, the attribute information of the video to be predicted, the current time information and space information, and the corresponding relationship with the click-through rate.

[0092] Here, the click rate prediction network can be a meta-network. The meta-network is a neural network architecture that can learn how to quickly adapt to new tasks under limited training data and computing resources. The meta-network is usually configured with static network parameters.

[0093] In some optional methods, the click rate of the video to be predicted is generated based on the target historical video behavior information, the attribute information of the video to be predicted, the current time information and the space information, including: generating the click rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, the item information mounted on the video to be predicted, the user information, the current time information and the space information.

[0094] In this implementation, the execution entity can input the target historical video behavior information, the attribute information of the video to be predicted, the item information mounted on the video to be predicted, the user information, the current time information and the space information into the preset click rate prediction network to generate the click rate of the video to be predicted.

[0095] Among them, the item information of the video to be predicted, that is, the item information of the video to be predicted, may include information such as the category, brand logo, item logo, etc. of the item, and the user information may include information such as the user's gender, age, and statistical characteristics.

[0096] This implementation method generates the click-through rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, the item information mounted on the video to be predicted, the user information, the current time information and the space information. It fully considers the impact of the item information and user information mounted on the video to be predicted on the click-through rate, and further improves the accuracy of the click-through rate prediction.

[0097] Continue to see Figure 3 , Figure 3 is a flow chart of the click rate prediction method according to this embodiment.

[0098] exist Figure 3In the application scenario, the execution subject can adopt a network 301 based on the attention mechanism, for example, multiple self-attention layers (Self-attention Layer), determine the attention matrix corresponding to each sequence in the video attribute information sequence 302, the spatial information sequence 303 and the time information sequence 304 of the user's historical video click behavior, fuse the attention matrices corresponding to each sequence to obtain a fused attention matrix; based on the fused attention matrix, generate a historical video behavior sequence 305; further, the execution subject can generate target historical video behavior information 307 associated with the video to be predicted based on the historical video behavior sequence 305 and the attribute information 306 of the video to be predicted; further, the execution subject can perform splicing 312 based on the target historical video behavior information 307, the attribute information 306 of the video to be predicted, the item information 308 mounted on the video to be predicted, the user information 309, the current time information 310 and the space information 311, and input the splicing result into the target meta-network 313 to generate the click rate 314 of the video to be predicted, wherein the target meta-network is configured with a parameter patch 315 associated with the current time information and space information.

[0099] The click-through rate prediction method disclosed in the present invention determines the attention matrix corresponding to each sequence in the video attribute information sequence, spatial information sequence and time information sequence of the user's historical video click behavior, and fuses the attention matrices corresponding to each sequence to obtain a fused attention matrix. Based on the fused attention matrix, a historical video behavior sequence is generated; based on the historical video behavior sequence and the attribute information of the video to be predicted, target historical video behavior information associated with the video to be predicted is generated; based on the target historical video behavior information, the attribute information of the video to be predicted, the current time information and spatial information, the click-through rate of the video to be predicted is generated, thereby effectively improving the accuracy of the generated click-through rate.

[0100] Continue to refer Figure 4 , which shows a process 400 according to another embodiment of the click rate prediction method in the present application. The click rate prediction method includes the following steps:

[0101] Step 401, determine the attention matrix corresponding to each sequence in the video attribute information sequence, spatial information sequence and time information sequence of the user's historical video click behavior, and fuse the attention matrices corresponding to each sequence to obtain a fused attention matrix.

[0102] In this embodiment, the implementation details and technical effects of step 401 can refer to the description of step 201 and will not be repeated here.

[0103] Step 402: Generate a historical video behavior sequence based on the fused attention matrix.

[0104] In this embodiment, the implementation details and technical effects of step 402 can refer to the description of step 202 and will not be repeated here.

[0105] Step 403: Generate target historical video behavior information associated with the video to be predicted based on the historical video behavior sequence and the attribute information of the video to be predicted.

[0106] In this embodiment, the implementation details and technical effects of step 403 may refer to the description of step 203 and will not be repeated here.

[0107] Step 404, using the target meta-network, based on the target historical video behavior information, the attribute information of the video to be predicted, the current time information and the spatial information, the click rate of the video to be predicted is generated.

[0108] In this embodiment, the execution subject can obtain the click rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, the current time information, the spatial information and the target meta-network.

[0109] Specifically, the execution subject can directly input the target historical video behavior information, the attribute information of the video to be predicted, the current time information and the spatial information into the target meta-network to obtain the click rate of the video to be predicted.

[0110] The target meta-network is configured with parameter patches associated with current temporal information and spatial information, and the parameter patches may include weight parameter patches and bias parameter patches.

[0111] Specifically, the execution subject can use the current time information e t and spatial information l Generate parameter patches associated with current temporal and spatial information through the meta-network

[0112] Among them, W represents weight, b represents bias, is the weight parameter patch associated with the current temporal and spatial information, is the bias parameter patch associated with the current temporal and spatial information.

[0113] Furthermore, the attention-based network and the target meta-network can be solved by the following cross entropy loss function L ce Training results in:

[0114]

[0115] Among them, B represents the batch sample, x i represents the input feature, y i Indicates a click on a label.

[0116] In some optional embodiments, parameter patches associated with current time information and spatial information are generated in the following manner: a first parameter patch is determined based on parameter patches corresponding to each sub-time information in sub-time information of at least two different time scales; a second parameter patch is determined based on parameter patches corresponding to each sub-space information in sub-space information of at least two different spatial scales; and a parameter patch associated with the current time information and spatial information is determined based on the first parameter patch and the second parameter patch.

[0117] In this implementation, the spatial information may include subspace information of at least two different spatial scales, for example, location subspace information, city subspace information, province subspace information, etc., and the temporal information may include sub-time information of at least two different time scales, for example, timestamp sub-time information, hour sub-time information, week sub-time information, etc.

[0118] The execution entity may first determine the parameter patch corresponding to each sub-time information in the sub-time information of at least two different time scales; and then determine the first parameter patch based on the parameter patch corresponding to each sub-time information in the sub-time information of at least two different time scales.

[0119] Furthermore, the execution entity may determine a parameter patch corresponding to each subspace information in the subspace information of at least two different spatial scales; and then determine a second parameter patch based on the parameter patch corresponding to each subspace information in the subspace information of at least two different spatial scales.

[0120] Further, based on the first parameter patch and the second parameter patch, a parameter patch associated with the current time information and space information is determined.

[0121] Specifically, the execution subject may determine the first parameter patch based on the parameter patch corresponding to each sub-time information in the three different time scales (timestamp sub-time information, hour sub-time information, week sub-time information). and Based on the parameter patches corresponding to each subspace information in the three subspace information of different spatial scales (location subspace information, city subspace information, and province subspace information), the second parameter patch is determined and

[0122] Among them, t∈{hour,weekday,timestamp},l∈{POI,city_grade,province},

[0123]

[0124]

[0125] Here, k represents the kth hidden layer. If the input dimension is m and the output dimension is n, in order to save the number of parameters, is the shared parameter, f=m*n.

[0126] Furthermore, the parameter patch associated with the current temporal information and spatial information It can be expressed by the following formula:

[0127]

[0128]

[0129] in, are the static parameters of the meta-network.

[0130] This implementation method determines a first parameter patch based on a parameter patch corresponding to each sub-time information in sub-time information of at least two different time scales; determines a second parameter patch based on a parameter patch corresponding to each sub-space information in sub-space information of at least two different spatial scales; and determines a parameter patch associated with current time information and space information based on the first parameter patch and the second parameter patch, taking into account the influence of spatial information and time information of different scales on the generated parameter patch, thereby improving the accuracy of the generated parameter patch associated with the current time information and space information.

[0131] from Figure 4 It can be seen that the process 400 of the click-through rate prediction method in this embodiment reflects the use of a target meta-network to generate the click-through rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, the current time information and the space information, wherein the target meta-network is configured with a parameter patch associated with the current time information and the space information, that is, by setting the network parameters of the target meta-network, the network parameters can be dynamically changed according to the time information and the space information, so as to adaptively learn the video preferences under different time and space, thereby effectively improving the accuracy of the click-through rate of the generated video to be predicted.

[0132] Further references Figure 5 As an implementation of the methods shown in the above figures, the present application provides an embodiment of a click rate prediction device. Figure 2 , corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0133] like Figure 5 As shown, the click rate prediction device 500 of this embodiment includes: a fusion module 501, an acquisition module 502, a generation module 503 and a prediction module 504.

[0134] Among them, the fusion module 501 can be configured to determine the attention matrix corresponding to each sequence in the video attribute information sequence, spatial information sequence and time information sequence of the user's historical video click behavior, and fuse the attention matrices corresponding to each sequence to obtain a fused attention matrix.

[0135] The acquisition module 502 is configured to generate a historical video behavior sequence based on the fused attention matrix.

[0136] The generation module 503 may be configured to generate target historical video behavior information associated with the video to be predicted based on the historical video behavior sequence and the attribute information of the video to be predicted.

[0137] The prediction module 504 may be configured to generate a click rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, and the current time information and space information.

[0138] In some optional embodiments of this embodiment, the prediction module is further configured to use a target meta-network to generate a click rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, and the current time information and space information.

[0139] In some optional aspects of this embodiment, the acquisition module is further configured to generate an initial video behavior sequence based on the fused attention matrix; and to generate a historical video behavior sequence based on the initial video behavior sequence, the time interval sequence and the spatial interval sequence.

[0140] In some optional modes of this embodiment, the prediction module is further configured to generate the click rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, the item information mounted on the video to be predicted, the user information, the current time information and the space information.

[0141] In some optional embodiments of this embodiment, the fusion module is further configured to adopt a network based on the attention mechanism to determine the attention matrix corresponding to the video attribute information sequence of the user's historical video click behavior, each time information subsequence in at least two time information subsequences of different scales, and each spatial information subsequence in at least two spatial information subsequences of different scales.

[0142] In some optional methods of this embodiment, parameter patches associated with current time information and space information are generated in the following manner: a first parameter patch is determined based on parameter patches corresponding to each sub-time information in sub-time information of at least two different time scales; a second parameter patch is determined based on parameter patches corresponding to each sub-space information in sub-space information of at least two different spatial scales; and a parameter patch associated with the current time information and space information is determined based on the first parameter patch and the second parameter patch.

[0143] In some optional embodiments of this embodiment, the fusion module is further configured to adopt a network based on a multi-head attention mechanism to determine the attention matrix corresponding to each sequence in the video attribute information sequence, spatial information sequence and temporal information sequence of the user's historical video click behavior.

[0144] It should be noted that the collection, collection, updating, analysis, processing, use, transmission, storage and other aspects of user personal information involved in the technical solution of this disclosure are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information, network security and national security.

[0145] According to an embodiment of the present application, the present application also provides an electronic device, a readable storage medium and a computer program product.

[0146] like Figure 6 , which is a block diagram of an electronic device according to the click rate prediction method of an embodiment of the present application.

[0147] 600 is a block diagram of an electronic device according to a click-through rate prediction method according to an embodiment of the present application. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described and / or required herein.

[0148] like Figure 6As shown, the electronic device includes: one or more processors 601, a memory 602, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components are interconnected using different buses and can be installed on a common mainboard or installed in other ways as needed. The processor can process instructions executed in the electronic device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In other embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple electronic devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 6 A processor 601 is taken as an example.

[0149] The memory 602 is a non-transitory computer-readable storage medium provided in the present application. The memory stores instructions executable by at least one processor to enable the at least one processor to perform the click-through rate prediction method provided in the present application. The non-transitory computer-readable storage medium of the present application stores computer instructions, which are used to enable a computer to perform the click-through rate prediction method provided in the present application.

[0150] The memory 602 is a non-transient computer-readable storage medium that can be used to store non-transient software programs, non-transient computer executable programs and modules, such as the program instructions / modules corresponding to the click rate prediction method in the embodiment of the present application (for example, the attached Figure 5 The processor 601 executes various functional applications and data processing of the server by running the non-transient software programs, instructions and modules stored in the memory 602, that is, the click rate prediction method in the above method embodiment is implemented.

[0151] The memory 602 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and applications required for at least one function; the data storage area may store data created by the use of the click rate prediction electronic device, etc. In addition, the memory 602 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory 602 may optionally include a memory remotely arranged relative to the processor 601, and these remote memories may be connected to the click rate prediction electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0152] The electronic device of the click rate prediction method may further include: an input device 603 and an output device 604. The processor 601, the memory 602, the input device 603 and the output device 604 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0153] The input device 603 can receive input digital or character information, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a track ball, a joystick, and other input devices. The output device 604 may include a display device, an auxiliary lighting device (e.g., an LED), and a tactile feedback device (e.g., a vibration motor). The display device may include, but is not limited to, a liquid crystal display (LCD), a light emitting diode (LED) display, and a plasma display. In some embodiments, the display device may be a touch screen.

[0154] Various implementations of the systems and techniques described herein can be realized in digital electronic circuit systems, integrated circuit systems, dedicated ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0155] These computer programs (also referred to as programs, software, software applications, or code) include machine instructions for programmable processors and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or means (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0156] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0157] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0158] A computer system may include clients and servers. Clients and servers are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship to each other.

[0159] The technical solution according to the embodiment of the present application effectively improves the accuracy of the click rate of the generated video to be predicted.

[0160] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution disclosed in this application can be achieved, and this document is not limited here.

[0161] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. A click rate prediction method, the method comprising: Determine the attention matrix corresponding to each sequence in the video attribute information sequence, the spatial information sequence, and the temporal information sequence of the user's historical video click behavior, and fuse the attention matrices corresponding to the sequences to obtain a fused attention matrix; Based on the fused attention matrix, generating a historical video behavior sequence; Based on the historical video behavior sequence and the attribute information of the video to be predicted, generating target historical video behavior information associated with the video to be predicted; Based on the target historical video behavior information, the attribute information of the video to be predicted, the current time information and the space information, the click rate of the video to be predicted is generated.

2. The method according to claim 1, wherein: The step of generating the click rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, and the current time information and space information includes: A target meta-network is adopted to generate a click rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, and the current time information and space information. The target meta-network is configured with a parameter patch associated with the current time information and space information.

3. The method according to claim 1, wherein: The generating a historical video behavior sequence based on the fused attention matrix includes: Based on the fused attention matrix, generating an initial video behavior sequence; Based on the initial video behavior sequence, the time interval sequence and the spatial interval sequence, a historical video behavior sequence is generated, wherein the time interval sequence is composed of the time information sequence of the user's historical video click behavior and the time interval of the current time information, and the spatial interval sequence is composed of the spatial information sequence of the user's historical video click behavior and the spatial interval of the current spatial information.

4. The method according to claim 1, wherein: The step of generating the click rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, and the current time information and space information includes: The click rate of the video to be predicted is generated based on the target historical video behavior information, the attribute information of the video to be predicted, the item information mounted on the video to be predicted, the user information, the current time information and the space information.

5. The method according to claim 1, wherein: The spatial information sequence includes at least two spatial information subsequences of different spatial scales, the temporal information sequence includes at least two temporal information subsequences of different temporal scales, and the attention matrix corresponding to each of the video attribute information sequence, the spatial information sequence and the temporal information sequence for determining the user's historical video click behavior includes: A network based on the attention mechanism is adopted to determine the attention matrices corresponding to the video attribute information sequence of the user's historical video click behavior, each time information subsequence in at least two time information subsequences of different scales, and each spatial information subsequence in at least two spatial information subsequences of different scales.

6. The method according to claim 2, wherein: The time information includes at least two sub-time information of different time scales; the space information includes at least two sub-space information of different space scales, and the parameter patch associated with the current time information and space information is generated by: Determining a first parameter patch based on parameter patches corresponding to each of the sub-time information in the at least two sub-time information of different time scales; Determining a second parameter patch based on parameter patches corresponding to each of the subspace information of at least two different spatial scales; Based on the first parameter patch and the second parameter patch, a parameter patch associated with current time information and spatial information is determined.

7. The method according to any one of claims 1 to 6, wherein: The attention matrix corresponding to each sequence in the video attribute information sequence, the spatial information sequence and the temporal information sequence for determining the user's historical video click behavior includes: A network based on a multi-head attention mechanism is used to determine the attention matrix corresponding to each sequence in the video attribute information sequence, spatial information sequence and temporal information sequence of the user's historical video click behavior.

8. A click rate prediction device, comprising: A fusion module is configured to determine the attention matrix corresponding to each of the video attribute information sequence, the spatial information sequence and the temporal information sequence of the user's historical video click behavior, and fuse the attention matrices corresponding to the sequences to obtain a fused attention matrix; The acquisition module is configured to generate a historical video behavior sequence based on the fused attention matrix; A generating module configured to generate target historical video behavior information associated with the video to be predicted based on the historical video behavior sequence and the attribute information of the video to be predicted; The prediction module is configured to generate a click rate of the video to be predicted based on the target historical video behavior information, the attribute information of the video to be predicted, and the current time information and space information.

9. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor so that the at least one processor can perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images