Video popularity prediction method, device, equipment and medium

Abstract

The embodiments of the present disclosure disclose a method, apparatus, device and medium for predicting video popularity. The method includes: obtaining target metadata and current playback data corresponding to the video to be predicted; performing feature extraction on the target metadata to determine content feature information, and performing feature extraction on the current playback data to determine timing feature information; based on the release time of the video to be predicted, determining the current release duration corresponding to the video to be predicted at the current prediction time; inputting the content feature information, timing feature information and current release duration into a preset feature fusion model for information fusion, and determining the fused target video feature information; inputting the target video feature information into a preset popularity prediction model for popularity prediction, and determining the popularity prediction result corresponding to the video to be predicted. The technical solution of the embodiments of the present disclosure can improve the accuracy of video popularity prediction.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to Internet technologies, and in particular to a method, apparatus, device, and medium for predicting video popularity. Background Art

[0002] With the rapid development of mobile internet, more and more users enjoy watching videos on video platforms. Tens of millions of videos are produced and shared on these platforms every day. Predicting video popularity can be a powerful way to manage this massive amount of video content. Currently, there is an urgent need for an effective method to predict video popularity and improve its accuracy. Summary of the Invention

[0003] The embodiments of the present disclosure provide a video popularity prediction method, apparatus, device, and medium to improve the accuracy of video popularity prediction.

[0004] In a first aspect, an embodiment of the present disclosure provides a video popularity prediction method, comprising:

[0005] Get the target metadata and current playback data corresponding to the video to be predicted;

[0006] Extracting features from the target metadata to determine content feature information, and extracting features from the currently played data to determine timing feature information;

[0007] Determining the current release duration of the video to be predicted at the current prediction time based on the release time of the video to be predicted;

[0008] Inputting the content feature information, the timing feature information and the current release duration into a preset feature fusion model for information fusion, and determining the fused target video feature information;

[0009] The target video feature information is input into a preset popularity prediction model to perform popularity prediction, and a popularity prediction result corresponding to the video to be predicted is determined.

[0010] In a second aspect, the embodiments of the present disclosure further provide a video popularity prediction device, comprising:

[0011] The data acquisition module is used to obtain the target metadata and current playback data corresponding to the video to be predicted;

[0012] A feature extraction module is used to extract features from the target metadata to determine content feature information, and to extract features from the currently played data to determine timing feature information;

[0013] A current release duration determination module, configured to determine, based on the release time of the video to be predicted, the current release duration corresponding to the video to be predicted at the current prediction time;

[0014] A feature fusion module is used to input the content feature information, the timing feature information and the current release duration into a preset feature fusion model to perform information fusion and determine the fused target video feature information;

[0015] The popularity prediction module is used to input the target video feature information into a preset popularity prediction model to perform popularity prediction and determine the popularity prediction result corresponding to the video to be predicted.

[0016] In a third aspect, an embodiment of the present disclosure further provides an electronic device, the electronic device comprising:

[0017] one or more processors;

[0018] a memory for storing one or more programs;

[0019] When the one or more programs are executed by the one or more processors, the one or more processors implement the video popularity prediction method provided by any embodiment of the present disclosure.

[0020] In a fourth aspect, an embodiment of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the video popularity prediction method provided in any embodiment of the present disclosure.

[0021] The disclosed embodiment extracts features from the target metadata and current playback data corresponding to the video to be predicted, determines content feature information and timing feature information, and based on the release time of the video to be predicted, determines the current release duration corresponding to the video to be predicted at the current prediction time, and inputs the content feature information, timing feature information and current release duration into a preset feature fusion model, so that the preset feature fusion model can dynamically consider the impact of the content feature information and timing feature information of the video to be predicted on the popularity of the video based on the current release duration, realizes dynamic fusion of information, and makes the fused target video feature information more fully characterize the popularity of the video, and then inputs the target video feature information into the preset popularity prediction model for popularity prediction, so as to obtain more accurate popularity prediction results and improve the accuracy of video popularity prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that the originals and elements are not necessarily drawn to scale.

[0023] Figure 1 This is a flow chart of a video popularity prediction method provided by the first embodiment of the present disclosure;

[0024] Figure 2 This is a flow chart of a video popularity prediction method provided by the second embodiment of the present disclosure;

[0025] Figure 3 This is an example of a knowledge graph involved in the second embodiment of the present disclosure;

[0026] Figure 4 This is an extraction example of a preset content feature extraction model involved in the second embodiment of the present disclosure;

[0027] Figure 5 This is an extraction example of a preset time series feature extraction model involved in the second embodiment of the present disclosure;

[0028] Figure 6 This is a flow chart of a video popularity prediction method provided by the third embodiment of the present disclosure;

[0029] Figure 7 This is an information fusion example of a preset feature fusion model involved in the third embodiment of the present disclosure;

[0030] Figure 8 This is a schematic diagram of the structure of a video popularity prediction device provided by the fourth embodiment of the present disclosure;

[0031] Figure 9 This is a structural diagram of an electronic device provided in Example 5 of the present disclosure. DETAILED DESCRIPTION

[0032] The following describes embodiments of the present disclosure in more detail with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0033] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0034] As used herein, the term "including" and its variations are open-ended, i.e., "including but not limited to." The term "based on" means "based, at least in part, on." The term "one embodiment" means "at least one embodiment," the term "another embodiment" means "at least one additional embodiment," and the term "some embodiments" means "at least some embodiments." Other terms are defined in the following description.

[0035] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0036] It should be noted that the modifications of "one" and "multiple" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, they should be understood as "one or more".

[0037] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0038] Example 1

[0039] Figure 1 This is a flow chart of a video popularity prediction method provided in the first embodiment of the present disclosure. This embodiment is applicable to the case of predicting the popularity of videos released by users. The method can be executed by a video popularity prediction device, which can be implemented in software and / or hardware and integrated into an electronic device. Figure 1 As shown, the method specifically includes the following steps:

[0040] S110: Obtain target metadata and current playback data corresponding to the video to be predicted.

[0041] Among them, the video to be predicted may refer to any video posted by the user on the video platform. For example, the video to be predicted may refer to an original video contributed by the user. The target metadata may be static data used to describe the video to be predicted itself. For example, the target metadata may include but is not limited to: at least one of the video keywords, video length, video release time, video category information and video author information corresponding to the video to be predicted. The video author information may include but is not limited to: at least one of the number of fans, the total number of videos released, the number of days the account was created, the number of views of released videos, the number of likes of released videos and the number of comments on released videos. The current playback data may refer to the dynamic playback data of the video to be predicted at the current prediction moment. For example, the current playback data may include the video playback volume per hour in the preset historical time period before the current prediction moment. For example, the current playback data may be: the playback volume sequence of the video to be predicted i in the past n hours

[0042] S120 , extracting features from the target metadata to determine content feature information, and extracting features from the currently played data to determine timing feature information.

[0043] Among them, both content feature information and time series feature information can be represented in the form of vectors.

[0044] Specifically, embodiments of the present disclosure can utilize a preset content feature extraction model to extract features from target metadata, obtaining content feature information that can fully and effectively characterize the target metadata, such as a content feature vector. Embodiments of the present disclosure can utilize a preset timing feature extraction model to extract features from current playback data, obtaining timing feature information that can fully and effectively characterize the current playback data, such as a timing feature vector. The preset timing feature extraction model can be, but is not limited to, an LSTM (Long Short-Term Memory) long short-term memory network model.

[0045] S130: Based on the release time of the video to be predicted, determine the current release duration corresponding to the video to be predicted at the current prediction time.

[0046] The current prediction time may refer to the time when the popularity of the video to be predicted needs to be predicted after the video to be predicted is released, that is, the triggering time of the prediction operation. The embodiment of the present disclosure can make predictions at any time after the video is released. The current release duration may refer to the time interval between the release time of the video to be predicted and the current prediction time, which can be used to characterize the age of the video to be predicted. Specifically, the current prediction time t predicted and the release time t of the video to be predicted publish The difference between them is taken as the current release time Δt, that is, Δt=t predicted -tpublish .

[0047] S140: Input the content feature information, the timing feature information, and the current release duration into a preset feature fusion model for information fusion, and determine the fused target video feature information.

[0048] The preset feature fusion model can be a pre-set model for dynamically fusing the current release duration with content feature information and timing feature information. The preset feature fusion model used in the disclosed embodiments is a model pre-trained based on sample data. The target video feature information can be a joint representation of content features and timing features, which can also be represented in the form of a vector.

[0049] It's important to note that the impact of content and timing features on video popularity changes dynamically throughout a video's lifecycle. For example, in the early stages of a video's release, the primary factor determining its future popularity is its content features. However, as a video ages, its timing features—i.e., its most recent playback information—have an increasing influence on its future popularity, while the influence of its content features decreases. Therefore, when predicting video popularity, we need to dynamically consider the impact of both content and timing features on video popularity to improve the accuracy of popularity predictions.

[0050] Specifically, the content feature information, timing feature information, and current release time corresponding to the video to be predicted can be input into a trained preset feature fusion model for dynamic information fusion. The preset feature fusion model can dynamically fuse both the content feature information and the timing feature information with the current release time, and then splice or overlay the content feature information and timing feature information obtained after dynamic fusion to obtain jointly represented target video feature information, namely, a target video feature vector. The preset feature fusion model can dynamically control the size of the information valve based on the current release time, so as to dynamically consider the impact of the content feature information and timing feature information of the video to be predicted on the video popularity, thereby achieving dynamic information fusion. For example, if the current release time is longer, it is determined that the content feature information has a smaller impact on the video popularity and the timing feature information has a greater impact on the video popularity. As a result, the fused target video feature information contains less content feature information and more timing feature information, thereby enabling the obtained target video feature information to more fully represent the video popularity.

[0051] S150: Input the target video feature information into a preset popularity prediction model to perform popularity prediction, and determine a popularity prediction result corresponding to the video to be predicted.

[0052] Among them, the preset popularity prediction model can be a pre-set model for predicting the popularity of the video after it is released in the future. For example, the preset popularity prediction model can be a simple single-layer feedforward neural network model. The preset popularity prediction model used in the embodiment of the present disclosure is a model that has been pre-trained based on sample data. The popularity prediction result can refer to the total number of views of the video to be predicted within a preset future time after the current prediction moment, that is, the total number of views. For example, the popularity prediction result can refer to the total number of views of the video to be predicted i in the next m hours. The popularity prediction result can also include hot videos or cold videos. Among them, a hot video refers to a video whose predicted total number of views in the future is greater than or equal to a preset threshold. A cold video can refer to a video whose predicted total number of views in the future is less than a preset threshold.

[0053] Specifically, by inputting the dynamically fused target video feature information into a preset popularity prediction model for popularity prediction, and based on the output of the preset popularity prediction model, a more accurate popularity prediction result can be obtained, thereby improving the accuracy of video popularity prediction. For example, the preset popularity prediction model can be characterized as: Among them, ReLU is the activation function, F is the target video feature information, W F and b F are all trained model parameters. The total number of views predicted by the model.

[0054] The technical solution of the embodiment of the present disclosure performs feature extraction on the target metadata and current playback data corresponding to the video to be predicted respectively, determines the content feature information and timing feature information, and determines the current release duration corresponding to the video to be predicted at the current prediction time based on the release time of the video to be predicted, and inputs the content feature information, timing feature information and current release duration into a preset feature fusion model, so that the preset feature fusion model can dynamically consider the influence of the content feature information and timing feature information of the video to be predicted on the popularity of the video based on the current release duration, realizes dynamic fusion of information, and makes the fused target video feature information more fully characterize the popularity of the video, and then inputs the target video feature information into the preset popularity prediction model for popularity prediction, so as to obtain more accurate popularity prediction results and improve the accuracy of video popularity prediction.

[0055] Example 2

[0056] Figure 2This is a flowchart of a video popularity prediction method provided in Example 2 of the present disclosure. This example, based on the previous example, optimizes the steps of "extracting features from target metadata to determine content feature information" and "extracting features from currently played data to determine timing feature information." Explanations of terms that are identical or corresponding to those in the previous examples are omitted here.

[0057] See also Figure 2 The video popularity prediction method provided in this embodiment specifically includes the following steps:

[0058] S210: Obtain target metadata and current playback data corresponding to the video to be predicted.

[0059] S220: Obtain a knowledge graph pre-constructed based on metadata corresponding to each video.

[0060] Specifically, the embodiments of the present disclosure can discretize continuous metadata in advance, such as video length, number of fans, etc., and construct a knowledge graph based on the discretized metadata corresponding to each video. Figure 3 An example of a knowledge graph is given. Figure 3 As shown in the figure, the knowledge graph can be used to uniformly represent all metadata corresponding to all videos and the relationship between metadata, thereby connecting the metadata of the videos to form a unified knowledge representation, so as to improve the effect of subsequent content feature extraction.

[0061] S230. Based on the target metadata, determine the target subgraph of the video to be predicted in the knowledge graph.

[0062] Specifically, based on the target metadata corresponding to the video to be predicted i, the first-order neighbors of the video to be predicted i in the knowledge graph are obtained, and based on the first-order neighbors, the knowledge graph is determined. The target subgraph in That is, the subgraph composed of all first-order neighbors.

[0063] S240: Input the target subgraph into a preset content feature extraction model to perform feature extraction and determine content feature information.

[0064] The preset content feature extraction model may be a pre-set model for extracting content features from a target subgraph to obtain content feature information. The preset content feature extraction model in this embodiment is pre-trained based on sample data. For example, the preset content feature extraction model may include: a structure transformation submodel and an information aggregation submodel. The structure transformation submodel may be a model for converting a graph structure into a tree-like result. The information aggregation submodel may be a model for propagating and aggregating information in a tree-like structure to obtain the final extracted content feature information.

[0065] Specifically, the target subgraph can be The input is fed into the preset content feature extraction model for feature extraction, and based on the output of the preset content feature extraction model, the extracted content feature information h can be obtained. E .

[0066] Exemplarily, S240 may include: inputting the target subgraph into the structural transformation submodel for structural transformation to determine a target tree with a tree structure; inputting the target tree into the information aggregation submodel, propagating and aggregating the node information of the target tree from bottom to top, and determining content feature information based on the aggregated root node information.

[0067] Specifically, Figure 4 An extraction example of a preset content feature extraction model is given. Figure 4 As shown, the target subgraph Input into the structural transformation sub-model, the structural transformation sub-model can transform the target sub-graph The vertices and edges in the tree are converted into corresponding nodes, and the neighbors under the same relationship are used as nodes on the same layer, thus generating a target tree with a tree structure. The target tree Input into the information aggregation sub-model, the information aggregation sub-model can propagate and aggregate the node information of the target tree layer by layer to the root node of the target tree (i.e. Figure 4 Node e1 in the root node), obtain the root node information corresponding to the root node, and use the root node information as the extracted content feature information h E .

[0068] Exemplarily, bottom-up propagation and aggregation of node information of the target tree may include: starting from the bottom layer of the target tree, taking each parent node in each layer as the current parent node; determining the child node attention weight corresponding to each current child node based on the current parent node information and each current child node information corresponding to the current parent node; based on the child node attention weight, propagating each current child node information upward to the current parent node, and determining the target propagation information corresponding to the current parent node; aggregating the current parent node information with the target propagation information, and determining the target node information corresponding to the aggregated current parent node.

[0069] Specifically, the information aggregation sub-model can be an information aggregation sub-model with a self-attention mechanism, so that the importance of each node can be known based on the attention weight, thereby improving the interpretability of the model. The information aggregation sub-model can propagate the information of each child node corresponding to each parent node in each layer to the corresponding parent node layer by layer in a bottom-up propagation direction, and aggregate the information with the parent node information until it is aggregated to the top root node. The embodiment of the present disclosure can use each parent node in each layer as the current parent node to perform information propagation and aggregation operations. For example, the current parent node e k Each corresponding current child node e j The child node attention weight π(e k ,e j ) can be determined based on the following:

[0070] π(e k ,e j )=LeakyReLU(W1(W2e k ||W3e j ))

[0071] Among them, e j Can refer to the current parent node e k A current child node of . W, W2 and W3 are all model parameters obtained after training. It should be noted that due to e k and e j are different types of nodes, namely, one is an entity node and the other is a relationship node, so k and e j Two different linear transformation matrices are introduced to improve the accuracy of attention processing. This embodiment can also use the softmax function to calculate the current parent node e k The corresponding child node attention weights of all current child nodes are normalized, that is, the child node attention weights obtained after normalization are: in, Represents the current set of child nodes.

[0072] After determining the child node attention weight corresponding to each current child node, the child node attention weight can be used to control the amount of information propagated by each current child node to the current parent node. Specifically, the weighted addition of each current child node information and the corresponding child node attention weight can be performed, and the result of the addition can be used as the target propagation information corresponding to the current parent node. Right now By aggregating the information of the current parent node itself and the target propagation information transmitted, the target node information corresponding to the current parent node is obtained, and the current parent node information is updated to the target node information. This embodiment can use a double interaction method to implement the aggregation function. The aggregation function It can be expressed as follows:

[0073]

[0074] Where g is a nonlinear function, such as the LeakyReLU function. ⊙ is the dot product, which is the bitwise multiplication of vectors. W4 and W5 are model parameters obtained after training.

[0075] S250: Input the current playback data into a preset temporal feature extraction model with an attention mechanism for feature extraction to determine temporal feature information.

[0076] Specifically, Figure 5 An extraction example of a preset time series feature extraction model is given, such as Figure 5 As shown, the preset temporal feature extraction model with attention mechanism can be an LSTM model with attention mechanism. Input into the LSTM model, through its gate mechanism, including memory gate, input gate and forget gate, LSTM can remember what should be remembered and forget what should be forgotten, thereby obtaining the hidden vector sequence generated by LSTM And use the attention mechanism to select important vectors in the hidden vector sequence to learn more informative context representations. For example, the i-th hidden vector in the hidden vector sequence H The attention weight And the output time series feature information h C This can be determined as follows:

[0077]

[0078]

[0079]

[0080] Among them, q T 、V c 、v c The disclosed embodiment extracts temporal features by utilizing a preset temporal feature extraction model with an attention mechanism, thereby being able to determine the importance of each playback information based on the attention weight, thereby further improving the interpretability of the model.

[0081] S260: Based on the release time of the video to be predicted, determine the current release duration corresponding to the video to be predicted at the current prediction time.

[0082] S270: Input the content feature information, the timing feature information, and the current release duration into a preset feature fusion model for information fusion, and determine the fused target video feature information.

[0083] S280: Input the target video feature information into a preset popularity prediction model to perform popularity prediction, and determine a popularity prediction result corresponding to the video to be predicted.

[0084] The technical solution of this embodiment connects video metadata based on a knowledge graph to form a unified knowledge representation, thereby generating stable and robust content feature information and further improving the accuracy of popularity prediction. By extracting temporal features within a pre-set temporal feature extraction model with an attention mechanism, the importance of each playback information can be determined based on attention weights, further improving the interpretability of the model.

[0085] Example 3

[0086] Figure 6 This is a flowchart of a video popularity prediction method provided in Example 3 of the present disclosure. Building on the previous examples, this example predefines a feature fusion model that includes an influence weight determination submodel and a feature fusion submodel. Furthermore, the information fusion process is described in detail. Explanations of terms that are identical or corresponding to those in the previous examples are omitted here.

[0087] See also Figure 6 The video popularity prediction method provided in this embodiment specifically includes the following steps:

[0088] S610: Obtain target metadata and current playback data corresponding to the video to be predicted.

[0089] S620: Extract features from the target metadata to determine content feature information, and extract features from the currently played data to determine timing feature information.

[0090] S630: Based on the release time of the video to be predicted, determine the current release duration corresponding to the video to be predicted at the current prediction time.

[0091] S640: Input the current publishing time into the influence weight determination sub-model to determine the first influence weight corresponding to the content feature information and the second influence weight corresponding to the timing feature information.

[0092] Among them, the influence weight determination sub-model can be a model for dynamically determining the influence of content features and timing features on popularity. The first influence weight can be used to characterize the influence of content features on video popularity at the current prediction moment, that is, the importance of content feature information. The second influence weight can be used to characterize the influence of timing features on video popularity at the current prediction moment, that is, the importance of timing feature information. For example, as the length of video release increases, the importance of content features gradually decreases, that is, the first influence becomes smaller and smaller, while the importance of timing features gradually increases, that is, the second influence weight becomes larger and larger.

[0093] Specifically, Figure 7 An example of information fusion using a preset feature fusion model is given. Figure 7 As shown, the current publishing time is input into the influence weight determination sub-model to determine the influence, and according to the output of the influence weight determination sub-model, the first influence weight corresponding to the content feature information and the second influence weight corresponding to the timing feature information are obtained.

[0094] Exemplarily, the influence weight determination sub-model can be an information valve based on a learnable preset decay function to control the inflow of different information at different times. For example, the influence weight determination sub-model in this embodiment can include a preset decay function obtained after training. The preset decay function can be used to simulate the decay of the importance of information over time, and it can refer to a function of any decay form. For example, the preset decay function can be, but is not limited to, an exponential decay function. Exponential decay function It can be expressed as follows:

[0095]

[0096] Among them, Δt is the current release time, θ, and These are all trainable model parameters, and the parameter θ can be used to control the attenuation function The decay rate.

[0097] Exemplarily, step S640 may include: determining a first influence weight corresponding to the content feature information based on a preset attenuation function and the current publishing duration; and determining a second influence weight corresponding to the timing feature information based on the first influence weight.

[0098] Specifically, the influence weight determination sub-model can be based on a preset attenuation function And the current release time Δt input to determine the current attenuation value The current attenuation value can be used as the first influence weight corresponding to the content feature information. The sum of the first influence weight and the second influence weight is 1, so the difference between 1 and the first influence weight can be used as the second influence weight corresponding to the temporal feature information. This allows the dynamic variability of the influence of content features and temporal features to be simulated based on a preset time function.

[0099] S650. Input the content feature information, timing feature information, the first influence weight and the second influence weight into the feature fusion sub-model, perform weighted processing on the content feature information and the timing feature information based on the first influence weight and the second influence weight respectively, and fuse the target content feature information and the target timing feature information obtained after processing to determine the fused target video feature information.

[0100] Specifically, the content feature information, the temporal feature information, the first influence weight and the second influence weight are input into the feature fusion sub-model (such as Figure 7 The feature fusion sub-model performs dynamic fusion on the input content feature information based on the input first impact weight to obtain the processed target content feature information. For example, the target content feature information The feature fusion sub-model performs weighted processing on the input time series feature information based on the input second influence weight to obtain the processed target time series feature information. For example, the target time series feature information The embodiment of the present disclosure may perform splicing or bitwise addition processing on the obtained target content feature information and target timing feature information to obtain final target video feature information.

[0101] Exemplarily, the feature fusion sub-model may be a feature fusion sub-model with an attention mechanism. The step of "performing a fusion process on the target content feature information and the target temporal feature information obtained after processing to determine the fused target video feature information" in S650 may include: performing attention processing on the target content feature information and the target temporal feature information obtained after processing to determine the content attention weight corresponding to the target content feature information and the temporal attention weight corresponding to the target temporal feature information; and determining the target video feature information based on the target content feature information, the target temporal feature information, the content attention weight, and the temporal attention weight.

[0102] Specifically, in addition to considering the impact of release time on popularity prediction, the content interaction between different features can also be considered to further improve the accuracy and interpretability of the prediction results. The embodiment of the present disclosure can use the attention mechanism to process the internal interaction between content features and time series features. Figure 7 As shown, the feature fusion sub-model obtains the target content feature information and target temporal feature information Afterwards, the target content feature information can be determined based on the attention mechanism Attention The corresponding content attention weight α E and target temporal feature information The corresponding temporal attention weight α C . It can be based on the content attention weight α E and temporal attention weight α C , respectively for the target content feature information and target temporal feature information Perform weighted processing and add the processed results bit by bit to obtain the final video joint representation, that is, the target video feature information

[0103] Exemplarily, performing attention processing on the target content feature information and the target temporal feature information obtained after processing to determine the content attention weight corresponding to the target content feature information and the temporal attention weight corresponding to the target temporal feature information may include:

[0104] The target content feature information and the target temporal feature information are concatenated and linearly transformed to obtain global feature information; the target content feature information and the global feature information are dot-product processed to determine the content attention weight corresponding to the target content feature information, and the target temporal feature information and the global feature information are dot-product processed to determine the temporal attention weight corresponding to the target temporal feature information.

[0105] Specifically, the learnable linear transformation matrix W can be g Applied to target content feature information and target temporal feature information Feature information obtained after splicing On the other hand, we can obtain global feature information, such as a global feature vector r, that is, The target content feature information can be Perform dot product processing with the global feature information r to obtain the content attention weight, that is, The target time series characteristic information can be Perform dot product processing with the global feature information r to obtain the temporal attention weight, that is, The embodiment of the present disclosure can normalize the obtained content attention weight and temporal attention weight to obtain the normalized content attention weight and temporal attention weight, that is,

[0106] It should be noted that by utilizing a feature fusion sub-model with an attention mechanism for feature fusion, the importance of content features and temporal features can be determined based on content attention weights and temporal attention weights, further improving the model's interpretability. For example, when a popularity prediction results in a hot video, the reason why the video was predicted as a hot video can be determined based on the weight. This sophisticated learning process makes the model interpretable, enabling video platform managers to make better strategic decisions and better manage platform resources.

[0107] S660: Input the target video feature information into a preset popularity prediction model to perform popularity prediction, and determine a popularity prediction result corresponding to the video to be predicted.

[0108] The technical solution of this embodiment uses the influence weight determination sub-model to determine the first influence weight corresponding to the content feature information and the second influence weight corresponding to the timing feature information based on the current release time, and uses the feature fusion sub-model to weightedly process and fuse the content feature information and the timing feature information based on the first influence weight and the second influence weight, respectively, so as to flexibly control the inflow of different information at different times, realize the dynamic fusion of information, and further improve the accuracy of popularity prediction.

[0109] The following is an embodiment of a video popularity prediction device provided by an embodiment of the present disclosure. The device and the video popularity prediction method of the above embodiment belong to the same inventive concept. For details not fully described in the embodiment of the video popularity prediction device, please refer to the video popularity prediction method of the above embodiment.

[0110] Example 4

[0111] Figure 8 This is a schematic diagram of the structure of a video popularity prediction device provided by the fourth embodiment of the present disclosure. This embodiment is applicable to the case of predicting the popularity of videos released by users. Figure 8 As shown, the device specifically includes: a data acquisition module 810, a feature extraction module 820, a current release duration determination module 830, a feature fusion module 840 and a popularity prediction module 850.

[0112] Among them, the data acquisition module 810 is used to obtain the target metadata and current playback data corresponding to the video to be predicted; the feature extraction module 820 is used to extract features from the target metadata to determine the content feature information, and to extract features from the current playback data to determine the timing feature information; the current release duration determination module 830 is used to determine the current release duration corresponding to the video to be predicted at the current prediction moment based on the release moment of the video to be predicted; the feature fusion module 840 is used to input the content feature information, timing feature information and current release duration into a preset feature fusion model for information fusion, and determine the fused target video feature information; the popularity prediction module 850 is used to input the target video feature information into a preset popularity prediction model for popularity prediction, and determine the popularity prediction result corresponding to the video to be predicted.

[0113] The technical solution of the embodiment of the present disclosure performs feature extraction on the target metadata and current playback data corresponding to the video to be predicted respectively, determines the content feature information and timing feature information, and determines the current release duration corresponding to the video to be predicted at the current prediction time based on the release time of the video to be predicted, and inputs the content feature information, timing feature information and current release duration into a preset feature fusion model, so that the preset feature fusion model can dynamically consider the influence of the content feature information and timing feature information of the video to be predicted on the popularity of the video based on the current release duration, realizes dynamic fusion of information, and makes the fused target video feature information more fully characterize the popularity of the video, and then inputs the target video feature information into the preset popularity prediction model for popularity prediction, so as to obtain more accurate popularity prediction results and improve the accuracy of video popularity prediction.

[0114] Based on the above technical solution, the feature extraction module 820 includes:

[0115] A knowledge graph acquisition unit, configured to acquire a knowledge graph pre-built based on metadata corresponding to each video;

[0116] A target subgraph determination unit, configured to determine a target subgraph of the video to be predicted in the knowledge graph based on the target metadata;

[0117] The content feature extraction unit is used to input the target subgraph into a preset content feature extraction model to extract features and determine content feature information.

[0118] On the basis of the above technical solutions, the preset content feature extraction model includes: a structure transformation sub-model and an information aggregation sub-model;

[0119] Content feature extraction unit, including:

[0120] A target tree determination subunit is used to input the target subgraph into the structure transformation submodel for structural transformation, and determine a target tree with a tree structure;

[0121] The content feature extraction sub-unit is used to input the target tree into the information aggregation sub-model, propagate and aggregate the node information of the target tree from the bottom up, and determine the content feature information based on the aggregated root node information.

[0122] Based on the above technical solutions, the content feature extraction subunit is specifically used to:

[0123] Starting from the bottom layer of the target tree, each parent node in each layer is used as the current parent node; based on the current parent node information and each current child node information corresponding to the current parent node, the child node attention weight corresponding to each current child node is determined; based on the child node attention weight, each current child node information is propagated upward to the current parent node to determine the target propagation information corresponding to the current parent node; the current parent node information and the target propagation information are aggregated to determine the target node information corresponding to the aggregated current parent node.

[0124] Based on the above technical solutions, the feature extraction module 820 further includes:

[0125] The temporal feature extraction unit is used to: input the current playback data into a preset temporal feature extraction model with an attention mechanism for feature extraction, and determine the temporal feature information.

[0126] On the basis of the above technical solutions, the preset feature fusion model includes: an influence weight determination sub-model and a feature fusion sub-model;

[0127] The feature fusion module 840 includes:

[0128] An influence weight determination unit, configured to input the current publishing time into an influence weight determination sub-model to determine a first influence weight corresponding to the content feature information and a second influence weight corresponding to the timing feature information;

[0129] The feature fusion unit is used to input content feature information, timing feature information, a first influence weight and a second influence weight into a feature fusion sub-model, perform weighted processing on the content feature information and the timing feature information based on the first influence weight and the second influence weight respectively, and fuse the target content feature information and the target timing feature information obtained after processing to determine the fused target video feature information.

[0130] On the basis of the above technical solutions, the influence weight determination sub-model includes a preset attenuation function obtained after training;

[0131] The influence weight determination unit is specifically used to: determine the first influence weight corresponding to the content feature information based on a preset attenuation function and the current release time; and determine the second influence weight corresponding to the timing feature information based on the first influence weight.

[0132] Based on the above technical solutions, the feature fusion sub-model is a feature fusion sub-model with an attention mechanism;

[0133] Feature fusion unit, including:

[0134] An attention processing subunit is used to perform attention processing on the target content feature information and the target temporal feature information obtained after processing, and determine the content attention weight corresponding to the target content feature information and the temporal attention weight corresponding to the target temporal feature information;

[0135] The target group video feature information determination subunit is used to determine the target video feature information based on the target content feature information, the target temporal feature information, the content attention weight and the temporal attention weight.

[0136] Based on the above technical solutions, the attention processing subunit is specifically used to:

[0137] The target content feature information and the target temporal feature information are concatenated and linearly transformed to obtain global feature information; the target content feature information and the global feature information are dot-product processed to determine the content attention weight corresponding to the target content feature information, and the target temporal feature information and the global feature information are dot-product processed to determine the temporal attention weight corresponding to the target temporal feature information.

[0138] The video popularity prediction device provided by the embodiments of the present disclosure can execute the video popularity prediction method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the video popularity prediction method.

[0139] It is worth noting that in the embodiment of the above-mentioned video popularity prediction device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0140] Example 5

[0141] Reference below Figure 9 , which shows a structural diagram of an electronic device 900 suitable for implementing the embodiments of the present disclosure. Figure 9 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0142] like Figure 9 As shown, the electronic device 900 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage device 908 into a random access memory (RAM) 903. Various programs and data required for the operation of the electronic device 900 are also stored in the RAM 903. The processing device 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0143] Typically, the following devices may be connected to the I / O interface 905: an input device 906 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 907 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 908 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 909. The communication device 909 may allow the electronic device 900 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 9 The electronic device 900 is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead.

[0144] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 909, or installed from the storage device 908, or installed from the ROM 902. When the computer program is executed by the processing device 901, the above-mentioned functions defined in the method of the embodiment of the present disclosure are performed.

[0145] The electronic device provided in the embodiment of the present disclosure and the video popularity prediction method provided in the above embodiment belong to the same inventive concept. For technical details not fully described in the embodiment of the present disclosure, please refer to the above embodiment, and the embodiment of the present disclosure has the same beneficial effects as the above embodiment.

[0146] Example 6

[0147] An embodiment of the present disclosure provides a computer storage medium having a computer program stored thereon. When the program is executed by a processor, the video popularity prediction method provided by the above embodiment is implemented.

[0148] It should be noted that the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, device, or component. In the present disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0149] In some embodiments, the client and server can communicate using any currently known or future developed network protocol, such as HTTP (HyperText Transfer Protocol), and can be interconnected with 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"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0150] The computer-readable medium may be included in the server, or may exist independently without being incorporated into the server.

[0151] The above-mentioned computer-readable medium carries one or more programs. When the above-mentioned one or more programs are executed by the server, the server: obtains the target metadata and current playback data corresponding to the video to be predicted; performs feature extraction on the target metadata to determine content feature information, and performs feature extraction on the current playback data to determine timing feature information; based on the release time of the video to be predicted, determines the current release duration corresponding to the video to be predicted at the current prediction time; inputs the content feature information, the timing feature information and the current release duration into a preset feature fusion model for information fusion, and determines the fused target video feature information; inputs the target video feature information into a preset popularity prediction model for popularity prediction, and determines the popularity prediction result corresponding to the video to be predicted.

[0152] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages, or a combination thereof, including, but not limited to, object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0153] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0154] The units involved in the embodiments described in this disclosure may be implemented by software or hardware. In some cases, the name of a unit does not constitute a limitation on the unit itself. For example, an editable content display unit may also be described as an "editing unit."

[0155] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0156] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0157] According to one or more embodiments of the present disclosure, [Example 1] provides a video popularity prediction method, including:

[0158] Get the target metadata and current playback data corresponding to the video to be predicted;

[0159] Extracting features from the target metadata to determine content feature information, and extracting features from the currently played data to determine timing feature information;

[0160] Determining the current release duration of the video to be predicted at the current prediction time based on the release time of the video to be predicted;

[0161] Inputting the content feature information, the timing feature information and the current release duration into a preset feature fusion model for information fusion, and determining the fused target video feature information;

[0162] The target video feature information is input into a preset popularity prediction model to perform popularity prediction, and a popularity prediction result corresponding to the video to be predicted is determined.

[0163] According to one or more embodiments of the present disclosure, [Example 2] provides a video popularity prediction method, further comprising:

[0164] Optionally, feature extraction is performed on the target metadata to determine content feature information, including:

[0165] Obtain a knowledge graph pre-built based on the metadata corresponding to each video;

[0166] Based on the target metadata, determining a target subgraph of the video to be predicted in the knowledge graph;

[0167] The target subgraph is input into a preset content feature extraction model to perform feature extraction and determine content feature information.

[0168] According to one or more embodiments of the present disclosure, [Example 3] provides a video popularity prediction method, further comprising:

[0169] Optionally, the preset content feature extraction model includes: a structure transformation sub-model and an information aggregation sub-model;

[0170] The step of inputting the target subgraph into a preset content feature extraction model to extract features and determine content feature information includes:

[0171] Inputting the target subgraph into the structure transformation submodel to perform structure transformation, and determining a target tree having a tree structure;

[0172] The target tree is input into the information aggregation sub-model, the node information of the target tree is propagated and aggregated from bottom to top, and content feature information is determined based on the aggregated root node information.

[0173] According to one or more embodiments of the present disclosure, [Example 4] provides a video popularity prediction method, further comprising:

[0174] Optionally, the bottom-up propagation and aggregation of the node information of the target tree includes:

[0175] Starting from the bottom layer of the target tree, each parent node in each layer is used as the current parent node;

[0176] Based on the current parent node information and each current child node information corresponding to the current parent node, determine the child node attention weight corresponding to each current child node;

[0177] Based on the attention weights of the child nodes, the information of each current child node is propagated upward to the current parent node, and the target propagation information corresponding to the current parent node is determined;

[0178] Aggregate the current parent node information with the target propagation information to determine the target node information corresponding to the aggregated current parent node.

[0179] According to one or more embodiments of the present disclosure, [Example 5] provides a video popularity prediction method, further comprising:

[0180] Optionally, extracting features from the currently played data to determine timing feature information includes:

[0181] The current playback data is input into a preset temporal feature extraction model with an attention mechanism for feature extraction to determine temporal feature information.

[0182] According to one or more embodiments of the present disclosure, [Example 6] provides a video popularity prediction method, further comprising:

[0183] Optionally, the preset feature fusion model includes: an influence weight determination sub-model and a feature fusion sub-model;

[0184] The step of inputting the content feature information, the timing feature information, and the current release duration into a preset feature fusion model for information fusion, and determining the fused target video feature information, includes:

[0185] Inputting the current publishing time into the influence weight determination sub-model to determine a first influence weight corresponding to the content feature information and a second influence weight corresponding to the timing feature information;

[0186] The content feature information, the timing feature information, the first influence weight and the second influence weight are input into the feature fusion sub-model, the content feature information and the timing feature information are weighted based on the first influence weight and the second influence weight respectively, and the target content feature information and target timing feature information obtained after processing are fused to determine the fused target video feature information.

[0187] According to one or more embodiments of the present disclosure, [Example 7] provides a video popularity prediction method, further comprising:

[0188] Optionally, the influence weight determination sub-model includes a preset attenuation function obtained after training;

[0189] The determining of the first influence weight corresponding to the content feature information and the second influence weight corresponding to the timing feature information includes:

[0190] Determining a first influence weight corresponding to the content feature information based on the preset attenuation function and the current publishing time;

[0191] Based on the first impact weight, a second impact weight corresponding to the time series feature information is determined.

[0192] According to one or more embodiments of the present disclosure, [Example 8] provides a video popularity prediction method, further comprising:

[0193] Optionally, the feature fusion sub-model is a feature fusion sub-model with an attention mechanism;

[0194] The step of fusing the target content feature information and the target temporal feature information obtained after processing to determine the fused target video feature information includes:

[0195] Performing attention processing on the target content feature information and the target temporal feature information obtained after processing, and determining a content attention weight corresponding to the target content feature information and a temporal attention weight corresponding to the target temporal feature information;

[0196] Target video feature information is determined based on the target content feature information, the target temporal feature information, the content attention weight, and the temporal attention weight.

[0197] According to one or more embodiments of the present disclosure, [Example 9] provides a video popularity prediction method, further comprising:

[0198] Optionally, performing attention processing on the target content feature information and target temporal feature information obtained after processing to determine a content attention weight corresponding to the target content feature information and a temporal attention weight corresponding to the target temporal feature information includes:

[0199] performing splicing and linear transformation processing on the target content feature information and the target time series feature information to obtain global feature information;

[0200] Perform dot product processing on the target content feature information and the global feature information to determine the content attention weight corresponding to the target content feature information, and perform dot product processing on the target temporal feature information and the global feature information to determine the temporal attention weight corresponding to the target temporal feature information.

[0201] According to one or more embodiments of the present disclosure, [Example 10] provides a video popularity prediction device, including:

[0202] The data acquisition module is used to obtain the target metadata and current playback data corresponding to the video to be predicted;

[0203] A feature extraction module is used to extract features from the target metadata to determine content feature information, and to extract features from the currently played data to determine timing feature information;

[0204] A current release duration determination module, configured to determine, based on the release time of the video to be predicted, the current release duration corresponding to the video to be predicted at the current prediction time;

[0205] A feature fusion module is used to input the content feature information, the timing feature information and the current release duration into a preset feature fusion model to perform information fusion and determine the fused target video feature information;

[0206] The popularity prediction module is used to input the target video feature information into a preset popularity prediction model to perform popularity prediction and determine the popularity prediction result corresponding to the video to be predicted.

[0207] The above description is merely a preferred embodiment of the present disclosure and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also includes other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned disclosed concepts. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this disclosure.

[0208] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details have been included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0209] Although the subject matter has been described in language specific to structural features and / or methodological logical acts, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are merely example forms of implementing the claims.

Claims

1. A video popularity prediction method, characterized in that: include: Get the target metadata and current playback data corresponding to the video to be predicted; Extracting features from the target metadata to determine content feature information, and extracting features from the currently played data to determine timing feature information; Determining the current release duration of the video to be predicted at the current prediction time based on the release time of the video to be predicted; The content feature information, the timing feature information and the current release time are input into a preset feature fusion model for information fusion to determine the fused target video feature information, wherein the preset feature fusion model includes an influence weight determination sub-model and a feature fusion sub-model; the influence weight determination sub-model is used to determine a first influence weight corresponding to the content feature information and a second influence weight corresponding to the timing feature information based on the current release time, and the feature fusion sub-model is used to perform weighted processing and fusion processing based on the first influence weight, the second influence weight, the content feature information and the timing feature information to determine the fused target video feature information; The target video feature information is input into a preset popularity prediction model to perform popularity prediction, and a popularity prediction result corresponding to the video to be predicted is determined.

2. The method according to claim 1, characterized in that Extracting features from the target metadata to determine content feature information includes: Obtain a knowledge graph pre-built based on the metadata corresponding to each video; Based on the target metadata, determining a target subgraph of the video to be predicted in the knowledge graph; The target subgraph is input into a preset content feature extraction model to perform feature extraction and determine content feature information.

3. The method according to claim 2, characterized in that The preset content feature extraction model includes: a structure transformation sub-model and an information aggregation sub-model; The step of inputting the target subgraph into a preset content feature extraction model to extract features and determine content feature information includes: Inputting the target subgraph into the structure transformation submodel to perform structure transformation, and determining a target tree having a tree structure; The target tree is input into the information aggregation sub-model, the node information of the target tree is propagated and aggregated from bottom to top, and content feature information is determined based on the aggregated root node information.

4. The method according to claim 3, characterized in that The bottom-up propagation and aggregation of node information of the target tree includes: Starting from the bottom layer of the target tree, each parent node in each layer is used as the current parent node; Based on the current parent node information and each current child node information corresponding to the current parent node, determine the child node attention weight corresponding to each current child node; Based on the attention weights of the child nodes, the information of each current child node is propagated upward to the current parent node, and the target propagation information corresponding to the current parent node is determined; Aggregate the current parent node information with the target propagation information to determine the target node information corresponding to the aggregated current parent node.

5. The method according to claim 1, wherein The extracting features of the currently played data to determine timing feature information includes: The current playback data is input into a preset temporal feature extraction model with an attention mechanism for feature extraction to determine temporal feature information.

6. The method according to claim 1, characterized in that The step of inputting the content feature information, the timing feature information, and the current release duration into a preset feature fusion model for information fusion, and determining the fused target video feature information, includes: Inputting the current publishing time into the influence weight determination sub-model to determine a first influence weight corresponding to the content feature information and a second influence weight corresponding to the timing feature information; The content feature information, the timing feature information, the first influence weight and the second influence weight are input into the feature fusion sub-model, the content feature information and the timing feature information are weighted based on the first influence weight and the second influence weight respectively, and the target content feature information and target timing feature information obtained after processing are fused to determine the fused target video feature information.

7. The method according to claim 6, characterized in that The influence weight determination sub-model includes a preset attenuation function obtained after training; The determining of the first influence weight corresponding to the content feature information and the second influence weight corresponding to the timing feature information includes: Determining a first influence weight corresponding to the content feature information based on the preset attenuation function and the current publishing time; Based on the first impact weight, a second impact weight corresponding to the time series feature information is determined.

8. The method according to claim 6, characterized in that The feature fusion sub-model is a feature fusion sub-model with an attention mechanism; The step of fusing the target content feature information and the target temporal feature information obtained after processing to determine the fused target video feature information includes: Performing attention processing on the target content feature information and the target temporal feature information obtained after processing, and determining a content attention weight corresponding to the target content feature information and a temporal attention weight corresponding to the target temporal feature information; Target video feature information is determined based on the target content feature information, the target temporal feature information, the content attention weight, and the temporal attention weight.

9. The method according to claim 8, characterized in that The performing attention processing on the target content feature information and the target temporal feature information obtained after the processing, and determining the content attention weight corresponding to the target content feature information and the temporal attention weight corresponding to the target temporal feature information, includes: performing splicing and linear transformation processing on the target content feature information and the target time series feature information to obtain global feature information; Perform dot product processing on the target content feature information and the global feature information to determine the content attention weight corresponding to the target content feature information, and perform dot product processing on the target temporal feature information and the global feature information to determine the temporal attention weight corresponding to the target temporal feature information.

10. A video popularity prediction device, characterized in that: include: The data acquisition module is used to obtain the target metadata and current playback data corresponding to the video to be predicted; A feature extraction module is used to extract features from the target metadata to determine content feature information, and to extract features from the currently played data to determine timing feature information; A current release duration determination module, configured to determine, based on the release time of the video to be predicted, the current release duration corresponding to the video to be predicted at the current prediction time; a feature fusion module, configured to input the content feature information, the timing feature information, and the current release duration into a preset feature fusion model for information fusion, and determine the fused target video feature information, wherein the preset feature fusion model includes an influence weight determination submodel and a feature fusion submodel; the influence weight determination submodel is configured to determine a first influence weight corresponding to the content feature information and a second influence weight corresponding to the timing feature information based on the current release duration; the feature fusion submodel is configured to perform weighted processing and fusion processing based on the first influence weight, the second influence weight, the content feature information, and the timing feature information, and determine the fused target video feature information; The popularity prediction module is used to input the target video feature information into a preset popularity prediction model to perform popularity prediction and determine the popularity prediction result corresponding to the video to be predicted.

11. An electronic device, characterized in that: The electronic device comprises: one or more processors; a memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the video popularity prediction method according to any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the video popularity prediction method according to any one of claims 1 to 9 is implemented.

Images