Publishing time period estimation method and device, equipment and storage medium

Abstract

The invention relates to a publishing time period estimation method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring video semantic features of a target video to be published to a video platform; determining a first video set similar to the target video in a platform video library; determining video platform side statistical information according to the video popularity related information of the video platform in each preset time period and the video page view of each video in the first video set in each preset time period; based on the video semantic features of the target video and the video platform side statistical information, outputting an estimated playing amount of the target video in each preset time period; according to the pre-estimated playing amount of the target video in each preset time period, the release time period of the target video is determined, and the accuracy of the pre-estimated release time period is improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a release period estimation method, apparatus, computer equipment, storage medium and computer program product. Background Art

[0002] With the rapid development of Internet technology and computer technology, various video platforms have provided support for video creators to create videos, publish videos, and users to browse videos, meeting different video needs.

[0003] In order to increase the number of views of videos released by creators, some research has been proposed in the related art on the release period of videos, so as to estimate the release period of videos. However, the information considered in the related art when estimating the release period is relatively simple, resulting in the estimated release period being inaccurate. Summary of the invention

[0004] Based on this, it is necessary to provide a release period estimation method, device, computer equipment, storage medium and computer program product to address the above technical problems. When estimating the release period, it can take into account that videos of different types of content have different audience groups in different time periods and the release periods of other videos of the same type of content, thereby improving the accuracy of the estimated release period.

[0005] In a first aspect, the present application provides a method for estimating a release period. The method comprises:

[0006] Obtain video semantic features of a target video to be published to a video platform;

[0007] Determine a first video set in a platform video library that is similar to the target video;

[0008] Determine the video platform side statistical information according to the video popularity related information of the video platform in each preset time period and the video views of each video in the first video set in each preset time period;

[0009] Based on the video semantic features of the target video and the statistical information on the video platform side, outputting the estimated playback volume of the target video in each preset time period;

[0010] The release period of the target video is determined according to the estimated playback volume of the target video in each preset time period.

[0011] In a second aspect, the present application also provides a release period estimation device. The device comprises:

[0012] A semantic feature acquisition module, used to acquire the video semantic features of the target video to be published to the video platform;

[0013] A video set determination module, used to determine a first video set similar to the target video in the platform video library;

[0014] A statistical information acquisition module, configured to determine statistical information on the video platform side according to information related to video popularity of the video platform in each preset time period and the number of video views of each video in the first video set in each preset time period;

[0015] An output module, configured to output an estimated playback volume of the target video in each preset time period based on the video semantic features of the target video and the statistical information on the video platform side;

[0016] The release period determination module is used to determine the release period of the target video according to the estimated playback volume of the target video in each preset period.

[0017] In some embodiments, the semantic feature acquisition module is used to acquire a target video to be published to the video platform; and to extract semantic features from the target video to obtain video semantic features of the target video.

[0018] In some embodiments, the target video is to be released through a target account; the video collection determination module is used to obtain platform videos released through other accounts from the platform video library; the other accounts are accounts other than the target account; semantic features are extracted from the obtained platform videos to obtain video semantic features of each of the platform videos; based on the similarity between the video semantic features of the target video and the video semantic features of each of the platform videos, similar videos to the target video are screened out from the obtained platform videos; based on the screened similar videos, a first video collection is obtained.

[0019] In some embodiments, the statistical information acquisition module is used to count the number of videos published on the video platform in each preset time period; count the number of accounts active on the video platform in each preset time period; count the video views of each video in the first video set in each preset time period; and use the counted number of videos published on the video platform in each preset time period, the number of accounts active on the video platform in each preset time period, and the video views of each video in the first video set in each preset time period as the statistical information on the video platform side.

[0020] In some embodiments, the output module is used to generate a first information type code according to a first information type, and to generate a second information type code according to a second information type, wherein the first information type is used to represent the information type of the semantic features of the video, and the second information type is used to represent the information type of the statistical information on the video platform side; generate a first vector type code according to a vector type corresponding to the first type of statistical vector, and generate a first type of statistical vector on the platform side according to the number of videos published on the video platform in each preset time period; generate a second vector type code according to a vector type corresponding to the second type of statistical vector, and generate a platform side according to the number of accounts active on the video platform in each preset time period. A second type of statistical vector; generating a third vector type code according to the vector type corresponding to the third type of statistical vector, and generating a third type of statistical vector on the platform side according to the statistical video views of each video in the first video set in each preset time period; through a playback volume estimation model, according to the first information type code and the video semantic features, the second information type code, the first vector type code and the first type of statistical vector on the platform side, the second information type code, the second vector type code and the second type of statistical vector on the platform side, and the second information type code, the third vector type code and the third type of statistical vector on the platform side, output the estimated playback volume of the target video in each preset time period. In some embodiments, the output module is used to obtain video side input data according to the first information type code and the video semantic features of the target video; obtain first-dimensional input data on the platform side according to the second information type code, the first vector type code and the first type of statistical vector on the platform side; obtain second-dimensional input data on the platform side according to the second information type code, the second vector type code and the second type of statistical vector on the platform side; obtain third-dimensional input data on the platform side according to the second information type code, the third vector type code and the third type of statistical vector on the platform side; and output the estimated playback volume of the target video in each preset time period according to the video side input data, the first-dimensional input data on the platform side, the second-dimensional input data on the platform side and the third-dimensional input data on the platform side through a playback volume estimation model.

[0021] In some embodiments, the target video is to be released through a target account, and the video set determination module is further used to determine a second video set that is similar to the target video in historical videos released through the target account; the statistical information acquisition module is further used to determine the account-side statistical information based on the account popularity-related information of the target account in each preset time period and the video views of each video in the second video set in each preset time period; the output module is used to output the estimated playback volume of the target video in each preset time period based on the video semantic features of the target video, the video platform-side statistical information and the account-side statistical information.

[0022] In some embodiments, the video collection determination module is also used to obtain historical videos released by the target account; perform semantic feature extraction on the obtained historical videos to obtain video semantic features of each of the historical videos; filter out similar videos of the target video from the obtained historical videos based on the similarity between the video semantic features of the target video and the video semantic features of each of the historical videos; and obtain a second video collection based on the filtered similar videos.

[0023] In some embodiments, the statistical information acquisition module is also used to count the number of videos posted on the video platform by the target account in each preset time period; count the number of active accounts regarding videos posted by the target account in each preset time period; count the video views of each video in the second video set in each preset time period; and use the counted number of videos posted on the video platform in each preset time period, the number of accounts active on the video platform in each preset time period, and the video views of each video in the second video set in each preset time period as the account-side statistical information.

[0024] In some embodiments, the output module is used to generate a first information type code according to a first information type, generate a second information type code according to a second information type, and generate a third information type code according to a third information type, wherein the first information type is used to represent the information type of the semantic feature of the video, the second information type is used to represent the information type of the statistical information on the video platform side, and the third information type is used to represent the information type of the statistical information on the account side; generate a first vector type code according to the vector type corresponding to the first type of statistical vector, and generate a first type of statistical vector on the platform side according to the statistical number of videos posted on the video platform in each preset time period, and generate a first type of statistical vector on the account side according to the statistical number of videos posted on the video platform by the target account in each preset time period; generate a second vector type code according to the vector type corresponding to the second type of statistical vector, and generate a second type of statistical vector on the platform side according to the statistical number of accounts active on the video platform in each preset time period, and generate a second type of statistical vector on the account side according to the statistical number of active accounts regarding videos posted by the target account in each preset time period. Statistical vector; generating a third vector type code according to the vector type corresponding to the third type of statistical vector, and generating a third type of statistical vector on the platform side according to the statistical video views of each video in the first video set in each preset time period, and generating a third type of statistical vector on the account side according to the statistical video views of each video in the second video set in each preset time period; through the playback volume estimation model, according to the first information type code and the video semantic features, the second information type code, the first vector type code and the first type of statistical vector on the platform side, the second information type code, the second vector type code and the second type of statistical vector on the platform side, the second information type code, the third vector type code and the third type of statistical vector on the platform side, the third information type code, the first vector type code and the first type of statistical vector on the account side, the third information type code, the second vector type code and the second type of statistical vector on the account side, the third information type code, the third vector type code and the third type of statistical vector on the account side, the third information type code, the third vector type code and the third type of statistical vector on the account side, the third information type code, the third vector type code and the third type of statistical vector on the account side, the

[0025] In some embodiments, the release period determination module is used to determine the preset period corresponding to the maximum value of the estimated playback volume of each preset period based on the estimated playback volume of the target video in each preset period; if the preset period corresponding to the maximum value is a non-prohibited release period, then the preset period corresponding to the maximum value is determined as the release period of the target video.

[0026] In some embodiments, the estimated playback volume is output by a playback volume estimation model based on the video semantic features of the target video and the statistical information on the video platform side. The device also includes a model training module, which is used to obtain training samples, and the training samples include sample videos, the annotated release time period of the sample videos, and the sum of the annotated playback volumes of the sample videos in each preset time period; obtain the video semantic features of the sample videos; determine a third video set similar to the sample video in the platform video library; determine the statistical information on the video platform side based on the video popularity related information of the video platform in each preset time period and the video views of each video in the third video set in each preset time period; output the estimated playback volume of the sample video in each preset time period based on the video semantic features and the statistical information on the video platform side through the playback volume estimation model; update the playback volume estimation model based on the difference between the estimated playback volume of the sample video corresponding to the annotated release time period and the sum of the annotated playback volume in each preset time period.

[0027] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the above-mentioned release period estimation method are implemented.

[0028] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above-mentioned release period estimation method when executed by a processor.

[0029] In a fifth aspect, the present application further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned release period estimation method are implemented.

[0030] The above-mentioned release period estimation method, device, computer equipment, storage medium and computer program product obtain the video semantic features of the target video to be released to the video platform; determine the first video set similar to the target video in the platform video library, that is, determine the first video set that has been released and is of the same type as the target video from the platform video library. According to the video platform's video heat related information in each preset period and the video views of each video in the first video set in each preset period, the video platform side statistical information is determined. In this way, the video platform side statistical information can more comprehensively reflect the user's attention to the same type of video in different time periods, thereby intuitively reflecting the audience groups of the same type of video in different time periods. Based on this, based on the video semantic features of the target video and the video platform side statistical information, it can be considered that videos of different types of content have different audience groups and the release periods of other videos of the same type of content in different time periods, so that the playback volume of each preset period can be accurately estimated, and the estimated playback volume of the target video in each preset period can be output; therefore, according to the estimated playback volume of the target video in each preset period, the release period of the target video can be more accurately determined, which improves the accuracy of estimating the release period of the target video. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A diagram showing an application environment of a release period estimation method in an embodiment;

[0032] Figure 2 It is a flowchart of a method for estimating a release period in one embodiment;

[0033] Figure 3A A schematic diagram of training a model for image description in one embodiment;

[0034] Figure 3B A schematic diagram of training a model for image description in another embodiment;

[0035] Figure 4 A schematic diagram of a flow chart of steps for determining statistical information on a video platform side in one embodiment;

[0036] Figure 5 A schematic diagram of a flow chart of a step for determining an estimated playback volume in one embodiment;

[0037] Figure 6 A schematic diagram of a flow chart of a step for determining an estimated playback volume in another embodiment;

[0038] Figure 7 A flowchart of a step of determining account-side statistical information in one embodiment;

[0039] Figure 8A schematic diagram of a flow chart of a step of determining an estimated playback volume in another embodiment;

[0040] Fig. 9 A schematic diagram of a flow chart of a step for determining an estimated playback volume in another embodiment;

[0041] Fig.10 A schematic diagram of a flow chart of a model training step in an embodiment;

[0042] Fig.11A is a flowchart of steps for determining a video set in one embodiment;

[0043] Fig. 11B A release period determination process in one embodiment;

[0044] Fig.12 It is a structural block diagram of a release period estimation device in one embodiment;

[0045] Fig.13 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0047] Usually, after a creator has created a video, he or she needs to choose a release time for the video. However, the information considered in the relevant technology for estimating the release time is relatively simple, which may easily cause the release time period to be unable to match different audience groups. For example, motivational videos such as news videos and positive energy videos are suitable for release in the morning, while emotional movies and other videos with emotional appeals are suitable for release in the evening. After adopting the relevant technology, videos with emotional appeals may be released in the morning, and motivational videos may be sent at night. Obviously, the relevant technology cannot accurately match the release time period with the user, and cannot accurately estimate the release time period of the video to be released.

[0048] The release period estimation method provided by the embodiment of the present application obtains the video semantic features of the target video to be released to the video platform; determines the first video set similar to the target video in the platform video library, that is, determines the first video set that has been released and is of the same type as the target video from the platform video library. According to the video platform's video heat related information in each preset period and the video views of each video in the first video set in each preset period, the video platform side statistical information is determined. In this way, the video platform side statistical information can more comprehensively reflect the user's attention to the same type of video in different time periods, thereby intuitively reflecting the audience groups of the same type of video in different time periods. Based on this, based on the video semantic features of the target video and the video platform side statistical information, it can be considered that videos of different types of content have different audience groups and the release periods of other videos of the same type of content in different time periods, so that the playback volume of each preset period can be accurately estimated, and the estimated playback volume of the target video in each preset period can be output; therefore, according to the estimated playback volume of the target video in each preset period, the release period of the target video can be more accurately determined, which improves the accuracy of estimating the release period of the target video.

[0049] The release period estimation method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, can be integrated on the server 104, or can be integrated on the cloud or other servers. The terminal 102 and the server 104 can both execute the release period prediction method separately, and the terminal 102 and the server 104 can also assist in executing the release period prediction method.

[0050] In some embodiments, the release period estimation method provided in the embodiments of the present application can be executed by the server 104. The server 104 obtains the video semantic features of the target video to be released to the video platform; the server 104 determines the first video set similar to the target video in the platform video library; according to the video popularity related information of the video platform in each preset time period and the video views of each video in the first video set in each preset time period, the video platform side statistical information is determined; based on the video semantic features of the target video and the video platform side statistical information, the server 104 outputs the estimated playback volume of the target video in each preset time period; the server 104 determines the release period of the target video according to the estimated playback volume of the target video in each preset time period.

[0051] Optionally, the server 104 sends the publishing period of the target video to the terminal 102, and the terminal 102 publishes the target video in the publishing period.

[0052] Exemplarily, the terminal 102 deploys a client with a video publishing function, which may be a video client, an e-commerce client, a social application client, etc. After obtaining the target video to be published by the creator, the server 104 uses the publishing period estimation method provided in the embodiment of the present application to determine the publishing period of the target video, and sends the publishing period to the terminal 102, and publishes the target video in the client with a video publishing function through the target account.

[0053] The terminal 102 may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented by an independent server or a server cluster consisting of multiple servers, or by a cloud server.

[0054] In one embodiment, Figure 2 As shown, a method for estimating a release period is provided, and the method is applied to a computer device (which may be Figure 1 The server 104 in the example, which may also be the terminal 102, is used as an example to illustrate, and the following steps are included:

[0055] Step S202, obtaining video semantic features of a target video to be published to a video platform.

[0056] The video platform is a platform that provides support for the distributed video function. Optionally, the video platform provides support for the video publishing function and video viewing function of the video client.

[0057] The target video is a video to be published on the video platform. The target video can be a short video or a long video, and there is no specific limitation. The video semantic feature is used to characterize the semantic content of the video. The more similar the video semantic features of any two videos are, the closer the semantic content of the any two videos is, and the greater the possibility that the any two videos belong to the same video type.

[0058] Optionally, the computer device obtains a target video to be published on the video platform, and determines the video semantic features of the target video, wherein the target video is a video to be published through a target account.

[0059] Exemplarily, after acquiring the target video, the computer device extracts semantic features of the target video to obtain video semantic features of the target video.

[0060] In some embodiments, obtaining the video semantic features of a target video to be published on a video platform includes: obtaining the target video to be published on the video platform; extracting semantic features from the target video to obtain the video semantic features of the target video. The semantic feature extraction model is a neural network model, and the semantic feature extraction model is a trained model. The semantic feature extraction model is used to extract the semantic features of the video.

[0061] Optionally, the computer device obtains a target video to be published to the video platform, and calls a semantic feature extraction model to extract semantic features of the target video, obtain video semantic features of the target video, and determine a video semantic feature vector of the video semantic features. The video semantic feature vector is a set of vector data used to characterize the video semantic features.

[0062] The semantic feature extraction model is an encoder-decoder structure, which is a model for image description, such as Figure 3A FIG. 1 is a schematic diagram of training a model for image description in an embodiment, wherein the encoder encodes the image and text respectively, and the decoder generates an image description based on the encoded information. In the process of training the semantic feature extraction model, sample pairs for training are obtained, and the sample pairs include sample pairs of text and image. Figure 3A In the example, the text in the sample pair is "two dogs running in the field" and the corresponding image. The image shows two different dogs running in the field. The image is encoded by the image encoder to obtain the image encoding result, and the image encoding result is processed by the confluence layer to obtain the confluence result. The text is segmented to obtain the segmentation result, such as Figure 3A , the word segmentation result is [s][two][dogs][in][field][running][cls], where [cls] is used to extract the unified representation of the text, and [s] is the start symbol; the word segmentation result is encoded by a single text encoder to obtain a text encoding result. The text encoding result is processed by a multimodal text decoder to obtain a text decoding result, such as Figure 3A As shown, the text decoding result is [two] [dogs] [in] [a field] [running] [ / s], and [ / s] is the end symbol. According to the merging result and the text encoding result, the contrast loss is determined; the merging result is cross-attention processed to determine the text generation result, and the generation loss is determined according to the text generation result and the text encoding result, and the contrast loss and the generation loss are used to train the model to obtain a trained model for image description.

[0063] Furthermore, the above-mentioned image description model can be applied to video classification tasks. In video classification tasks, in order to make the encoder-decoder structure model better perform video recognition and classification, such as Figure 3BFIG. 1 is a schematic diagram of training a model for image description in another embodiment. The cross entropy loss can be directly used to train the model for image description. Specifically, a sample video is obtained, and multiple video frames are uniformly sampled from the sample video, such as Figure 3B 16 video frames are sampled from the image. Each video frame is input into the shared image encoder of the image description model in turn to obtain the image encoding result corresponding to each video frame. The image encoding results corresponding to multiple video frames are processed by the confluence layer in the image description model to obtain the sample video semantic features corresponding to the sample video. Since a video may have multiple labels, for example, including i labels, the sample video semantic feature x corresponding to the i-th label i Therefore, the sigmoid function is used for activation processing, that is, the image encoding results corresponding to multiple video frames are processed through the confluence layer in the model for image description to obtain sample video semantic features corresponding to multiple labels. At this time, for each label, the sigmoid function is used to map the sample video semantic features corresponding to each label to obtain the mapping result p i , which can also be understood as the probability that the sample video belongs to the i-th label. The specific mapping process is as follows:

[0064]

[0065] For example, for a three-classification problem, i is 3. If the label (y1, y2, y3) of the sample video is (0, 1, 1), it means that the sample video belongs to the second category and the third category. The cross entropy loss L is constructed according to the mapping results corresponding to each label, for example, as shown below:

[0066]

[0067] y i Indicates the value of the i-th label. For example, if it belongs to this label, the corresponding value is 1; if it does not belong to this label, the corresponding value is 0. i is the probability that the sample video is the i-th label, and ∑(.) is the summation function. The model for image description is trained according to the cross entropy loss until the training stop condition is reached, and a trained model for image description is obtained, which is a semantic feature extraction model.

[0068] In this embodiment, after obtaining the target video to be published on the video platform, the semantic features of the target video are accurately extracted to obtain matching video semantic features. In this way, the same type of video as the target video can be quickly and accurately obtained from the video platform based on the video semantic features of the target video, effectively avoiding the situation where users are not interested in the target video after the target video is published during the publishing period, ensuring the playback volume of the target video during the publishing period, thereby improving the accuracy of estimating the publishing period of the target video.

[0069] Step S204: determine a first video set in the platform video library that is similar to the target video.

[0070] Among them, the platform video library includes multiple platform videos that have been published, and the first video set includes at least one platform video of the same video type as the target video. It can be understood that each platform video in the first video set is a published video. Optionally, each platform video in the first video set is published through a corresponding account, and the first video set includes platform videos published through the target account and platform videos published through other accounts, and the other accounts are different from the target account; optionally, the platform videos in the first video set do not include platform videos published through the target account, that is, they include platform videos published through at least one other account, without specific limitation.

[0071] Optionally, the computer device obtains multiple platform videos in the platform video library, and selects at least one platform video similar to the target video from the multiple platform videos, and determines the first video set based on the at least one platform video similar to the target video.

[0072] Exemplarily, after acquiring multiple platform videos in the platform video library, the computer device once screens out platform videos posted by other accounts from the multiple platform videos, and secondarily screens out at least one platform video similar to the target video from the once screened platform videos, and determines the first video set based on the at least one second screened platform video. After the platform videos are screened once, the processing amount of the subsequent second screened platform videos can be reduced, and the accuracy of the second screening can also be ensured.

[0073] Step S206, determining the statistical information on the video platform side according to the video popularity related information in each preset time period and the video views of each video in the first video set in each preset time period.

[0074] Among them, the preset time period is a time period of fixed duration, each preset time period is a plurality of continuous time periods of the same duration within the target time period, the duration of the target time period is greater than or equal to the duration of the preset time period, and after determining the target time period and the fixed duration, starting from the first preset time period of the target time period, a preset number of preset time periods are obtained in sequence, and the preset number is the ratio between the target time period and the fixed duration. For example, if the target time period is one day (i.e., 24 hours) and the fixed duration is 1 hour, there are 24 preset time periods, which are the first preset time period (0 o'clock to 1 o'clock), the second preset time period (1 o'clock to 2 o'clock), ..., the 24th preset time period (23 o'clock to 24 o'clock); for another example, if the target time period is 1 hour (60 minutes) and the fixed duration is 4 minutes, there are 15 preset time periods, which are the first preset time period (0th minute to 4th minute), the second preset time period (4th minute to 8th minute), ..., the 15th preset time period (56th minute to 60th minute).

[0075] The popularity of a video reflects the level of attention paid to the video. The information related to the popularity of the video in each preset period reflects the release of the video and the online status of the account in the preset period. It is understandable that the more videos are released in the preset period, the more videos will be accumulated in the video recommendation pool, and the user's attention will be distracted. For each creator, the more videos are released in the preset period, the easier it is for the videos released by the creator to be covered up, and the exposure of the video will be small. The more accounts are online in the preset period, the more it means that the preset period is an online peak period, and the fewer accounts are online in the preset period, the more it means that the preset period is an online trough period. Obviously, the exposure of videos in the online peak period is greater than the exposure of videos in the online trough period. Therefore, the information related to the popularity of the video in each preset period reflects the exposure of the video. The higher the exposure, the higher the attention paid to the video.

[0076] For each video, the higher the video views within the preset time period, the more suitable the video is for release within the preset time period. For example, for food-related videos, if the food-related videos have the highest views within the preset time period from 22:00 to 23:00, and the food-related videos have the lowest views within the preset time period from 9:00 to 10:00 in the morning, it means that for the preset time periods from 22:00 to 23:00 and from 9:00 to 10:00, the food-related videos are suitable for release from 10:00 to 11:00 at night, but not from 9:00 to 10:00 in the morning.

[0077] The statistical information on the video platform side includes statistical information on multiple dimensions of the platform side. Optionally, the statistical information on the video platform side includes statistical information on the platform video publishing dimension, statistical information on the account online dimension, and statistical information on the video browsing dimension. Based on this, the statistical information on the video platform side can reflect the audience groups of different types of videos in different time periods.

[0078] Optionally, the computer device determines multiple consecutive preset time periods of the same duration and obtains historical statistical data of the video platform, and determines, based on the historical statistical data of the video platform, information related to the video popularity of the video platform in each preset time period and the number of video views of each video in the first video set in each preset time period, wherein the information related to the video popularity of the video platform in each preset time period includes the number of videos released by the video platform in each preset time period and the number of online accounts.

[0079] The computer device determines the statistical information of the video platform in the dimension of video publishing and the statistical information of the account online in each preset time period according to the number of videos published and the number of accounts online in each preset time period. The computer device determines the statistical information of the video platform in the dimension of video browsing in each preset time period according to the number of video views of each video in the first video set in each preset time period. The computer device determines the statistical information of the video platform in the dimension of video publishing, the statistical information of the account online dimension and the statistical information of the video browsing dimension in each preset time period as the statistical information on the video platform side.

[0080] In some embodiments, Figure 4 FIG. 1 is a flow chart of a step of determining statistical information on the video platform side in an embodiment, wherein the statistical information on the video platform side is determined according to the video popularity related information in each preset time period and the video views of each video in the first video set in each preset time period, including:

[0081] Step S402, counting the number of videos published on the video platform in each preset time period.

[0082] Optionally, the computer device obtains historical statistical data of the video platform; wherein the historical statistical data includes target historical statistical data of multiple continuous target time periods of the same duration, each target time period includes multiple continuous preset time periods of the same duration, and each target historical statistical data includes the number of historical videos in each preset time period. For each preset time period, the computer device obtains the number of historical videos in the preset time period from each target historical statistical data, and determines the number of videos published on the video platform in the preset time period based on the number of each historical video.

[0083] Exemplarily, after determining the number of each historical video, the computer device uses the average of the number of historical videos as the number of videos published on the video platform during the preset period. Alternatively, the computer device counts the frequency of occurrence of each historical video number, and determines the number of historical videos corresponding to the highest frequency as the number of videos published on the video platform during the preset period.

[0084] Exemplarily, the computer device determines the current target period and determines the historical period before the current target period. The historical period includes multiple continuous target periods of the same duration, and the historical statistical data includes target historical statistical data of multiple continuous target periods of the same duration, such as the target period is 1 day, and there are 24 preset time periods. If the current target period is April 1, the historical period refers to the period within the previous three months (i.e., January 1 to March 31, a total of 90 days), and the historical statistical data includes target historical statistical data for each day in the previous three months. For the preset period from 0 to 1 o'clock, the average of the number of videos published on the video platform in the historical period is counted, and the average is used as the number of videos published on the video platform in the preset period from 0 to 1 o'clock.

[0085] Step S404, counting the number of accounts active on the video platform in each preset time period.

[0086] Among them, the account that is active on the video platform during the preset period refers to the account that meets the active condition during the preset period. Optionally, the active condition is that the online time of the account during the preset period reaches the preset time. If the online time of the account during the preset period reaches the preset time, the account is determined to be an account that meets the active condition. Optionally, the active condition is that the number of interactive behaviors of the account during the preset period reaches the preset number. If the number of interactive behaviors of the account during the preset period reaches the preset number, the account is determined to be an account that meets the active condition. Optionally, the active condition is that the online time of the account during the preset period reaches the preset time, and the number of interactive behaviors of the account during the preset period reaches the preset number. If the online time of the account during the preset period reaches the preset time, and the number of interactive behaviors reaches the preset number, the account is determined to be an account that meets the active condition. The interactive behavior can be the behavior of liking a video through an account, the behavior of commenting on a video through an account, or the behavior of collecting a video through an account, without specific limitation. Accounts that are active on video platforms are more inclined to play videos. Therefore, we use active conditions to filter accounts that are active on video platforms to avoid accounts that mistakenly enter the client and have no tendency to play videos from being counted, thereby ensuring the effectiveness of counting the number of accounts that are active on video platforms.

[0087] It is understandable that accounts that meet the active conditions must be online accounts, and accounts that meet the active conditions can be regarded as active accounts, which are valid online accounts. Therefore, the number of accounts active on the video platform during a preset period of time can be regarded as the number of online accounts that are valid online accounts.

[0088] Each target historical statistical data in the historical statistical data of the video platform also includes historical online account statistical data in the video platform in each preset period. The historical online account statistical data includes each online account and the online time data and interactive behavior data of each online account.

[0089] Optionally, for each preset time period, the computer device obtains historical online account statistics for the preset time period from each target historical statistical data. For each online account in each preset time period and each historical online account statistical data, the computer device determines whether the online account satisfies the active condition based on at least one of the online duration data and interactive behavior data corresponding to the online account. The computer device counts the number of online accounts that meet the active condition in each historical online account statistical data. The computer device determines the number of accounts active on the video platform in the preset time period based on the number of online accounts corresponding to the preset time period.

[0090] Exemplarily, if the active condition is that the online duration reaches a preset duration, the above step of determining whether the online account satisfies the active condition includes: for each preset time period and each online account, the computer device determines the online duration of the online account in the preset time period. If the online duration is greater than or equal to the preset duration, the online account is determined to be an account that satisfies the active condition.

[0091] If the active condition is that the number of interactive behaviors reaches a preset number, the above step of determining whether the online account meets the active condition includes: for each preset time period and each online account, the computer device determines the number of interactive behaviors of the online account in the preset time period. If the number of interactive behaviors is greater than or equal to the preset number, the online account is determined to be an account that meets the active condition.

[0092] If the active condition is that the online time reaches a preset time and the number of interactive behaviors reaches a preset number, the above step of determining whether the online account meets the active condition includes: for each preset time period and each online account, the computer device determines the online time and the number of interactive behaviors of the online account in the preset time period; if the online time is greater than or equal to the preset time and the number of interactive behaviors is greater than or equal to the preset number, the online account is determined to be an account that meets the active condition.

[0093] Exemplarily, for each preset period, after determining the number of online accounts corresponding to each historical online account statistical data of the preset period, the computer device uses the average of the number of online accounts corresponding to the preset period as the number of accounts active on the video platform during the preset period. For example, the target period is 1 day, and there are 24 preset periods. If the current target period is April 1, the historical period refers to the period within the previous three months (i.e., January 1 to March 31, a total of 90 days). For each of the previous 90 days, taking January 1 as an example, obtain the historical online account statistical data 1 of the preset period from 0:00 to 1:00 on January 1, and the historical online account statistical data 1 records the online time data and interactive behavior data of each online account from online account 1 to online account 10. For online account 1, according to at least one of the online time data and interactive behavior data of online account 1, determine whether online account 1 meets the active condition within the preset period from 0:00 to 1:00, and so on, until it is determined whether online account 10 meets the active condition within the preset period from 0:00 to 1:00. The computer device counts the number of online accounts that meet the active condition in the preset time period from 0:00 to 1:00 on January 1. Using similar steps as above, the number of online accounts in the preset time period from 0:00 to 1:00 every day in the previous 90 days is obtained, such as M1, M2, ..., M90; at this time, the computer device calculates the average of M1, M2, ..., M90 Computer equipment will mean The number of accounts active on the video platform during the preset time period from 0:00 to 1:00 is used. For other preset time periods, similar steps to the above-mentioned determination of the preset time period from 0:00 to 1:00 are adopted to count the number of accounts active on the video platform during each preset time period.

[0094] Step S406: counting the video views of each video in the first video set in each preset time period.

[0095] Among them, each target historical statistical data in the historical statistical data of the video platform also includes the video views of each video in the video platform in each preset time period. The video views reflect the number of times the video has been viewed.

[0096] Optionally, the computer device determines the number of video views of each video in the video platform in each preset time period based on historical statistical data of the video platform, and the computer device filters out the number of video views of each video belonging to the first video set in each preset time period based on the number of video views of each video in the video platform in each preset time period.

[0097] Step S408, the number of videos published on the video platform in each preset time period, the number of accounts active on the video platform in each preset time period, and the number of video views of each video in the first video set in each preset time period are counted as statistical information on the video platform side.

[0098] Among them, the amount of video data published on the video platform in each preset time period can be regarded as statistical information of the platform video publishing dimension; the number of accounts active on the video platform in each preset time period can be regarded as statistical information of the account online dimension; the video views of each video in the first video set in each preset time period can be regarded as statistical information of the video browsing dimension. Therefore, through the platform video publishing dimension, account online dimension and video browsing dimension, it is possible to fully reflect the statistical information of the video platform.

[0099] In this embodiment, by counting the number of videos published on the video platform in each preset time period in the dimension of counting platform video publishing, counting the number of accounts active on the video platform in each preset time period in the dimension of counting account online, and counting the number of video views of each video in the first video set in each preset time period in the dimension of counting video browsing, informative and comprehensive video platform-side statistical information is obtained. Based on this, the video platform-side statistical information can intuitively reflect the audience groups of different types of videos in different time periods, which is conducive to the subsequent accurate estimation of the release time period of the target video.

[0100] Step S208: output the estimated playback volume of the target video in each preset time period based on the video semantic features of the target video and the statistical information on the video platform side.

[0101] Among them, the playback volume reflects the number of times the video has been played.

[0102] Optionally, the computer device calls the playback volume estimation model, and determines the embedding result through the embedding layer of the playback volume estimation model according to the video semantic features of the target video and the statistical information on the video platform side. The estimated playback volume of the target video in each preset time period is determined through the encoder of the playback volume estimation model according to the embedding result, and the estimated playback volume of the target video in each preset time period is output. Among them, the playback volume estimation model is used to estimate the playback volume of the video in each preset time period.

[0103] The above embedding results are obtained by integrating the information of statistical information of different dimensions and the information of video semantic features, ensuring that the estimated playback volume of each preset time period can be estimated based on the embedding results with rich information. The playback volume estimation model is a trained model. The encoder of the playback volume estimation model is used for feature learning, and learns the complex relationship between features through different self-attention layers to make accurate playback volume estimates.

[0104] Step S210, determining the release period of the target video according to the estimated playback volume of the target video in each preset period.

[0105] Optionally, the computer device sorts the estimated playback volume of each preset time period from large to small, and sorts by playback volume, selects at least one preset time period with a higher playback volume sequence number from the multiple preset time periods, and determines the release time period of the target video based on the selected at least one preset time period. The higher the playback volume sequence number, the greater the corresponding estimated playback volume.

[0106] It can be understood that the estimated playback volume of at least one selected preset time period is greater than the estimated playback volume of each unselected preset time period.

[0107] Exemplarily, the computer device directly selects the preset time period corresponding to the first playback volume sequence number from the plurality of preset time periods as the release time period of the target video. The first playback volume sequence number means the largest playback volume.

[0108] Exemplarily, the computer device selects multiple preset time periods corresponding to the front number of playback volume from multiple preset time periods, and filters out the preset time periods that match the creator's publishing habits from the selected multiple preset time periods, and uses the filtered preset time periods as the publishing time periods of the target video.

[0109] In this example, the creator is the user who creates the target video, and the creator's publishing habits refer to the creator setting corresponding publishing time periods for different types of videos.

[0110] In the above-mentioned release period estimation method, by obtaining the video semantic features of the target video to be released to the video platform; determining the first video set similar to the target video in the platform video library, that is, determining the first video set that has been released and is of the same type as the target video from the platform video library. According to the video platform's video popularity related information in each preset period and the video views of each video in the first video set in each preset period, the video platform side statistical information is determined. In this way, the video platform side statistical information can more comprehensively reflect the user's attention to the same type of video in different time periods, thereby intuitively reflecting the audience groups of the same type of video in different time periods. Based on this, based on the video semantic features of the target video and the video platform side statistical information, it can be considered that videos of different types of content have different audience groups and the release periods of other videos of the same type of content in different time periods, so that the playback volume of each preset period can be accurately estimated, and the estimated playback volume of the target video in each preset period can be output; therefore, according to the estimated playback volume of the target video in each preset period, the release period of the target video can be more accurately determined, which improves the accuracy of estimating the release period of the target video.

[0111] In some embodiments, a target video is to be released through a target account; determining a first video set in a platform video library that is similar to the target video comprises: obtaining platform videos released through other accounts from the platform video library; the other accounts are accounts other than the target account; performing semantic feature extraction on the obtained platform videos to obtain video semantic features of each platform video; screening out similar videos of the target video from the obtained platform videos based on the similarity between the video semantic features of the target video and the video semantic features of each platform video; and obtaining a first video set based on the screened similar videos.

[0112] The target account refers to the account of the creator of the target video. The higher the similarity between the video semantic features of any two videos, the more similar the two videos are. In this embodiment, in order to distinguish whether the publisher of the video in the platform video library is the creator, the videos published by other accounts are regarded as platform videos. The video type of the similar video is the same as the video type of the target video.

[0113] Optionally, the computer device obtains platform videos published through other accounts from the platform video library. For each platform video, the computer device extracts semantic features of the platform video through a semantic feature extraction model to obtain video semantic features of the platform video. For any platform video, the computer device calculates the similarity between the video semantic features of the target video and the video semantic features of the any platform video. The computer device filters out a preset number of platform videos from multiple platform videos based on each similarity, and uses the filtered platform videos as similar videos. The computer device determines a first video set based on the multiple similar videos.

[0114] Exemplarily, after determining the video semantic features of each platform video, for any platform video, the computer device calculates the similarity between the video semantic features of the target video and the video semantic features of any platform video through cosine distance calculation based on the video semantic features of the target video and the video semantic features of the any platform video.

[0115] For example, let the video semantic features of the target video and the video semantic features of any platform video be x i and x j At this time, the similarity s of the two videos is calculated by the following cosine distance formula ij :

[0116]

[0117] in, Represents x i‖.‖ is the L2 norm calculation, which is used to determine the modulus of the video semantic feature. Of course, the similarity between the video semantic feature of the target video and the video semantic feature of any platform video can also be determined by the Euclidean distance.

[0118] After determining the similarity between the video semantic features of the target video and the video semantic features of any platform video, the computer device calculates the similarity between the video semantic features of the target video and the video semantic features of each platform video, and the computer device sorts the corresponding platform videos according to each similarity. The smaller the sequence number of the platform video, the greater the similarity between the platform video and the target video. The computer device selects a preset number of platform videos with high rankings according to the sorting of each platform video, and determines the first video set based on the selected multiple platform videos. For example, the first video set is determined based on the Top-K (top K ranked) platform videos.

[0119] In this embodiment, after extracting the video semantic features of the platform videos posted by other accounts, similar videos similar to the target video can be screened out from multiple platform videos according to the similarity between the video semantic features of the target video and the video semantic features of each platform video, so as to determine the first video set. That is, selecting platform videos of the same video type as the target video can effectively avoid users being uninterested in the target video after it is posted during the posting period, thereby ensuring the playback volume of the target video during the posting period, and thus improving the accuracy of estimating the posting period for the target video.

[0120] In some embodiments, Figure 5 FIG. 1 is a flow chart of a step of determining the estimated playback volume in one embodiment. Based on the video semantic features of the target video and the statistical information on the video platform, the estimated playback volume of the target video in each preset time period is output, including:

[0121] Step S502, generating a first information type code according to the first information type, and generating a second information type code according to the second information type, wherein the first information type is used to represent the information type of the semantic feature of the video, and the second information type is used to represent the information type of the statistical information on the video platform side.

[0122] The first information type reflects information at the feature level, and the second information type reflects information at the video platform level. The first information type and the second information type are used to distinguish information types, ensuring that the playback volume estimation model can learn information of different information types.

[0123] Optionally, the computer device determines an information type, the information type includes a first information type for characterizing a semantic feature of the video and a second information type for characterizing statistical information on the video platform side. The computer device generates a corresponding first information type code according to the first information type, and generates a corresponding second information type code according to the second information type.

[0124] Exemplarily, the computer device encodes a first information type corresponding to a first information type to which the video semantic feature of the target video belongs as code A, and encodes a second information type corresponding to a second information type to which the statistical information on the video platform side belongs as code B.

[0125] Step S504, generating a first vector type code according to the vector type corresponding to the first type statistical vector, and generating a first type statistical vector on the platform side according to the number of videos published on the video platform in each preset time period.

[0126] Among them, the vector type encoding is an encoding based on the corresponding vector type, and the first vector type is the vector type of the first type of statistical vector, which reflects the number of videos released on the video platform in each preset time period. The first type of statistical vector on the platform side is a first type of statistical vector that integrates the video release status of multiple accounts on the video platform, which can be understood as the first type of statistical vector on the video platform side. The first type of statistical vector on the platform side is a multidimensional vector, and each dimension represents a preset time period. For example, the preset time period from 0 o'clock to 1 o'clock, the preset time period from 1 o'clock to 2 o'clock,..., and the preset time period from 23 o'clock to 24 o'clock represent the first dimension, the second dimension,..., and the 24th dimension respectively.

[0127] Optionally, for each preset time period, the computer device determines the first type of platform-side value corresponding to the preset time period according to the number of videos published on the video platform during the preset time period. The computer device determines the first type of platform-side statistical vector according to the first type of platform-side value corresponding to each preset time period. The computer device generates a first vector type code corresponding to the first type of vector type. The first type of platform-side value corresponding to the preset time period represents the value of the dimension represented by the preset time period in the first type of platform-side statistical vector.

[0128] Since the numerical value of the number of videos is relatively large, in order to reduce the amount of calculation for estimating the release time period, exemplarily, for each preset time period, after determining the number of videos released on the video platform during the preset time period, the computer device performs a logarithmic operation on the number of videos released on the video platform during the preset time period, obtains the corresponding first operation result on the platform side, and uses the corresponding first operation result on the platform side as the first category of numerical value on the platform side corresponding to the preset time period.

[0129] For example, in the preset time period from 0:00 to 1:00, the number of videos posted on the video platform is 100, then a logarithm operation with a base of 10 is performed on 100 to obtain the first operation result log corresponding to the preset time period from 0:00 to 1:00 on the platform side. 10 (100)=2, that is, the first-category value on the platform side of the first dimension in the first-category statistical vector on the platform side is 2.

[0130] For example, the first vector type code corresponding to the first type of vector type is regarded as code 0.

[0131] Step S506, generating a second vector type code according to the vector type corresponding to the second type statistical vector, and generating a second type statistical vector on the platform side according to the number of accounts active on the video platform in each preset time period.

[0132] Among them, the second vector type is the vector type of the second type of statistical vector, and the second type of statistical vector reflects the number of accounts active on the video platform in each preset time period. The second type of statistical vector on the platform side is a second type of statistical vector that integrates the online status of multiple accounts in the video platform, which can be understood as the second type of statistical vector on the video platform side. The second type of statistical vector on the platform side is a multi-dimensional vector, and each dimension represents a preset time period.

[0133] Optionally, for each preset time period, the computer device determines the second type of platform-side value corresponding to the preset time period according to the number of accounts active on the video platform during the preset time period. The computer device determines the second type of platform-side statistical vector according to the second type of platform-side value corresponding to each preset time period. The second type of platform-side value of the dimension represented by the preset time period represents the value of the dimension represented by the preset time period in the second type of platform-side statistical vector.

[0134] Since a large number of accounts are active on the video platform in each preset time period, in order to reduce the amount of calculation for estimating the release time period, for each preset time period, after determining the number of accounts active on the video platform in the preset time period, the computer device performs a logarithmic operation on the number of accounts active on the video platform in the preset time period to obtain the corresponding platform-side second operation result, and uses the corresponding platform-side second operation result as the platform-side second category value corresponding to the preset time period.

[0135] For example, in the preset time period from 0:00 to 1:00, the number of accounts active on the video platform is 1000, then a logarithm operation with a base of 10 is performed on 1000 to obtain the second operation result log corresponding to the preset time period from 0:00 to 1:00 on the platform side. 10 (1000)=3, that is, the second-category value on the platform side of the first dimension in the second-category statistical vector on the platform side is 3.

[0136] For example, the second vector type code corresponding to the second category vector type is regarded as code 1.

[0137] Step S508, generating a third vector type code according to the vector type corresponding to the third type statistical vector, and generating a third type statistical vector on the platform side according to the statistical video views of each video in the first video set in each preset time period.

[0138] Among them, the third vector type is the vector type of the third type of statistical vector, and the third type of statistical vector reflects the video browsing situation in each preset time period. Each video in the first video set has a unique corresponding third type of statistical vector on the platform side, that is, the third type of statistical vector on the platform side corresponding to each video in the first video set reflects the browsing situation of each video in the first video set in each preset time period. The third type of statistical vector on the platform side can be understood as the third type of statistical vector on the video platform side. The third type of statistical vector on the platform side is a multi-dimensional vector, and each dimension represents a preset time period.

[0139] Optionally, for each preset time period, the computer device determines the third type of platform-side numerical value corresponding to the video in the first video set in the preset time period according to the video views of each video in the first video set in the preset time period. For each video in the first video set, the computer device determines the third type of platform-side statistical vector corresponding to the video in the first video set according to the third type of platform-side numerical value corresponding to the video in each preset time period. The third type of platform-side numerical value corresponding to the video in the first video set in the preset time period represents the numerical value of the dimension represented by the preset time period in the third type of platform-side statistical vector corresponding to the video in the first video set.

[0140] Since the video views of each video in the first video set in each preset time period are relatively large, in order to facilitate subsequent calculations, for each preset time period, after determining the video views of each video in the first video set in the preset time period, the computer device performs a logarithmic operation on the video views of each video in the first video set in the preset time period to obtain a corresponding platform-side third operation result, and determines the corresponding platform-side third operation result as the platform-side third category value corresponding to the video in the first video set in the preset time period.

[0141] For example, if the number of views of video 1 in the first video set from 0:00 to 1:00 is 10,000, then a logarithm operation with a base of 10 is performed on 10,000 to obtain the corresponding third operation result log on the platform side. 10 (10000)=4, that is, the platform side third category value of the video 1 in the preset time period from 0 o'clock to 1 o'clock is 4, that is, the platform side third category value of the first dimension in the platform side third category statistical vector is 4.

[0142] For example, the third vector type encoding corresponding to the third category of vector types is regarded as encoding 2.

[0143] Step S510, through the playback volume estimation model, according to the first information type code and video semantic features, the second information type code, the first vector type code and the first type of statistical vector on the platform side, the second information type code, the second vector type code and the second type of statistical vector on the platform side, and the second information type code, the third vector type code and the third type of statistical vector on the platform side, output the estimated playback volume of the target video in each preset time period.

[0144] Optionally, the computer device determines a corresponding video semantic feature vector according to the video semantic features of the target video. Through the embedding layer of the playback volume estimation model, the first information type code is converted into a vector corresponding to the first information type code, the second information type code is converted into a vector corresponding to the second information type code, the first vector type code is converted into a vector corresponding to the first vector type code, the second vector type code is converted into a vector corresponding to the second vector type code, and the third vector type code is converted into a vector corresponding to the third vector type code.

[0145] Through the encoder of the playback volume estimation model, based on the vector corresponding to the first information type encoding and the video semantic feature vector, the vector corresponding to the second information type encoding, the vector corresponding to the first vector type encoding and the first type of statistical vector on the platform side, the vector corresponding to the second information type encoding, the vector corresponding to the second vector type encoding and the second type of statistical vector on the platform side, and the vector corresponding to the second information type encoding, the vector corresponding to the third vector type encoding and the third type of statistical vector on the platform side of each video in the first video set, the estimated playback volume of the target video in each preset time period is obtained, and the estimated playback volume of the target video in each preset time period is output.

[0146] The embedding layer of the above-mentioned playback volume estimation model is used to express discrete variables (i.e., encodings) as continuous vectors. The vector corresponding to the first information type encoding and the vector corresponding to the second information type encoding can help the playback volume estimation model determine whether the information belongs to semantic feature information or statistical information on the video platform side. The vector corresponding to the first vector type encoding, the vector corresponding to the second vector type encoding, and the vector corresponding to the third vector type encoding can help the playback volume estimation model distinguish different types of statistical vectors.

[0147] The dimensions of the vector corresponding to the first information type encoding, the video semantic feature vector, the vector corresponding to the second information type encoding, the vector corresponding to the first vector type encoding, the first type of statistical vector on the platform side, the vector corresponding to the second vector type encoding, the second type of statistical vector on the platform side, the vector corresponding to the third vector type encoding and the third type of statistical vector on the platform side of each video in the first video set are all the same, which is the ratio between the target time period and the fixed duration of the preset time period. For example, if the target time period is 1 day and the fixed duration of the preset time period is 1 hour, then the dimension is 24.

[0148] Furthermore, the embedding layer of the playback volume estimation model includes a first embedding layer and a second embedding layer, the first embedding layer is used to determine the vector corresponding to the information type encoding, and the second embedding layer is used to determine the vector corresponding to the vector type encoding.

[0149] Through the first embedding layer of the playback volume estimation model, the vector corresponding to the first information type encoding and the vector corresponding to the second information type encoding are determined. Through the second embedding layer of the playback volume estimation model, the vector corresponding to the first vector type encoding, the vector corresponding to the second vector type encoding and the vector corresponding to the third vector type encoding are determined.

[0150] Exemplarily, let the first information type code be code A, let the second information type code be code B; let the first vector type code be code 0, let the second vector type code be code 1, and let the third vector type code be code 2; based on this, the vector corresponding to the first information type code (i.e., code A) is E(A), the vector corresponding to the second information type code (i.e., code B) is E(B), the vector corresponding to the first vector type code (i.e., code 0) is E(0), the vector corresponding to the second vector type code (i.e., code 1) is E(1), and the vector corresponding to the third vector type code (i.e., code 2) is E(2).

[0151] At this time, the encoder of the playback volume estimation model determines the estimated playback volume of the target video in each preset time period based on E(A) and the video semantic feature vector, E(B), E(0) and the first type of statistical vector on the platform side, E(B), E(1) and the second type of statistical vector on the platform side, and E(B), E(2) and the third type of statistical vector on the platform side.

[0152] In this embodiment, by generating a first information type code corresponding to the first information type and generating a second information type code corresponding to the second information type, different information types are significantly distinguished by coding. According to the number of videos published on the video platform in each preset time period, the number of accounts active on the video platform in each preset time period, and the video views of each video in the first video set in each preset time period, the corresponding platform side different types of statistical vectors (the first type of statistical vector on the platform side, the second type of statistical vector on the platform side, and the third type of statistical vector on the platform side) are determined respectively, and vector type codes that can significantly distinguish different types of statistical vectors on the platform side are generated. In this way, through the playback volume estimation model, according to the video semantic features and the information type code to which it belongs, the statistical vector of each type on the platform side, the corresponding information type code and the vector type code to which it belongs are comprehensively considered from the information dimension, the statistical vector type dimension and the vector dimension, so as to accurately estimate the playback volume of each preset time period, thereby improving the accuracy of estimating the release period of the target video.

[0153] In some embodiments, Figure 6 As shown, it is a flowchart of the estimated playback volume determination step in another embodiment. Through the playback volume estimation model, according to the first information type code and the video semantic feature, the second information type code, the first vector type code and the first type of statistical vector on the platform side, the second information type code, the second vector type code and the second type of statistical vector on the platform side, and the second information type code, the third vector type code and the third type of statistical vector on the platform side, the estimated playback volume of the target video in each preset time period is output, including:

[0154] Step S602: Obtain video side input data according to the first information type code and the video semantic features of the target video.

[0155] Optionally, the computer device determines a vector corresponding to the first information type code and a video semantic feature vector corresponding to the video semantic feature of the target video. The computer device fuses the vector corresponding to the first information type code and the video semantic feature vector corresponding to the video semantic feature of the target video to obtain video side input data.

[0156] Exemplarily, the computer device superimposes the vector corresponding to the first information type code and the video semantic feature vector corresponding to the video semantic feature of the target video to obtain the video side input data. For example, let the first information type code be code A, and the vector corresponding to the first information type code (i.e., code A) be E(A), then the video side input data is E(A)+video semantic feature vector. Among them, the vector corresponding to the first information type code and the video semantic feature vector corresponding to the video semantic feature of the target video have the same dimension.

[0157] Step S604, obtaining the first dimension input data on the platform side according to the second information type code, the first vector type code and the first type of statistical vector on the platform side.

[0158] Optionally, the computer device determines a vector corresponding to the second information type code and determines a vector corresponding to the first vector type code. The computer device fuses the vector corresponding to the second information type code, the vector corresponding to the first vector type code, and the first type of statistical vector on the platform side to obtain the first dimensional input data on the platform side.

[0159] Exemplarily, a computer device determines a first weight of a vector corresponding to an information type code and a second weight of a vector corresponding to a vector type code. If the information type code includes a first information type code, the weight of the vector corresponding to the second information type code is the first weight. If the vector type code includes a first vector type code, the weight of the vector corresponding to the first vector type code is the second weight.

[0160] The computer device performs weighted summation processing on the vector corresponding to the second information type code and the vector corresponding to the first vector type code according to the first weight and the second weight to obtain the first-dimensional weighted data on the platform side. The computer device superimposes the first-dimensional weighted data on the platform side with the first-category statistical vector on the platform side to obtain the first-dimensional input data on the platform side.

[0161] Among them, the first weight and the second weight are both less than or equal to the value 1, and the first weight and the second weight can be determined according to actual needs. For example, the computer device determines the first weight and the second weight according to actual needs. If the actual demand is to pay more attention to the details of information type encoding, the first weight is set to be greater than the second weight, and the first weight is less than or equal to the value 1. If the actual demand is to pay more attention to the details of vector type encoding, the first weight is set to be less than the second weight, and the second weight is less than or equal to the value 1. If the actual demand is to pay equal attention to the details of information type encoding and the details of vector type encoding, the first weight is set to be equal to the second weight, such as setting the first weight and the second weight to be both 1. At this time, the vector corresponding to the second information type encoding, the vector corresponding to the first vector type encoding, and the first type of statistical vector on the platform side are directly superimposed to obtain the first-dimensional input data on the platform side. For example: let the second information type code be code B, and the vector corresponding to the second information type code B be E(B); let the first vector type code be regarded as code 0, and the vector corresponding to the first vector type code 0 be E(0); the first weight and the second weight are L1 and L2 respectively, then the first dimension input data on the platform side is E(B)×L1+E(0)×L2+the first type statistical vector on the platform side.

[0162] Step S606, obtaining the second dimensional input data on the platform side according to the second information type code, the second vector type code and the second type of statistical vector on the platform side.

[0163] Optionally, the computer device determines a vector corresponding to the second information type code and a vector corresponding to the second vector type code, and the computer device fuses the vector corresponding to the second information type code, the vector corresponding to the second vector type code and the second type of statistical vector on the platform side to obtain the second dimensional input data on the platform side.

[0164] Wherein, the vector type code also includes a second vector type code, and the weight of the vector corresponding to the second vector type is the second weight.

[0165] Exemplarily, the computer device performs weighted summation processing on the vector corresponding to the second information type code and the vector corresponding to the second vector type code according to the first weight and the second weight to obtain the second-dimensional weighted data on the platform side. The computer device superimposes the second-dimensional weighted data on the platform side with the second type of statistical vector on the platform side to obtain the second-dimensional input data on the platform side.

[0166] Similarly, the computer device determines the first weight and the second weight according to actual needs. If the actual need is to pay more attention to the details of the information type encoding, the first weight is set to be greater than the second weight, and the first weight is less than or equal to the value 1. If the actual need is to pay more attention to the details of the vector type encoding, the first weight is set to be less than the second weight, and the second weight is less than or equal to the value 1. If the actual need is to pay equal attention to the details of the information type encoding and the details of the vector type encoding, the first weight is set to be equal to the second weight, such as setting the first weight and the second weight to be both 1. At this time, the vector corresponding to the second information type encoding, the vector corresponding to the second vector type encoding, and the second type of statistical vector on the platform side are directly superimposed to obtain the second dimensional input data on the platform side. For example: let the second information type encoding be encoding B, and the vector corresponding to the second information type encoding B is E(B); let the second vector type encoding be regarded as encoding 1, and the vector corresponding to the second vector type encoding 1 is E(1). The first weight and the second weight are L1 and L2 respectively, and the second dimensional input data on the platform side is E(B)×L1+E(1)×L2+the second type of statistical vector on the platform side.

[0167] Step S608, obtaining the third-dimensional input data on the platform side according to the second information type code, the third vector type code and the third type of statistical vector on the platform side.

[0168] Optionally, for each video in the first video set, the computer device determines a vector corresponding to the second information type code and determines a vector corresponding to the third vector type code, and the computer device fuses the vector corresponding to the second information type code, the vector corresponding to the third vector type code, and the third type of statistical vector on the platform side corresponding to the video to obtain third-dimensional input data on the platform side corresponding to the video.

[0169] Wherein, the vector type code also includes a third vector type code, and the weight of the vector corresponding to the second vector type is the second weight.

[0170] Exemplarily, for each video in the first video set, the computer device performs weighted summation processing on the vector corresponding to the second information type code and the vector corresponding to the third vector type code according to the first weight and the second weight to obtain the third-dimensional weighted data on the platform side. The computer device superimposes the third-dimensional weighted data on the platform side with the third-category statistical vector on the platform side corresponding to the video to obtain the third-dimensional input data on the platform side corresponding to the video.

[0171] Similarly, the computer device determines the first weight and the second weight according to actual needs. If the actual need is to pay more attention to the details of the information type encoding, the first weight is set to be greater than the second weight, and the first weight is less than or equal to the value 1. If the actual need is to pay more attention to the details of the vector type encoding, the first weight is set to be less than the second weight, and the second weight is less than or equal to the value 1. If the actual need is to pay equal attention to the details of the information type encoding and the details of the vector type encoding, the first weight is set to be equal to the second weight, such as setting the first weight and the second weight to be both 1. At this time, the vector corresponding to the second information type encoding, the vector corresponding to the third vector type encoding, and the third type of statistical vector on the platform side corresponding to the video are directly superimposed to obtain the third-dimensional input data on the platform side corresponding to the video. For example: for video 1 in the first video set, let the second information type code be code B, and the vector corresponding to the second information type code B be E(B); let the third vector type code be code 2, and the vector corresponding to the third vector type code 2 be E(2). The first weight and the second weight are L1 and L2 respectively. Then, the third-dimensional input data on the platform side corresponding to video 1 in the first video set is E(B)×L1+E(2)×L2+the third-type statistical vector on the platform side corresponding to video 1 in the first video set.

[0172] Step S610, outputting the estimated playback volume of the target video in each preset time period through the playback volume estimation model according to the video side input data, the platform side first dimension input data, the platform side second dimension input data and the platform side third dimension input data.

[0173] Among them, the video side input data, the first dimension input data on the platform side, the second dimension input data on the platform side and the third dimension input data on the platform side corresponding to each video in the first video set can all be regarded as vectors with the same dimension, and the dimension is the ratio between the fixed duration of the target time period and the preset time period. For example, if the target time period is 1 day and the fixed duration of the preset time period is 1 hour, then the dimension is 24 dimensions.

[0174] Optionally, the computer device combines the video side input data, the platform side first dimension input data, the platform side second dimension input data and the platform side third dimension input data corresponding to each video in the first video set to obtain the target input data. The computer device obtains the estimated playback volume of the target video in each preset time period through the encoder of the playback volume estimation model, and outputs the estimated playback volume of the target video in each preset time period.

[0175] In this embodiment, accurate video side input data is obtained by combining the information type coding and the video semantic features of the target video, and integrating the information type dimension corresponding to the video semantic features and the feature dimension of the video semantic features. For various statistical vectors on the platform side, the platform side input data corresponding to various statistical vectors on the platform side (such as the first type of statistical vectors on the platform side, the second type of statistical vectors on the platform side, and the third type of statistical vectors on the platform side) can be accurately obtained by integrating the platform side dimensions, the vector type dimensions of various statistical vectors, and the statistical vector dimensions of various statistical vectors. In this way, through the platform side input data and the video side input data corresponding to various statistical vectors on the platform side, the preset playback volume of each preset time period can be estimated from multiple dimensions, thereby improving the accuracy of estimating the release time period for the target video.

[0176] In some embodiments, the target video is to be released through the target account, and the method also includes: determining a second video set similar to the target video in historical videos released through the target account; determining account-side statistical information based on account popularity-related information of the target account in each preset time period and the video views of each video in the second video set in each preset time period.

[0177] Among them, each video in the second video set is a similar video similar to the target video, and each video in the second video set is released by the target account. Account popularity reflects the degree of attention to the target account. The account popularity related information of each preset time period reflects the situation of releasing videos through the target account during the preset time period and the online situation of the follower accounts that follow the target account. It is understandable that the more videos released by the target account during the preset time period, it means that the creator (that is, the account is the target account) tends to fix the preset time period as the creator's release time period, and improve the loyalty of the follower accounts that follow the target account. The more online the follower accounts are in the preset time period, the higher the exposure of the target video will be if the target video is released during the preset time period.

[0178] For each video in the second video set, the higher the video views of the video in each preset time period, the more suitable the video is for being released in the preset time period.

[0179] The account-side statistical information includes statistical information of multiple dimensions on the account side. Optionally, the account-side statistical information includes statistical information of the account publishing dimension, statistical information of the follower account online dimension, and statistical information of the video browsing dimension of the videos published by the target account. Based on this, the account-side statistical information reflects the degree of attention paid to different types of videos related to the target account in different preset time periods.

[0180] Optionally, after determining a plurality of consecutive preset time periods of the same duration and obtaining historical statistical data of the video platform, the computer device determines the target account's account historical statistical data from the video platform's historical statistical data. The computer device determines the target account's account popularity related information in each preset time period and the video views of each video in the second video set in each preset time period based on the target account's account historical statistical data. The target account's account popularity related information in each preset time period includes the number of videos posted by the target account in each preset time period and the number of online accounts following the account.

[0181] The computer device determines the statistical information of the account publishing dimension and the statistical information of the online dimension of the followed account in each preset time period according to the number of video publishings published by the target account and the number of online accounts of the followed account in each preset time period. The computer device determines the statistical information of the video browsing dimension of the videos published by the target account in each preset time period according to the video browsing volume of each video in the second video set in each preset time period, and the computer device determines the statistical information of the account publishing dimension, the statistical information of the online dimension of the followed account, and the statistical information of the video browsing dimension of the videos published by the target account in each preset time period as the account side statistical information.

[0182] In this embodiment, by determining a second video set that is similar to the target video in historical videos released by the target account, and based on the account popularity-related information of the target account in each preset time period and the video views of each video in the second video set in each preset time period, the account-side statistical information can be comprehensively and accurately determined from multiple dimensions related to the target account, which is conducive to subsequently determining the publishing time period that matches the target account based on the account-side statistical information, thereby improving the accuracy of the publishing time period.

[0183] In some embodiments, based on the video semantic features of the target video and the statistical information on the video platform side, the estimated playback volume of the target video in each preset time period is output, including: based on the video semantic features of the target video, the statistical information on the video platform side and the statistical information on the account side, the estimated playback volume of the target video in each preset time period is output.

[0184] Optionally, after the computer device executes step S206 and determines the account-side statistical information, the computer device calls the playback volume estimation model, and determines the embedding result through the embedding layer of the playback volume estimation model according to the video semantic features of the target video, the video platform-side statistical information and the account-side statistical information. The estimated playback volume of the target video in each preset time period is determined according to the embedding result through the encoder of the playback volume estimation model, and the estimated playback volume of the target video in each preset time period is output.

[0185] In this embodiment, based on the video semantic features of the target video, the video platform-side statistical information and the account-side statistical information, the estimated playback volume of the target video in each preset time period is output. In this way, considering that videos of different types of content have different audience groups in different time periods and the release time periods of other videos of the same type of content, combined with the account-side statistical information that can reflect the release habits of the target account, the fixed release time period of the target account is matched, which greatly increases the playback volume of the target video released by the target account during the release period.

[0186] In some embodiments, determining a second video set similar to a target video from historical videos posted by a target account includes: acquiring historical videos posted by the target account; performing semantic feature extraction on the acquired historical videos to obtain video semantic features of each historical video; screening out similar videos to the target video from the acquired historical videos based on similarities between the video semantic features of the target video and the video semantic features of each historical video; and obtaining a second video set based on the screened similar videos.

[0187] Optionally, the computer device obtains multiple historical videos published by the target account from the platform video library. For each historical video, the semantic feature extraction model is called to extract semantic features of the historical video to obtain video semantic features of the historical video. For each historical video, the computer device calculates the similarity between the video semantic features of the target video and the video semantic features of the historical video, and the computer device selects a preset number of historical videos from the multiple historical videos according to the similarities, and determines the second video set based on the selected historical videos.

[0188] Exemplarily, after determining the video semantic features of each historical video, for each historical video, the computer device calculates the similarity between the video semantic features of the target video and the video semantic features of the historical video by cosine distance calculation. Of course, the Euclidean distance can also be used to calculate the similarity between the video semantic features of the target video and the video semantic features of the historical video.

[0189] After determining the similarity between the video semantic features of the target video and the video semantic features of each historical video, the computer device sorts the historical videos according to the similarity between the video semantic features of the target video and the video semantic features of each historical video. The smaller the sequence number of the historical video, the greater the similarity between the historical video and the target video. The computer device selects a preset number of historical videos with high rankings according to the sorting of each historical video, and determines a second video set based on the selected multiple historical videos. For example, the second video set is determined based on the Top-K (top K ranked) historical videos. In order to ensure the effectiveness of the subsequent release period estimate, it is necessary to ensure that the number of videos in the first video set is the same as the number of videos in the second video set.

[0190] In this embodiment, based on the extracted video semantic features of each historical video posted by the target account, historical videos similar to the target video can be accurately screened to accurately determine the second video set, and then based on the second video set, account-side statistical information can be efficiently counted. In this way, based on the account-side statistical information, the fixed posting time period of the target account is matched, which greatly increases the playback volume of the target video posted by the target account during the posting time period.

[0191] In some embodiments, Figure 7 FIG. 1 is a flow chart of a step of determining account-side statistical information in an embodiment. The method further includes:

[0192] Step S702, counting the number of videos posted on the video platform by the target account in each preset time period.

[0193] Optionally, the computer device determines the target account's account historical statistics from the historical statistics of the video platform. The target account's account historical statistics include multiple target account historical statistics of consecutive target time periods of the same duration. Each target account historical statistics includes the number of target historical videos published by the target account in each preset time period. For each preset time period, the computer device obtains the number of target historical videos in the preset time period from the historical statistics of each target account, and determines the number of videos published on the video platform by the target account in the preset time period based on the number of each target historical video.

[0194] Exemplarily, after determining the number of each target historical video, the computer device uses the average of the number of multiple target historical videos as the number of videos published on the video platform by the target account during the preset period. Alternatively, the computer device counts the frequency of occurrence of each target historical video number, and determines the target historical video number corresponding to the highest frequency as the number of videos published on the video platform by the target account during the preset period.

[0195] For example, if the target period is 1 day and the fixed duration of the preset period is 1 hour, there are 24 preset periods. If the current target period is April 1, the historical period refers to the period within the previous three months (i.e., January 1 to March 31, a total of 90 days). For the preset period from 0:00 to 1:00, the average number of videos that have been posted on the video platform through the target account is counted, and the average is used as the number of videos posted on the video platform through the target account during the preset period from 0:00 to 1:00.

[0196] Step S704, counting the number of active accounts regarding videos posted by the target account in each preset time period.

[0197] As mentioned above, the accounts that follow the target account are regarded as the follow accounts. The active accounts of the videos released by the target account mentioned in this step refer to the accounts that meet the active conditions in the preset period. The active conditions are that the online time of the account in the preset period reaches the preset time, or the number of interactive behaviors of the account in the preset period reaches the preset number, or the online time of the account in the preset period reaches the preset time, and the number of interactive behaviors of the account in the preset period reaches the preset number.

[0198] Each target account historical statistical data in the target account account historical statistical data also includes historical follow-up account online statistical data for each preset period. The historical follow-up account online statistical data includes each follow-up account and each follow-up account's online duration data and interactive behavior data.

[0199] Optionally, for each preset time period, the computer device obtains the online statistics of the historical follow-up accounts for the preset time period from the historical statistics of each target account. For each follow-up account in each preset time period and each historical follow-up account online statistics, the computer device determines whether the follow-up account is an account that meets the active condition based on at least one of the online time data and interactive behavior data corresponding to the follow-up account. The computer device counts the number of follow-up accounts that meet the active condition in the online statistics of each historical follow-up account, and the computer device determines the number of active accounts for the videos posted by the target account in the preset time period based on the number of each follow-up account corresponding to the preset time period.

[0200] Exemplarily, if the active condition is that the online duration reaches a preset duration, the above step of determining whether the follow account is an account that meets the active condition includes: for each preset time period and each follow account, the computer device determines the online duration of the follow account in the preset time period. If the online duration of the follow account is greater than or equal to the preset duration, the follow account is determined to be an account that meets the active condition.

[0201] If the active condition is that the number of interactive behaviors reaches a preset number, the above step of determining whether the follow account is an account that meets the active condition includes: for each preset time period and each follow account, the computer device determines the number of interactive behaviors of the follow account in the preset time period. If the number of interactive behaviors of the follow account is greater than or equal to the preset number, the follow account is determined to be an account that meets the active condition. The interactive behavior of the follow account refers to the behavior of the follow account interacting with the videos posted by the target account.

[0202] If the active condition is that the online time reaches a preset time and the number of interactive behaviors reaches a preset number, the above step of determining whether the follow account is an account that meets the active condition includes: for each preset time period and each follow account, if the online time of the follow account is greater than or equal to the preset time, and the number of interactive behaviors of the follow account is greater than or equal to the preset number, then the follow account is determined to be an account that meets the active condition.

[0203] Exemplarily, the target period is 1 day, and there are 24 preset periods. If the current target period is April 1, the historical period refers to the period within the previous three months (i.e., January 1 to March 31, a total of 90 days). The number of follow accounts of follow accounts that meet the active condition in the preset period from 0:00 to 1:00 every day in the previous 90 days is determined in sequence. For the preset period from 0:00 to 1:00, the computer device uses the average of the number of follow accounts for 90 days as the number of active accounts for the preset period from 0:00 to 1:00 regarding videos posted by the target account.

[0204] Step S706: counting the video views of each video in the second video set in each preset time period.

[0205] Optionally, the computer device determines the number of video views of each video belonging to the second video set in each preset time period based on the number of video views of each video in the video platform in each preset time period.

[0206] Step S708, the number of videos published on the video platform in each preset time period, the number of accounts active on the video platform in each preset time period, and the number of video views of each video in the second video set in each preset time period are counted as account-side statistical information.

[0207] Among them, the number of videos published on the video platform in each preset time period can be regarded as statistical information of the account publishing dimension, the number of accounts active on the video platform in each preset time period can be regarded as statistical information of the online dimension of the followed account, and the video views of each video in the second video set in each preset time period can be regarded as statistical information of the video browsing dimension of the videos published by the target account.

[0208] In this embodiment, by counting the number of videos published on the video platform in each preset period of the account publishing dimension, counting the number of accounts active on the video platform in each preset period of the follow account online dimension, and the video views of each video in the second video set in each preset period of the video viewing dimension of the video published by the target account, it is possible to obtain information-rich and comprehensive account-side statistical information. In this way, based on accurate account-side statistical information, the fixed publishing period of the target account can be matched, greatly improving the playback volume of the target video published by the target account during the publishing period.

[0209] In some embodiments, Figure 8 FIG. 1 is a flow chart of another embodiment of determining the estimated playback volume step. Based on the video semantic features of the target video and the statistical information on the video platform side, the estimated playback volume of the target video in each preset time period is output, including:

[0210] Step S802, generate a first information type code according to the first information type, generate a second information type code according to the second information type, and generate a third information type code according to the third information type, the first information type is used to represent the information type of the semantic feature of the video, the second information type is used to represent the information type of the statistical information on the video platform side, and the third information type is used to represent the information type of the statistical information on the account side.

[0211] Among them, the third information type reflects the information on the account side. The types of information are distinguished by the first information type, the second information type and the third information type to ensure that the playback volume estimation model can learn information of different information types.

[0212] Optionally, the computer device determines the information type, which includes a first information type for characterizing the semantic features of the video, a second information type for characterizing the statistical information on the video platform side, and a third information type for characterizing the statistical information on the account side. The computer device generates a corresponding first information type code based on the first information type, and generates a corresponding second information type code based on the second information type. The computer device generates a corresponding third information type code based on the third information type.

[0213] Exemplarily, the first information type encoding corresponding to the first information type to which the video semantic features of the target video belong is encoded as code A, the second information type encoding corresponding to the second information type to which the statistical information on the video platform side belongs is encoded as code B, and the third information type encoding corresponding to the third information type to which the statistical information on the account side belongs is encoded as code C.

[0214] Step S804, generating a first vector type code according to the vector type corresponding to the first type of statistical vector, and generating a first type of statistical vector on the platform side according to the statistical number of videos published on the video platform in each preset time period, and generating a first type of statistical vector on the account side according to the statistical number of videos published on the video platform through the target account in each preset time period.

[0215] Among them, the first type of statistical vector on the account side is the first type of statistical vector that reflects the video release situation of the target account, which can be understood as the first type of statistical vector on the account side. The first type of statistical vector includes the first type of statistical vector on the platform side and the first type of statistical vector on the account side. The first type of statistical vector on the account side is a multi-dimensional vector, and each dimension represents a preset time period. The first vector type encoding and the generation steps of the first type of statistical vector on the platform side can refer to the embodiment of step S504.

[0216] Optionally, for each preset time period, the computer device determines the first category value of the account side corresponding to the preset time period according to the number of videos published on the video platform by the target account during the preset time period. The computer device determines the first category statistical vector on the account side according to the first category value of the account side corresponding to each preset time period. The first category value on the account side corresponding to the preset time period represents the value of the dimension represented by the preset time period in the first category statistical vector on the account side.

[0217] Similarly, in order to reduce the amount of calculation, the computer device performs logarithmic operation on the number of videos published on the video platform through the target account during the preset time period, obtains the corresponding account-side first operation result, and uses the corresponding account-side first operation result as the account-side first category value corresponding to the preset time period.

[0218] Step S806, generating a second vector type code according to the vector type corresponding to the second type of statistical vector, and generating a second type of statistical vector on the platform side according to the number of accounts that are active on the video platform in each preset time period, and generating a second type of statistical vector on the account side according to the number of active accounts regarding videos released by the target account in each preset time period.

[0219] Wherein, refer to the description of the second vector type code and the second type of statistical vector on the platform side in step S506. The second type of statistical vector on the account side is the second type of statistical vector that focuses on the online status of the account, which can be understood as the second type of statistical vector on the account side. The second type of statistical vector includes the second type of statistical vector on the platform side and the second type of statistical vector on the account side. The second type of statistical vector on the account side is a multi-dimensional vector, and each dimension represents a preset time period.

[0220] Optionally, for each preset time period, the computer device determines the second type of account-side value corresponding to the preset time period according to the number of active accounts about videos posted by the target account in the preset time period. The computer device determines the second type of account-side statistical vector according to the second type of account-side value corresponding to each preset time period.

[0221] Similarly, in order to reduce the amount of calculation, for each preset time period, the computer device performs a logarithmic operation on the number of active accounts regarding videos released by the target account in the preset time period, obtains the corresponding account-side second operation result, and uses the corresponding account-side second operation result as the account-side second category value corresponding to the preset time period.

[0222] Step S808, generating a third vector type code according to the vector type corresponding to the third type of statistical vector, and generating a third type of statistical vector on the platform side according to the number of video views of each video in the first video set in each preset time period, and generating a third type of statistical vector on the account side according to the number of video views of each video in the second video set in each preset time period.

[0223] Among them, referring to the description of the third vector type encoding and the third type of statistical vector on the platform side in step S508, the third type of statistical vector on the account side is a third type of statistical vector that reflects the browsing status of each video in the second video set in each preset time period. It can be understood as the third type of statistical vector on the account side. The third type of statistical vector includes the first type of statistical vector on the platform side and the third type of statistical vector on the account side. The third type of statistical vector on the account side is a multi-dimensional vector, and each dimension represents a preset time period.

[0224] Optionally, the computer device determines the third category value on the account side corresponding to the video in the second video set in the preset time period according to the video views of each video in the second video set in the preset time period. The computer device determines the third category statistical vector on the account side corresponding to the video in the second video set according to the third category value on the account side corresponding to each preset time period. The third category value on the account side corresponding to the video in the second video set in the preset time period represents the value of the dimension represented by the preset time period in the third category statistical vector on the account side corresponding to the video in the second video set.

[0225] Similarly, in order to reduce the amount of calculation, for each preset time period and each video in the second video set, the computer device performs a logarithmic operation on the number of video views of the video in the preset time period, obtains the corresponding account-side third operation result, and determines the corresponding account-side third operation result as the account-side third category value corresponding to the video in the second video set in the preset time period.

[0226] Step S810, through the playback volume estimation model, according to the first information type code and video semantic features, the second information type code, the first vector type code and the first type of statistical vector on the platform side, the second information type code, the second vector type code and the second type of statistical vector on the platform side, the second information type code, the third vector type code and the third type of statistical vector on the platform side, the third information type code, the first vector type code and the first type of statistical vector on the account side, the third information type code, the second vector type code and the second type of statistical vector on the account side, the third information type code, the third vector type code and the third type of statistical vector on the account side, output the estimated playback volume of the target video in each preset time period.

[0227] Optionally, the computer device determines a corresponding video semantic feature vector according to the video semantic features of the target video. Through the embedding layer of the playback volume estimation model, the first information type code is converted into a vector corresponding to the first information type code, the second information type code is converted into a vector corresponding to the second information type code, the third information type code is converted into a vector corresponding to the third information type code, the first vector type code is converted into a vector corresponding to the first vector type code, the second vector type code is converted into a vector corresponding to the second vector type code, and the third vector type code is converted into a vector corresponding to the third vector type code.

[0228] Through the encoder of the playback volume estimation model, according to the vector corresponding to the first information type encoding and the video semantic feature vector, the vector corresponding to the second information type encoding, the vector corresponding to the first vector type encoding and the first type of statistical vector on the platform side, the vector corresponding to the second information type encoding, the vector corresponding to the second vector type encoding and the second type of statistical vector on the platform side, the vector corresponding to the second information type encoding, the vector corresponding to the third vector type encoding and the third type of statistical vector on the platform side of each video in the first video set, the vector corresponding to the third information type encoding, the vector corresponding to the first vector type encoding and the first type of statistical vector on the account side, the vector corresponding to the third information type encoding, the vector corresponding to the second vector type encoding and the second type of statistical vector on the account side, the vector corresponding to the third information type encoding, the vector corresponding to the third vector type encoding and the third type of statistical vector on the account side of each video in the second video set, the estimated playback volume of the target video in each preset time period is obtained, and the estimated playback volume of the target video in each preset time period is output.

[0229] It should be noted that the vector corresponding to the first information type encoding, the vector corresponding to the second information type encoding, and the vector corresponding to the third information type encoding can help the playback volume estimation model distinguish whether the information belongs to semantic feature information, platform-side statistical information, or account-side statistical information.

[0230] The dimensions of the vector corresponding to the first information type encoding, the video semantic feature vector, the vector corresponding to the second information type encoding, the vector corresponding to the second information type encoding, the vector corresponding to the first vector type encoding, the first type of statistical vector on the platform side, the vector corresponding to the second vector type encoding, the second type of statistical vector on the platform side, the vector corresponding to the third vector type encoding, the third type of statistical vector on the platform side, the first type of statistical vector on the account side, the second type of statistical vector on the account side and the third type of statistical vector on the account side mentioned above are all the same, which are the ratios of the fixed durations of the target time period and the preset time period.

[0231] Through the first embedding layer of the playback volume estimation model, the vector corresponding to the first information type encoding is determined, the vector corresponding to the second information type encoding is determined, and the vector corresponding to the third information type encoding is determined. Through the second embedding layer of the playback volume estimation model, the vector corresponding to the first vector type encoding is determined, the vector corresponding to the second vector type encoding is determined, and the vector corresponding to the third vector type encoding is determined.

[0232] In this embodiment, by generating a first information type code, generating a second information type code and a third information type code, different information types are significantly distinguished by coding. After determining the corresponding different types of statistical vectors on the platform side (the first type of statistical vector on the platform side, the second type of statistical vector on the platform side, and the third type of statistical vector on the platform side), the vector type codes of the statistical vectors of different types on the platform side are generated. Similarly, after determining the corresponding different types of statistical vectors on the account side (the first type of statistical vector on the account side, the second type of statistical vector on the account side, and the third type of statistical vector on the account side), the vector type codes of the statistical vectors of different types on the account side are generated. In this way, through the playback volume estimation model, according to the video semantic features and the information type code to which it belongs, the statistical vector of each type on the platform side, the corresponding information type code and the vector type code to which it belongs, the statistical vector of each type on the account side, the corresponding information type code and the vector type code to which it belongs, comprehensively consider the information dimension, the statistical vector type dimension and the vector dimension, so as to accurately estimate the playback volume of each preset time period, greatly improving the playback volume of the target video released by the target account during the release period.

[0233] In some embodiments, Fig. 9 FIG. 8 is a flow chart of another embodiment of the step of determining the estimated playback volume. Step S810 includes:

[0234] Step S902: Obtain video side input data according to the first information type code and the video semantic features of the target video.

[0235] The embodiment of the above step S902 refers to step S602.

[0236] Step S904: Obtain the first-dimensional input data on the platform side according to the second information type code, the first vector type code and the first type of statistical vector on the platform side.

[0237] The embodiment of the above step S904 refers to step S604.

[0238] Step S906, obtaining the second dimensional input data on the platform side according to the second information type code, the second vector type code and the second type of statistical vector on the platform side.

[0239] The embodiment of the above step S906 refers to step S606.

[0240] Step S908, obtaining the third-dimensional input data on the platform side according to the second information type code, the third vector type code and the third type of statistical vector on the platform side.

[0241] The embodiment of the above step S908 refers to step S608.

[0242] Step S910, obtaining the first dimension input data on the account side according to the third information type code, the first vector type code and the first type of statistical vector on the account side.

[0243] Optionally, the computer device fuses the vector corresponding to the third information type code, the vector corresponding to the first vector type code and the first type of statistical vector on the account side to obtain the first dimension input data on the account side.

[0244] Exemplarily, the information type encoding further includes a third information type encoding, and the weight of the vector corresponding to the third information type is the first weight.

[0245] The computer device performs weighted summation processing on the vector corresponding to the third information type code and the vector corresponding to the first vector type code according to the first weight and the second weight to obtain the first-dimensional weighted data on the account side. The computer device superimposes the first-dimensional weighted data on the account side with the first-category statistical vector on the account side to obtain the first-dimensional input data on the account side.

[0246] For example: let the third information type code be code C, and the vector corresponding to the second information type code C be E(C); let the first vector type code be regarded as code 0, and the vector corresponding to the first vector type code 0 be E(0); the first weight and the second weight are L1 and L2 respectively, then the first dimension input data on the account side is E(C)×L1+E(0)×L2+the first type statistical vector on the account side.

[0247] Step S912, obtaining the second dimension input data on the account side according to the third information type code, the second vector type code and the second type of statistical vector on the account side.

[0248] Optionally, the computer device fuses the vector corresponding to the third information type code, the vector corresponding to the second vector type code and the second type of statistical vector on the account side to obtain the second dimensional input data on the account side.

[0249] Exemplarily, according to the first weight and the second weight, a vector corresponding to the third information type code and a vector corresponding to the second vector type code are weighted and summed to obtain the second-dimensional weighted data on the account side. The computer device superimposes the second-dimensional weighted data on the account side with the second-type statistical vector on the account side to obtain the second-dimensional input data on the account side.

[0250] For example: let the third information type code be code C, and the vector corresponding to the second information type code C be E(C); let the second vector type code be code 1, and the vector corresponding to the second vector type code 1 be E(1). The first weight and the second weight are L1 and L2 respectively. Then the second dimensional input data on the account side is E(C)×L1+E(1)×L2+the second type statistical vector on the account side.

[0251] Step S914, obtaining the third-dimensional input data on the account side according to the third information type code, the third vector type code and the third type of statistical vector on the account side.

[0252] Optionally, for each video in the second video set, the computer device fuses the vector corresponding to the third information type code, the vector corresponding to the third vector type code, and the third type of statistical vector on the account side to obtain the third dimensional input data on the account side corresponding to the video in the second video set.

[0253] Exemplarily, for each video in the second video set, a vector corresponding to the third information type code and a vector corresponding to the third vector type code are weighted and summed according to the first weight and the second weight to obtain account-side third-dimensional weighted data. The computer device superimposes the account-side third-dimensional weighted data with the third-category statistical vector of the account side corresponding to the video in the second video set to obtain account-side third-dimensional input data corresponding to the video in the second video set.

[0254] For example: for video P in the second video set, let the third information type code be code C, and the vector corresponding to the second information type code C be E(C); let the third vector type code be regarded as code 2, and the vector corresponding to the second vector type code 2 be E(2), the first weight and the second weight are L1 and L2 respectively, then the second dimensional input data on the account side is E(C)×L1+E(2)×L2+the third type statistical vector on the account side corresponding to video P.

[0255] Step S916, through the playback volume estimation model, according to the video side input data, the platform side first dimension input data, the platform side second dimension input data, the platform side third dimension input data, the account side first dimension input data, the account side second dimension input data and the account side third dimension input data, output the estimated playback volume of the target video in each preset time period.

[0256] Among them, the video side input data, the first dimension input data on the platform side, the second dimension input data on the platform side, the third dimension input data on the platform side corresponding to each video in the first video set, the first dimension input data on the account side, the second dimension input data on the account side, and the third dimension input data on the account side corresponding to each video in the second video set are all vectors of the same dimension, and the dimension is the ratio between the fixed duration of the target time period and the preset time period. For example, if the target time period is 1 day and the fixed duration of the preset time period is 1 hour, then the dimension is 24 dimensions.

[0257] Optionally, the computer device combines the video side input data, the platform side first dimension input data, the platform side second dimension input data, the platform side third dimension input data corresponding to each video in the first video set, the account side first dimension input data, the account side second dimension input data, and the account side third dimension input data corresponding to each video in the second video set to obtain the target input data. The computer device obtains the estimated playback volume of the target video in each preset time period according to the target input data through the encoder of the playback volume estimation model, and outputs the estimated playback volume of the target video in each preset time period.

[0258] In this embodiment, through the platform-side input data corresponding to various statistical vectors on the platform side, the account-side input data corresponding to various statistical vectors on the account side, and the video-side input data, the preset playback volume of each preset time period can be estimated from multiple dimensions, and the release time period matching the target account can be determined while improving the accuracy of the estimated release time period for the target video.

[0259] In some embodiments, the release time period of the target video is determined based on the estimated playback volume of the target video in each preset time period, including: determining the preset time period corresponding to the maximum value of the estimated playback volume in each preset time period based on the estimated playback volume of the target video in each preset time period; if the preset time period corresponding to the maximum value is a non-prohibited release time period, then the preset time period corresponding to the maximum value is determined as the release time period of the target video.

[0260] Among them, the prohibited publishing period is a period that is not suitable for publishing any video. Exemplarily, a specific date is not suitable for publishing any video, and all preset time periods on the specific date are regarded as prohibited publishing periods. Exemplarily, the corresponding prohibited time period is determined based on the user information of the creator; wherein the user information includes the age of the creator. For example, if the age of the creator is less than the age threshold, it indicates that the creator is a minor. In order to prevent the creator from overusing the video platform, the early morning period is regarded as the corresponding prohibited time period, such as prohibiting the creator from publishing any video in the early morning.

[0261] Optionally, the computer device selects the maximum value of the estimated playback volume from multiple preset time periods, and determines the current time of the current date. If the current time is not after the preset time period corresponding to the maximum value in the current date, it is determined whether the current date is a specific date. If it is not a specific date, it is determined that the preset time period corresponding to the maximum value in the current date is a non-prohibited release time period, and the preset time period corresponding to the maximum value in the current date is determined as the release time period of the target time period. If it is a specific date, it is determined whether the date after the current date is a specific date. If the date after the current date is not a specific date, it is determined that the preset time period corresponding to the maximum value in the next date is a non-prohibited release time period, and the preset time period corresponding to the maximum value in the next date is determined as the release time period of the target time period.

[0262] If the current time is after the preset time period corresponding to the maximum value, it is determined whether the next date after the current date corresponding to the current time is a specific date. If it is not a specific date, it is determined that the preset time period corresponding to the maximum value in the next date is a non-prohibited time period, and the preset time period corresponding to the maximum value in the next date is determined as the release time period of the target time period.

[0263] As mentioned above, each preset time period is a plurality of continuous time periods of the same duration within the target time period. The target time period is a time period in the same day. Therefore, if the preset time period corresponding to the maximum value is 9:00 to 10:00, the current time of the current date is 8:00, and it is determined that the current date is not a specific date, then it is determined that 9:00 to 8:00 in the current date is a non-prohibited time period, and 9:00 to 10:00 in the current date is used as the release time period of the target video. If the current time of the current date is 15:00, it is determined whether the next date is a specific date. If the next date is not a specific date, it is determined that 9:00 to 10:00 in the next date is a non-prohibited release time period, and 9:00 to 10:00 in the next date is used as the release time period of the target video.

[0264] Optionally, the computer device sequentially obtains the maximum value of the estimated playback volume from multiple preset time periods, obtains the user information of the creator, and determines the prohibited time period corresponding to the user information. If the preset time period corresponding to the maximum value is not the prohibited time period corresponding to the user information, the computer device determines the preset time period corresponding to the maximum value as the release time period of the target video. The user information includes at least one of the user age and the user's release habits, and the prohibited time period corresponding to the user information can be determined based on at least one of the user age and the user's release habits.

[0265] In this example, based on the estimated playback volume of the target video in each preset time period, the preset time period corresponding to the maximum value of the estimated playback volume in each preset time period is determined, that is, the time period with the maximum playback volume is located. If the preset time period corresponding to the maximum value is a non-prohibited release time period, the preset time period corresponding to the maximum value is determined as the release time period of the target video, thus ensuring the playback volume of the target video.

[0266] In some embodiments, the release period of the target video is determined based on the estimated playback volume of the target video in each preset period, including: sorting each preset period according to the estimated playback volume of the target video in each preset period, and the larger the estimated playback volume, the higher the sequence number of the corresponding preset period. Select multiple preset periods with higher sequence numbers from multiple preset periods. Determine the historical release period of multiple historical videos released by the target account, and determine the release period of the target video from multiple preset periods with higher sequence numbers based on multiple historical release periods and prohibited periods.

[0267] Optionally, after determining the serial number of each preset time period, start selecting from the first serial number and sequentially obtain a first number of preset time periods. The computer device determines the historical release time periods of multiple historical videos released by the target account, calculates the mean and variance corresponding to the multiple historical release time periods, and selects a second number of preset time periods from the first number of preset time periods based on the mean, variance and prohibited time periods. Among them, none of the second number of preset time periods is a non-prohibited time period, and the difference between the second number of preset time periods and the mean is less than the difference threshold. The computer device selects the preset time period corresponding to the maximum value based on the estimated playback volume of the second number of preset time periods selected as the release time period of the target video.

[0268] In this embodiment, each preset time period is sorted by the estimated playback volume of the target video in each preset time period. The larger the estimated playback volume, the higher the serial number of the corresponding preset time period. Multiple preset time periods with higher serial numbers are selected from multiple preset time periods. A comprehensive consideration is made based on the historical release time periods and prohibited time periods when the creator has released videos in the past, and a secondary selection is made to ensure that the secondary selected preset time period is close to the creator's historical release time period and avoids release during the prohibited time period. Then, the release time period is finally determined from the secondary selected preset time period to ensure the playback volume of the target video.

[0269] In some embodiments, Fig.10 FIG. 1 is a flow chart of a model training step in an embodiment. The estimated playback volume is output by a playback volume estimation model based on the video semantic features of the target video and the statistical information on the video platform side. The training steps of the playback volume estimation model include:

[0270] Step S1002: obtaining training samples, where the training samples include sample videos, annotated release periods of the sample videos, and the sum of annotated playback volumes of the sample videos in each preset period.

[0271] The marked release time period is the actual release time period of the sample video. For example, if the sample video is released at 9:20, then the preset time period from 9 to 10 o'clock is the marked release time period.

[0272] Step S1004, obtaining video semantic features of the sample video; and determining a third video set in the platform video library that is similar to the sample video.

[0273] Optionally, the computer device determines the video semantic features of the sample video and the video semantic features of each video in the platform video library through a semantic feature extraction model, and calculates the similarity between the video semantic features of the sample video and the video semantic features of each video in the platform video library to determine a third video set.

[0274] Step S1006, determining the statistical information on the video platform side according to the video popularity related information in each preset time period and the video views of each video in the third video set in each preset time period.

[0275] Optionally, the computer device counts the number of videos posted on the video platform in each preset time period; counts the number of accounts active on the video platform in each preset time period; counts the video views of each video in the third video set in each preset time period; and uses the counted number of videos posted on the video platform in each preset time period, the number of accounts active on the video platform in each preset time period, and the video views of each video in the third video set in each preset time period as statistical information on the video platform side.

[0276] Step S1008, through the playback volume estimation model, based on the video semantic features and the statistical information on the video platform side, output the estimated playback volume of the sample video in each preset time period; according to the difference between the estimated playback volume of the sample video corresponding to the marked release period and the sum of the marked playback volume in each preset time period, update the playback volume estimation model.

[0277] Optionally, a corresponding first information type code and a second information type code are generated according to the first information type and the second information type, respectively, the first information type is used to represent the information type of the video semantic features of the sample video, and the second information type is used to represent the information type of the statistical information on the video platform side; a first vector type code is generated, and according to the statistical number of videos posted on the video platform in each preset time period, a first type of statistical vector on the platform side is generated, and the first vector type code is used to represent the vector type of the first type of statistical vector; a second vector type code is generated, and according to the statistical number of accounts active on the video platform in each preset time period, a second type of statistical vector on the platform side is generated, and the second vector type code is used to represent the vector type of the second type of statistical vector. The vector type of the second type of statistical vector; generating a third vector type code, and generating a third type of statistical vector on the platform side according to the video views of each video in the first video set in each preset time period, the third vector type code is used to represent the vector type of the third type of statistical vector; through the playback volume estimation model, according to the first information type code and the video semantic features of the sample video, the second information type code, the first vector type code and the first type of statistical vector on the platform side, the second information type code, the second vector type code and the second type of statistical vector on the platform side, and the second information type code, the third vector type code and the third type of statistical vector on the platform side, the estimated playback volume of the sample video in each preset time period is output.

[0278] The above model training is performed with the training goal of minimizing the difference between the estimated playback volume of the sample video corresponding to the annotated release period and the sum of the annotated playback volume in each preset period.

[0279] For example, if the sum of the annotated playback volume in each preset time period is Q, the estimated playback volume of the sample video corresponding to the annotated release period is The following formula is used to determine the difference between the two: Then the difference L1 is as follows:

[0280]

[0281] when Then the difference L2 is as follows:

[0282]

[0283] Among them, c is a constant. It should be noted that the above-mentioned difference calculation method can effectively handle outliers, avoid the influence of outliers on model training, and is also robust.

[0284] In this embodiment, after calculating the estimated playback volume of the sample video in each preset time period, in order to ensure that the model can estimate a more accurate playback volume, the sum of the estimated playback volumes of all estimated time periods is used as a standard to ensure that the estimated playback volume corresponding to the marked release time period is close to the sum of the estimated playback volumes of all estimated time periods, thereby improving the accuracy of the playback volume estimation model for the playback volume.

[0285] The present application also provides an application scenario, which applies the above-mentioned release period estimation method. Specifically, the application of the release period estimation method in this application scenario is as follows: in a scenario involving a short video playback application, after the creator creates the target video, in order to ensure the exposure of the target video, it is necessary to determine a suitable release period. Based on this, the method of the embodiment of the present application can accurately estimate the playback volume of each preset time period to improve the accuracy of the release period estimation. Specifically, the video semantic features of the target video to be released to the video platform are obtained; the first video set similar to the target video in the platform video library is determined; according to the video platform's video heat related information in each preset time period, the video views of each video in the first video set in each preset time period, the video platform side statistical information is determined; based on the video semantic features of the target video and the video platform side statistical information, the estimated playback volume of the target video in each preset time period is output; according to the estimated playback volume of the target video in each preset time period, the release period of the target video is determined. In this way, it is possible to take into account that videos of different types of content have different audiences in different time periods and the release periods of other videos of the same type of content, thereby accurately estimating the playback volume of each preset time period, thereby improving the accuracy of estimating the release period of the target video.

[0286] Of course, it is not limited to this. The release period estimation method provided in the present application can also be applied in other application scenarios. For example, in a scenario involving product promotion, after the promoter has produced a target video about the product, in order to ensure the playback volume of the target video, the release period estimation method provided in the embodiment of the present application can be used to accurately estimate the release period of the target video to ensure the effect of the product promotion.

[0287] The above application scenarios are merely illustrative illustrations. It is to be understood that the application of the release period estimation method provided in each embodiment of the present application is not limited to the above scenarios.

[0288] In a specific embodiment, the implementation subject of this embodiment is a computer device. The specific description is as follows:

[0289] Step S1: Determine a video set. The video set includes a first video set and a second video set.

[0290] like Fig.11A The figure shows a flowchart of the steps of determining a video collection in an embodiment, specifically: a computer device obtains a target video to be published on a video platform, and extracts semantic features of the target video through a semantic feature extraction model to obtain video semantic features of the target video. Platform videos published by other accounts are obtained from the platform video library; other accounts are accounts other than the target account; semantic features are extracted from the obtained platform videos through a semantic feature extraction model to obtain video semantic features of each platform video; for each platform video, the similarity between the video semantic features of the target video and the video semantic features of each platform video is calculated, and similar videos of the target video are screened out from the obtained platform videos according to the similarity between the video semantic features of the target video and the video semantic features of each platform video; a first video collection is obtained according to the screened similar videos.

[0291] Acquire historical videos released by a target account; extract semantic features from the acquired historical videos using a semantic feature extraction model to obtain video semantic features of each historical video; for each historical video, calculate the similarity between the video semantic features of the target video and the video semantic features of each historical video, and filter out similar videos of the target video from the acquired historical videos based on the similarities between the video semantic features of the target video and the video semantic features of each historical video; obtain a second video set based on the filtered similar videos.

[0292] The above-mentioned semantic feature extraction model is a trained model for image description, and its training steps include: obtaining a sample video and uniformly sampling multiple video frames from the sample video. Each video frame is sequentially input into the shared image encoder of the model for image description to obtain the image encoding result corresponding to each video frame. The image encoding results corresponding to the multiple video frames are processed by the confluence layer in the model for image description to obtain the sample video semantic features corresponding to the sample video. Since a video may have multiple labels, the sigmoid function is used for activation processing, that is, the image encoding results corresponding to the multiple video frames are processed by the confluence layer in the model for image description to obtain the sample video semantic features corresponding to the multiple labels. At this time, for each label, the sample video semantic features corresponding to each label are mapped by the sigmoid function to obtain the mapping result, and the cross entropy loss is constructed according to the mapping results corresponding to each label. The model for image description is trained according to the cross entropy loss until the training conditions are met to obtain the trained model for image description, and the trained model for image description is a semantic feature extraction model.

[0293] Step S2: Obtaining platform-side statistical information. Specifically, the computer device counts the number of videos published on the video platform in each preset time period; counts the number of accounts active on the video platform in each preset time period; counts the video views of each video in the first video set in each preset time period; and uses the counted number of videos published on the video platform in each preset time period, the number of accounts active on the video platform in each preset time period, and the video views of each video in the first video set in each preset time period as the video platform-side statistical information.

[0294] Step S3: Obtain account-side statistical information. Specifically, count the number of videos posted on the video platform by the target account in each preset time period; count the number of active accounts regarding videos posted by the target account in each preset time period; count the video views of each video in the second video set in each preset time period; and use the counted number of videos posted on the video platform in each preset time period, the number of accounts active on the video platform in each preset time period, and the video views of each video in the second video set in each preset time period as account-side statistical information.

[0295] Step S4: Determine the release period. Fig. 11B FIG. 1 is a process for determining a release period in an embodiment, specifically:

[0296] Step S4.1: Generate a first information type code based on the first information type, and generate a second information type code based on the second information type, the first information type is used to represent the information type of the semantic features of the video, the second information type is used to represent the information type of the statistical information on the video platform side, and the third information type is used to represent the information type of the statistical information on the account side.

[0297] Step S4.2: Generate a first vector type code according to the vector type corresponding to the first type of statistical vector, and generate a first type of statistical vector on the platform side according to the statistical number of videos published on the video platform in each preset time period, and generate a first type of statistical vector on the account side according to the statistical number of videos published on the video platform through the target account in each preset time period.

[0298] Step S4.3: Generate a second vector type code according to the vector type corresponding to the second type of statistical vector, and generate a second type of statistical vector on the platform side according to the number of accounts that are active on the video platform in each preset time period, and generate a second type of statistical vector on the account side according to the number of active accounts regarding videos released by the target account in each preset time period.

[0299] Step S4.4: Generate a third vector type code according to the vector type corresponding to the third type of statistical vector, and generate a third type of statistical vector on the platform side according to the number of video views of each video in the first video set in each preset time period, and generate a third type of statistical vector on the account side according to the number of video views of each video in the second video set in each preset time period.

[0300] Step S4.5: Obtain video side input data according to the first information type code and the video semantic features of the target video.

[0301] Optionally, the vector corresponding to the first information type code and the video semantic feature vector corresponding to the video semantic feature of the target video are superimposed to obtain the video side input data. For example, let the first information type code be code A, and the vector corresponding to the first information type code (i.e., code A) be E(A), then the video side input data is E(A)+video semantic feature vector.

[0302] Step S4.6, obtaining the first dimension input data on the platform side according to the second information type code, the first vector type code and the first type of statistical vector on the platform side.

[0303] Optionally, the vector corresponding to the second information type code, the vector corresponding to the first vector type code and the first type of statistical vector on the platform side are superimposed to obtain the first dimensional input data on the platform side.

[0304] For example: let the second information type code be code B, and the vector corresponding to the second information type code B be E(B); let the first vector type code be regarded as code 0, and the vector corresponding to the first vector type code 0 be E(0), then the first dimension input data on the platform side is E(B)+E(0)+the first type statistical vector on the platform side.

[0305] Step S4.7, obtaining the second dimensional input data on the platform side according to the second information type code, the second vector type code and the second type of statistical vector on the platform side.

[0306] Optionally, the vector corresponding to the second information type code, the vector corresponding to the second vector type code and the second type of statistical vector on the platform side are superimposed to obtain the second dimensional input data on the platform side.

[0307] For example: let the second information type code be code B, and the vector corresponding to the second information type code B be E(B); let the second vector type code be code 1, and the vector corresponding to the second vector type code 1 be E(1), then the second dimensional input data on the platform side is E(B)+E(1)+the second type of statistical vector on the platform side.

[0308] Step S4.8, obtaining the third-dimensional input data on the platform side according to the second information type code, the third vector type code and the third type of statistical vector on the platform side.

[0309] Optionally, for each video in the first video set, a vector corresponding to the second information type code, a vector corresponding to the third vector type code and a third type of statistical vector on the platform side corresponding to the video are superimposed to obtain third-dimensional input data on the platform side corresponding to the video.

[0310] For example: for video 1 in the first video set, let the second information type code be code B, and the vector corresponding to the second information type code B be E(B); let the third vector type code be code 2, and the vector corresponding to the third vector type code 2 be E(2), then the third-dimensional input data on the platform side corresponding to video 1 is E(B)+E(2)+the third-category statistical vector 1 on the platform side corresponding to video 1. If there are k videos in the first video set, a similar method is used to determine that the third-dimensional input data on the platform side corresponding to video k is E(B)+E(2)+the third-category statistical vector k on the platform side corresponding to video k.

[0311] Step S4.9, obtaining the first dimension input data on the account side according to the third information type code, the first vector type code and the first type of statistical vector on the account side.

[0312] Optionally, the vector corresponding to the third information type code, the vector corresponding to the first vector type code and the first type of statistical vector on the account side are superimposed to obtain the first dimension input data on the account side.

[0313] For example: let the third information type code be code C, and the vector corresponding to the second information type code C be E(C); let the first vector type code be regarded as code 0, and the vector corresponding to the first vector type code 0 be E(0), then the first dimension input data on the account side is E(C)+E(0)+the first type statistical vector on the account side.

[0314] Step S4.10, obtaining the second dimension input data on the account side according to the third information type code, the second vector type code and the second type of statistical vector on the account side.

[0315] Optionally, the vector corresponding to the third information type code, the vector corresponding to the second vector type code and the second type of statistical vector on the account side are superimposed to obtain the second dimensional input data on the account side.

[0316] For example: let the third information type code be code C, and the vector corresponding to the second information type code C be E(C); let the second vector type code be code 1, and the vector corresponding to the second vector type code 1 be E(1), then the second dimension input data on the account side is E(C)+E(1)+the first type statistical vector on the account side.

[0317] Step S4.11, obtaining the third-dimensional input data on the account side according to the third information type code, the third vector type code and the third type of statistical vector on the account side.

[0318] Optionally, for each video in the second video set, the vector corresponding to the third information type encoding, the vector corresponding to the third vector type encoding and the third type of statistical vector on the account side are superimposed to obtain the third dimensional input data on the account side corresponding to the video in the second video set.

[0319] For example: for the first video in the second video set, let the third information type code be code C, and the vector corresponding to the second information type code C be E(C); let the third vector type code be code 2, and the vector corresponding to the second vector type code 2 be E(2), then the second dimension input data on the account side is E(C)+E(2)+the third category statistical vector 1 on the account side corresponding to the first video. If there are k videos in the second video set, a similar method is used to determine that the third dimension input data on the platform side corresponding to the kth video in the second video set is E(B)+E(2)+the third category statistical vector k on the account side corresponding to the kth video in the second video set.

[0320] Step S4.12, the computer device combines the video side input data, the first dimension input data on the platform side, the second dimension input data on the platform side, the third dimension input data on the platform side corresponding to each video in the first video set, the first dimension input data on the account side, the second dimension input data on the account side, and the third dimension input data on the account side corresponding to each video in the second video set to obtain the target input data. The computer device processes the target input data through multiple self-attention layers in the encoder of the playback volume estimation model to obtain the estimated playback volume of the target video in each preset time period, and outputs the estimated playback volume of the target video in each preset time period. For example, the target input data is sequentially passed through self-attention layer 1, ..., self-attention layer m to obtain a prediction vector, which includes the estimated playback volume of the target video in each preset time period.

[0321] The training steps of the above-mentioned playback volume estimation model include: obtaining training samples, the training samples include sample videos, the annotated release time period of the sample videos and the sum of the annotated playback volumes of the sample videos in each preset time period; obtaining video semantic features of the sample videos; determining a third video set similar to the sample video in the platform video library; determining the video platform side statistical information based on the video popularity related information of the video platform in each preset time period and the video views of each video in the third video set in each preset time period; outputting the estimated playback volume of the sample video in each preset time period through the playback volume estimation model based on the video semantic features and the video platform side statistical information; updating the playback volume estimation model based on the difference between the estimated playback volume corresponding to the sample video in the annotated release time period and the sum of the annotated playback volume in each preset time period.

[0322] Step S4.13, based on the estimated playback volume of the target video in each preset time period, determine the preset time period corresponding to the maximum value of the estimated playback volume in each preset time period; if the preset time period corresponding to the maximum value is a non-prohibited release time period, then the preset time period corresponding to the maximum value is determined as the release time period of the target video.

[0323] In this embodiment, by obtaining the video semantic features of the target video to be published to the video platform; determining the first video set similar to the target video in the platform video library, that is, determining the first video set that has been published and is of the same type as the target video from the platform video library. According to the video platform's video heat related information in each preset time period and the video views of each video in the first video set in each preset time period, the video platform side statistical information is determined. In this way, the video platform side statistical information can more comprehensively reflect the user's attention to the same type of video in different time periods, thereby intuitively reflecting the audience groups of the same type of video in different time periods. Based on this, based on the video semantic features of the target video and the video platform side statistical information, it can be considered that videos of different types of content have different audience groups and the release time periods of other videos of the same type of content in different time periods, so that the playback volume of each preset time period can be accurately estimated, and the estimated playback volume of the target video in each preset time period can be output; therefore, according to the estimated playback volume of the target video in each preset time period, the release time period of the target video can be more accurately determined, which improves the accuracy of estimating the release time period of the target video.

[0324] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0325] Based on the same inventive concept, the embodiment of the present application also provides a release period estimation device for implementing the release period estimation method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more release period estimation device embodiments provided below can refer to the limitations of the release period estimation method above, and will not be repeated here.

[0326] In one embodiment, Fig.12 As shown, a release period estimation device 1200 is provided, comprising: a semantic feature acquisition module 1202, a video set determination module 1204, a statistical information acquisition module 1206, an output module 1208 and a release period determination module 1210, wherein:

[0327] A semantic feature acquisition module 1202 is used to acquire video semantic features of a target video to be published to a video platform;

[0328] A video set determining module 1204 is used to determine a first video set similar to the target video in the platform video library;

[0329] The statistical information acquisition module 1206 is used to determine the statistical information on the video platform side according to the video popularity related information in each preset time period and the video views of each video in the first video set in each preset time period;

[0330] The output module 1208 is used to output the estimated playback volume of the target video in each preset time period based on the video semantic features of the target video and the statistical information of the video platform side;

[0331] The release period determination module 1210 is used to determine the release period of the target video according to the estimated playback volume of the target video in each preset period.

[0332] In some embodiments, the semantic feature acquisition module 1202 is used to acquire a target video to be published to a video platform; extract semantic features from the target video to obtain video semantic features of the target video.

[0333] In some embodiments, the target video is to be released through the target account; the video collection determination module 1204 is used to obtain platform videos released through other accounts from the platform video library; other accounts are accounts other than the target account; semantic features are extracted from the obtained platform videos to obtain video semantic features of each platform video; based on the similarity between the video semantic features of the target video and the video semantic features of each platform video, similar videos to the target video are screened out from the obtained platform videos; based on the screened similar videos, a first video collection is obtained.

[0334] In some embodiments, the statistical information acquisition module 1206 is used to count the number of videos published on the video platform in each preset time period; count the number of accounts active on the video platform in each preset time period; count the video views of each video in the first video set in each preset time period; and use the counted number of videos published on the video platform in each preset time period, the number of accounts active on the video platform in each preset time period, and the video views of each video in the first video set in each preset time period as statistical information on the video platform side.

[0335] In some embodiments, the output module 1208 is used to generate a first information type code according to the first information type, and to generate a second information type code according to the second information type, wherein the first information type is used to represent the information type of the semantic features of the video, and the second information type is used to represent the information type of the statistical information on the video platform side; generate the first vector type code according to the vector type corresponding to the first type of statistical vector, and generate the first type of statistical vector on the platform side according to the number of videos published on the video platform in each preset time period; generate the second vector type code according to the vector type corresponding to the second type of statistical vector, and generate the second vector type code according to the number of accounts active on the video platform in each preset time period. Generate a second type of statistical vector on the platform side; generate a third vector type code according to the vector type corresponding to the third type of statistical vector, and generate the third type of statistical vector on the platform side according to the statistical video views of each video in the first video set in each preset time period; through the playback volume estimation model, according to the first information type code and video semantic features, the second information type code, the first vector type code and the first type of statistical vector on the platform side, the second information type code, the second vector type code and the second type of statistical vector on the platform side, and the second information type code, the third vector type code and the third type of statistical vector on the platform side, the estimated playback volume of the target video in each preset time period is output.

[0336] In some embodiments, the output module 1208 is used to obtain video side input data according to the first information type code and the video semantic features of the target video; obtain the first dimension input data on the platform side according to the second information type code, the first vector type code and the first type of statistical vector on the platform side; obtain the second dimension input data on the platform side according to the second information type code, the second vector type code and the second type of statistical vector on the platform side; obtain the third dimension input data on the platform side according to the second information type code, the third vector type code and the third type of statistical vector on the platform side; through the playback volume estimation model, according to the video side input data, the first dimension input data on the platform side, the second dimension input data on the platform side and the third dimension input data on the platform side, the estimated playback volume of the target video in each preset time period is output.

[0337] In some embodiments, the target video is to be released through the target account, and the video set determination module 1204 is also used to determine a second video set similar to the target video in the historical videos released by the target account; the statistical information acquisition module 1206 is also used to determine the account-side statistical information based on the account popularity-related information of the target account in each preset time period and the video views of each video in the second video set in each preset time period; the output module 1208 is used to output the estimated playback volume of the target video in each preset time period based on the video semantic features of the target video, the video platform-side statistical information and the account-side statistical information.

[0338] In some embodiments, the video collection determination module 1204 is also used to obtain historical videos released by the target account; perform semantic feature extraction on the obtained historical videos to obtain video semantic features of each historical video; filter out similar videos of the target video from the obtained historical videos based on the similarity between the video semantic features of the target video and the video semantic features of each historical video; and obtain a second video collection based on the filtered similar videos.

[0339] In some embodiments, the statistical information acquisition module 1206 is also used to count the number of videos posted on the video platform by the target account in each preset time period; count the number of active accounts regarding videos posted by the target account in each preset time period; count the video views of each video in the second video set in each preset time period; and use the counted number of videos posted on the video platform in each preset time period, the number of accounts active on the video platform in each preset time period, and the video views of each video in the second video set in each preset time period as account-side statistical information.

[0340] In some embodiments, the output module 1208 is used to generate a first information type code according to the first information type, generate a second information type code according to the second information type, and generate a third information type code according to the third information type, the first information type is used to represent the information type of the semantic features of the video, the second information type is used to represent the information type of the statistical information on the video platform side, and the third information type is used to represent the information type of the statistical information on the account side; generate the first vector type code according to the vector type corresponding to the first type of statistical vector, and generate the first type of statistical vector on the platform side according to the number of videos published on the video platform in each preset time period according to the statistics, and generate the first type of statistical vector on the account side according to the number of videos published on the video platform through the target account in each preset time period; generate the second vector type code according to the vector type corresponding to the second type of statistical vector, and generate the second type of statistical vector on the platform side according to the number of accounts active on the video platform in each preset time period according to the statistics, and generate the second type of statistical vector on the platform side according to the number of active accounts regarding videos published through the target account in each preset time period. Generate a second type of statistical vector on the account side; generate a third vector type code according to the vector type corresponding to the third type of statistical vector, and generate a third type of statistical vector on the platform side according to the statistical video views of each video in the first video set in each preset time period, and generate a third type of statistical vector on the account side according to the statistical video views of each video in the second video set in each preset time period; through a playback volume estimation model, according to the first information type code and video semantic features, the second information type code, the first vector type code and the first type of statistical vector on the platform side, the second information type code, the second vector type code and the second type of statistical vector on the platform side, the second information type code, the third vector type code and the third type of statistical vector on the platform side, the third information type code, the first vector type code and the first type of statistical vector on the account side, the third information type code, the second vector type code and the second type of statistical vector on the account side, the third information type code, the third vector type code and the third type of statistical vector on the account side, the third information type code, the third vector type code and the third type of statistical vector on the account side, the

[0341] In some embodiments, the release period determination module 1210 is used to determine the preset period corresponding to the maximum value of the estimated playback volume of each preset period based on the estimated playback volume of the target video in each preset period; if the preset period corresponding to the maximum value is a non-prohibited release period, the preset period corresponding to the maximum value is determined as the release period of the target video.

[0342] In some embodiments, the estimated playback volume is output by a playback volume estimation model based on the video semantic features of the target video and the statistical information on the video platform side. The device also includes a model training module, which is used to obtain training samples, and the training samples include sample videos, the annotated release time period of the sample videos, and the sum of the annotated playback volumes of the sample videos in each preset time period; obtain the video semantic features of the sample videos; determine a third video set similar to the sample video in the platform video library; determine the statistical information on the video platform side based on the video popularity related information of the video platform in each preset time period and the video views of each video in the third video set in each preset time period; output the estimated playback volume of the sample video in each preset time period based on the video semantic features and the statistical information on the video platform side through the playback volume estimation model; update the playback volume estimation model based on the difference between the estimated playback volume corresponding to the sample video in the annotated release time period and the sum of the annotated playback volume in each preset time period.

[0343] Each module in the above-mentioned release period estimation device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0344] In one embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Fig.13 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a release period estimation method is implemented.

[0345] Those skilled in the art will understand that Fig.13 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0346] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0347] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0348] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0349] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0350] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0351] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0352] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for estimating a release period, characterized in that: The method comprises: Obtain video semantic features of a target video to be published to a video platform; Determine a first video set in a platform video library that is similar to the target video; Determine the video platform side statistical information according to the video popularity related information of the video platform in each preset time period and the video views of each video in the first video set in each preset time period; Based on the video semantic features of the target video and the statistical information on the video platform side, outputting the estimated playback volume of the target video in each preset time period; The release period of the target video is determined according to the estimated playback volume of the target video in each preset time period.

2. The method according to claim 1, characterized in that The step of obtaining the video semantic features of the target video to be published to the video platform includes: Obtain a target video to be published to the video platform; Semantic features are extracted from the target video to obtain video semantic features of the target video.

3. The method according to claim 1, characterized in that: The target video is to be published through the target account; The determining of a first video set in a platform video library that is similar to the target video includes: Acquire platform videos published by other accounts from the platform video library; the other accounts are accounts other than the target account; Extracting semantic features from the acquired platform videos to obtain video semantic features of each platform video; Filtering similar videos of the target video from the acquired platform videos according to the similarities between the video semantic features of the target video and the video semantic features of each of the platform videos; A first video set is obtained according to the screened similar videos.

4. The method according to claim 1, characterized in that The determining of the video platform-side statistical information according to the video popularity related information of the video platform in each preset time period and the video views of each video in the first video set in each preset time period includes: Counting the number of videos posted on the video platform during each preset time period; Count the number of accounts active on the video platform during each preset time period; Counting the number of video views of each video in the first video set in each preset time period; The number of videos published on the video platform in each preset time period, the number of accounts active on the video platform in each preset time period, and the number of video views of each video in the first video set in each preset time period are counted as statistical information on the video platform side.

5. The method according to claim 4, characterized in that The outputting the estimated playback volume of the target video in each preset time period based on the video semantic feature of the target video and the statistical information on the video platform side includes: Generate a first information type code according to the first information type, and generate a second information type code according to the second information type, wherein the first information type is used to represent the information type of the video semantic feature, and the second information type is used to represent the information type of the video platform side statistical information; Generate a first vector type code according to the vector type corresponding to the first type of statistical vector, and generate a first type of statistical vector on the platform side according to the number of videos published on the video platform in each preset time period. Generate a second vector type code according to the vector type corresponding to the second type of statistical vector, and generate a second type of statistical vector on the platform side according to the number of accounts active on the video platform in each preset time period; Generate a third vector type code according to the vector type corresponding to the third type of statistical vector, and generate a third type of statistical vector on the platform side according to the statistical video views of each video in the first video set in each preset time period; Through the playback volume estimation model, according to the first information type code and the video semantic features, the second information type code, the first vector type code and the first type of statistical vector on the platform side, the second information type code, the second vector type code and the second type of statistical vector on the platform side, and the second information type code, the third vector type code and the third type of statistical vector on the platform side, the estimated playback volume of the target video in each preset time period is output.

6. The method according to claim 5, characterized in that The playback volume estimation model outputs the estimated playback volume of the target video in each preset time period according to the first information type encoding and the video semantic feature, the second information type encoding, the first vector type encoding and the first type of statistical vector on the platform side, the second information type encoding, the second vector type encoding and the second type of statistical vector on the platform side, and the second information type encoding, the third vector type encoding and the third type of statistical vector on the platform side, including: Obtaining video side input data according to the first information type encoding and the video semantic features of the target video; Obtaining first-dimensional input data on the platform side according to the second information type code, the first vector type code and the first type of statistical vector on the platform side; Obtaining second-dimensional input data on the platform side according to the second information type code, the second vector type code and the second type of statistical vector on the platform side; Obtaining third-dimensional input data on the platform side according to the second information type code, the third vector type code and the third type of statistical vector on the platform side; Through the playback volume estimation model, the estimated playback volume of the target video in each preset time period is output according to the video side input data, the first dimension input data on the platform side, the second dimension input data on the platform side and the third dimension input data on the platform side.

7. The method according to claim 1, characterized in that The target video is to be published through a target account, and the method further includes: Determine a second video set similar to the target video from among historical videos posted by the target account; Determine account-side statistical information according to the account popularity related information of the target account in each preset time period and the video views of each video in the second video set in each preset time period; The outputting the estimated playback volume of the target video in each preset time period based on the video semantic feature of the target video and the statistical information on the video platform side includes: Based on the video semantic features of the target video, the video platform side statistical information and the account side statistical information, the estimated playback volume of the target video in each preset time period is output.

8. The method according to claim 7, characterized in that The determining of a second video set similar to the target video from among the historical videos published by the target account includes: Obtain historical videos posted by the target account; Extracting semantic features from the acquired historical videos to obtain video semantic features of each of the historical videos; Filtering similar videos of the target video from the acquired historical videos according to the similarities between the video semantic features of the target video and the video semantic features of each of the historical videos; A second video set is obtained based on the screened similar videos.

9. The method according to claim 7, characterized in that: The method further comprises: Counting the number of videos posted on the video platform by the target account in each preset time period; Counting the number of active accounts regarding videos posted by the target account in each preset time period; Counting the number of video views of each video in the second video set in each preset time period; The number of videos published on the video platform in each preset time period, the number of accounts active on the video platform in each preset time period, and the number of video views of each video in the second video set in each preset time period are counted as the account-side statistical information.

10. The method according to claim 9, characterized in that The outputting the estimated playback volume of the target video in each preset time period based on the video semantic feature of the target video and the statistical information on the video platform side includes: Generate a first information type code according to the first information type, generate a second information type code according to the second information type, and generate a third information type code according to the third information type, wherein the first information type is used to represent the information type of the video semantic feature, the second information type is used to represent the information type of the video platform side statistical information, and the third information type is used to represent the information type of the account side statistical information; Generate a first vector type code according to the vector type corresponding to the first type of statistical vector, generate a first type of statistical vector on the platform side according to the statistical number of videos published on the video platform in each preset time period, and generate a first type of statistical vector on the account side according to the statistical number of videos published on the video platform through the target account in each preset time period; Generate a second vector type code according to the vector type corresponding to the second type of statistical vector, generate a second type of statistical vector on the platform side according to the number of accounts active on the video platform in each preset time period, and generate a second type of statistical vector on the account side according to the number of active accounts about videos published by the target account in each preset time period; Generate a third vector type code according to the vector type corresponding to the third type of statistical vector, generate a third type of statistical vector on the platform side according to the statistical video views of each video in the first video set in each preset time period, and generate a third type of statistical vector on the account side according to the statistical video views of each video in the second video set in each preset time period; Through the playback volume estimation model, according to the first information type code and the video semantic features, the second information type code, the first vector type code and the first type of statistical vector on the platform side, the second information type code, the second vector type code and the second type of statistical vector on the platform side, the second information type code, the third vector type code and the third type of statistical vector on the platform side, the third information type code, the first vector type code and the first type of statistical vector on the account side, the third information type code, the second vector type code and the second type of statistical vector on the account side, the third information type code, the third vector type code and the third type of statistical vector on the account side, the estimated playback volume of the target video in each preset time period is output.

11. The method according to any one of claims 1 to 10, characterized in that The step of determining the release period of the target video according to the estimated playback volume of the target video in each preset period includes: According to the estimated playback volume of the target video in each preset time period, determine the preset time period corresponding to the maximum value of the estimated playback volume in each preset time period; If the preset time period corresponding to the maximum value is a non-prohibited release time period, the preset time period corresponding to the maximum value is determined as the release time period of the target video.

12. The method according to any one of claims 1 to 10, characterized in that The estimated playback volume is output by a playback volume estimation model based on the video semantic features of the target video and the statistical information on the video platform side. The training steps of the playback volume estimation model include: Acquire a training sample, wherein the training sample includes a sample video, a labeled release period of the sample video, and a sum of the labeled playback volume of the sample video in each preset period; Acquire video semantic features of the sample video; Determine a third video set in the platform video library that is similar to the sample video; Determine the video platform side statistical information according to the video popularity related information of the video platform in each preset time period and the video views of each video in the third video set in each preset time period; Outputting the estimated playback volume of the sample video in each preset time period through a playback volume estimation model based on the video semantic features and the video platform-side statistical information; The playback volume estimation model is updated according to the difference between the estimated playback volume of the sample video corresponding to the annotated release time period and the sum of the annotated playback volumes in each preset time period.

13. A release period estimation device, characterized in that: The device comprises: A semantic feature acquisition module, used to acquire the video semantic features of the target video to be published to the video platform; A video set determination module, used to determine a first video set similar to the target video in the platform video library; A statistical information acquisition module, configured to determine statistical information on the video platform side according to information related to video popularity of the video platform in each preset time period and the number of video views of each video in the first video set in each preset time period; An output module, configured to output an estimated playback volume of the target video in each preset time period based on the video semantic features of the target video and the statistical information on the video platform side; The release period determination module is used to determine the release period of the target video according to the estimated playback volume of the target video in each preset period.

14. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 12 are implemented.

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

16. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 12 are implemented.

Images