Method for identifying video author, method and device for recommending video resource

Abstract

The disclosure provides a video author identification method, a video resource recommendation method and device, and relates to the technical field of computers, in particular to the technical field of video technology and intelligent recommendation. The video author identification method comprises: determining first image features of a plurality of video works of a video author according to classification results of cover image of the plurality of video works; determining first video works of a target type from the plurality of video works according to text features and the first image features of the plurality of video works; wherein the text features are obtained according to label information of the video works; determining first work features of the video author according to a proportion of the first video works in the plurality of video works; determining second work features of the video author according to a proportion of second video works containing target attributes in the plurality of video works; and determining an author type label of the video author according to the first work features and the second work features. The disclosure is helpful for mining of video authors and video resources.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of computer, and particularly relates to the technical field of video technology and intelligent recommendation. BACKGROUND

[0002] When a user browses a video, a video platform can select video resources related to the user's interest from a large amount of video resources for recommendation. In order to improve the recommendation effect, it is necessary to mine the video resources in advance to form various types of video resource pools. SUMMARY

[0003] The present disclosure provides a video author identification method, a video resource recommendation method, a device, an apparatus and a storage medium.

[0004] According to an aspect of the present disclosure, a video author identification method is provided, comprising:

[0005] According to the classification result of the cover image of the multiple video works of the video author, a first image feature of the multiple video works is determined;

[0006] According to the text feature and the first image feature of the multiple video works, a first video work of a target type is determined from the multiple video works; wherein the text feature is obtained according to the label information of the video work;

[0007] According to the proportion of the first video work in the multiple video works, a first work feature of the video author is determined;

[0008] According to the proportion of the second video work containing the target attribute in the multiple video works, a second work feature of the video author is determined; and

[0009] According to the first work feature and the second work feature, an author type label of the video author is determined.

[0010] According to another aspect of the present disclosure, a video resource recommendation method is provided, comprising:

[0011] According to the author type label of the video author, a target video author is determined; wherein the author type label of the video author is obtained according to the above-mentioned video author identification method;

[0012] From the multiple video works of the target video author, a third video work is screened to form a video resource pool;

[0013] In response to a user request, resources are recalled from the video resource pool to obtain a recall result; and

[0014] According to the recall result, a video resource to be recommended is obtained.

[0015] According to another aspect of the present disclosure, there is provided an identification apparatus of a video author, comprising:

[0016] a first determination module configured to determine first image features of a plurality of video works of the video author according to classification results of cover images of the plurality of video works;

[0017] a first classification module configured to determine first video works of a target type from the plurality of video works according to text features and the first image features of the plurality of video works; wherein the text features are obtained according to label information of the video works;

[0018] a second determination module configured to determine first work features of the video author according to a proportion of the first video works in the plurality of video works;

[0019] a third determination module configured to determine second work features of the video author according to a proportion of second video works containing target attributes in the plurality of video works; and

[0020] a second classification module configured to determine an author type label of the video author according to the first work features and the second work features.

[0021] According to another aspect of the present disclosure, there is provided a recommendation apparatus of video resources, comprising:

[0022] an author determination module configured to determine a target video author according to an author type label of a video author; wherein the author type label of the video author is obtained by the identification apparatus of the video author;

[0023] a screening module configured to screen third video works from a plurality of video works of the target video author to form a video resource pool;

[0024] a recall module configured to perform resource recall from the video resource pool to obtain a recall result in response to a user request; and

[0025] a recommendation module configured to obtain video resources to be recommended according to the recall result.

[0026] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0027] at least one processor; and

[0028] a memory communicatively connected to the at least one processor; wherein

[0029] the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any of the embodiments of the present disclosure.

[0030] According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method according to any of the embodiments of the present disclosure.

[0031] According to another aspect of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements the method according to any of the embodiments of the present disclosure.

[0032] The present disclosure mines video resources by accurately identifying video authors, thereby improving the recommendation effect of video resources.

[0033] It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0034] The accompanying drawings are used to better understand the present scheme and do not limit the present disclosure. Among them:

[0035] Figure 1 is a flowchart of a video author identification method according to an embodiment of the present disclosure;

[0036] Figure 2 is a flowchart of a beauty class author mining process according to another embodiment of the present disclosure;

[0037] Figure 3 is a flowchart of a video resource recommendation method according to an embodiment of the present disclosure;

[0038] Figure 4 is a flowchart of a beauty class video recommendation method according to another embodiment of the present disclosure;

[0039] Figure 5 is a flowchart of a video author identification device according to an embodiment of the present disclosure;

[0040] Figure 6 is a flowchart of a video resource recommendation device according to an embodiment of the present disclosure;

[0041] Figure 7 is a block diagram of an electronic device for implementing an embodiment of the present disclosure. DETAILED DESCRIPTION

[0042] Exemplary embodiments of the present disclosure are described herein with reference to the accompanying drawings, which are presented for the purpose of illustration and description. They are not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. Accordingly, persons of ordinary skill in the art will appreciate that many modifications and variations are possible in light of the teachings herein. Also, descriptions and details of well-known functions and structures have been omitted for the sake of clarity and conciseness.

[0043] In the related art, video content can be classified into categories (also referred to as vertical categories) such as movies, food, technology, automobiles, games, parenting, and outdoor activities. Each category can be further divided into more sub-categories. The classification method can also have other dimensions. For example, knowledge videos can be understood as popular science and explanation of any knowledge in the form of videos, which can cover vertical categories such as technology, automobiles, and parenting. It can be popular science of knowledge in the field of technology, popular science of knowledge in the field of animals and plants, or popular science of knowledge in the field of parenting education. For another example, appearance videos can be understood as videos with pleasing appearance and good appearance. It usually contains elements such as "high appearance" and "good", which can cover vertical categories such as emotion, life, food, automobiles, humor, and people. It is difficult to filter such videos based on conventional category classification.

[0044] Currently, there are mainly two schemes for mining such resources: one is based on manual annotation, and the other is based on label matching. The second scheme manually selects a number of labels related to the target type, and then uses the labels to define resources of the target type. The above methods cannot solve the problems of excessive resource consumption and high error rate.

[0045] To at least partially solve one or more of the above problems and other potential problems, embodiments of the present disclosure provide a video author identification method and a video resource recommendation method. The technical solutions of the embodiments of the present disclosure can improve the accuracy of video author identification, facilitate the mining of video resources of a target type from the perspective of a video author, and improve the recommendation effect of video resources.

[0046] Figure 1 A flowchart of a video author identification method according to an embodiment of the present disclosure is shown in FIG. 1. As shown in FIG. 1, the method includes the following steps: Figure 1

[0047] S101, determining first image features of a plurality of video works according to classification results of cover image of the plurality of video works.

[0048] ​In the embodiments of the present disclosure, the cover image of the video work can be the first frame of the video work, can be an image specified by the video producer through a video editing software, or can be a thumbnail of the video work.

[0049] The cover image is classified, and the classification result can be a type label, such as a knowledge type, a beauty type, or other types. The cover image of a knowledge type video usually contains obvious text titles and graphic elements, and the text content can contain formulas, names of principles, etc. The cover image of a beauty type video usually contains exquisite clothes, delicate makeup, and beautiful characters. Through artificial or classification models, the classification of the cover image can be realized based on the characteristic elements in the cover image.

[0050] The first image feature can be obtained according to the classification result of the cover image.

[0051] In S102, a first video work of a target type is determined from a plurality of video works according to text features and first image features of the plurality of video works. The text features are obtained according to label information of the video works.

[0052] A video work usually contains text information such as a title, a brief introduction, a category, and a label. The category can refer to a type obtained based on a certain classification method, such as film and television, finance, technology, automobile, and digital. For short video works, the vertical categories can include hot spots, games, entertainment, sports, food, fashion, and second dimension. Each video work usually belongs to one vertical category. There can be multiple dimensions to describe or define the type of a video work, which is not limited here.

[0053] The label can be understood as a keyword describing the content of a video work, including but not limited to words describing video content, script content, theme, characters in the video content, occasion, photographer, shooting location, shooting form, related topics, etc. The title of the video work, the picture resolution, the video author, the video duration, the file size, and the file format can also be understood as a label describing the video work. The label can be selected or input by the video author, or generated after identifying the video content. By encoding various text information of the video author, the text features of the video work can be obtained.

[0054] Taking the vertical category and the label of the video work as the text features and taking the classification result of the cover image as the first image feature, a pre-trained classification model or other existing classification model can be used to classify the video work according to the text features and the first image features.

[0055] The target type can be knowledge-based, appearance-based, or other preset types. In one example, manually labeled videos of the target type can be used as positive samples, and randomly sampled videos of other non-target types can be used as negative samples to train a logistic regression (LR) binary classification model. The trained binary classification model can be used to identify whether a target video belongs to the target type. When training the model, the video can be divided into multiple sample datasets according to the vertical category of the video. For example, appearance-based videos under the film and television vertical category and non-appearance-based videos under the same vertical category can be used as the same sample dataset, and appearance-based videos under the food vertical category and non-appearance-based videos under the same vertical category can be used as the same sample dataset.

[0056] S103. Determine the characteristics of the video author's first work based on the proportion of the first video work among multiple video works.

[0057] In one example, if 60% of all the video works of a certain video creator are the first video works of the target type, then this percentage is used to determine the first work characteristic of the video creator.

[0058] S104. Determine the characteristics of the second work of the video author based on the proportion of the second video works containing the target attribute among multiple video works.

[0059] Target attributes can be one or more of vertical categories and tags. Specifically, it can involve selecting vertical categories or tags highly relevant to video works of the target type, and counting the number of second video works belonging to that vertical category or containing that tag to obtain the proportion of such works. For example, for knowledge-based videos, tags related to knowledge (knowledge tags) can be pre-selected from various tags as target tags, such as tags containing words like "principle," "formula," "decipher," "reveal," "encyclopedia," and "explanation," which are target attributes. For videos related to appearance, tags related to appearance (appearance tags) can be pre-selected from various tags as target tags, such as tags containing words like "pretty," "beautiful," "high-value," "beautiful woman," "graceful lady," and "stunningly beautiful." Knowledge tags or appearance tags can be obtained through statistical analysis of the tags contained in video samples labeled as the target type during manual annotation of video samples; this will not be exhaustive. The proportion of second video works containing the target tag among multiple video works is counted, and this proportion is used as a characteristic of the video author's second work.

[0060] S105. Based on the characteristics of the first work and the characteristics of the second work, determine the author type label of the video author.

[0061] According to the first work feature and the second work feature of the video author, the author type of the video author can be classified by using a pre-trained classification model to obtain an author type label. Similar to the training method of the LR model described above, another LR model can be trained using video authors of the target type annotated by humans as positive samples and video authors randomly sampled from other categories as negative samples. The trained LR model can be used to identify whether the video author is of the target type. It should be noted that the classification model and the model training method used in the embodiments of the present disclosure are only examples, and other models mastered by those skilled in the art can also be used to achieve the classification purpose through conventional training methods, and the type and structure of the model used are not limited herein.

[0062] The author type label is used to represent whether the video author is a high verticality author of the target type video. High verticality means that the video author focuses on producing videos of the target type. The video quality of a low verticality video author is usually low because he / she does not focus on producing videos of the same type. On the other hand, low verticality may also mean that the video author mainly produces works of others.

[0063] According to the scheme of the embodiments of the present disclosure, by classifying the video works of the video author, the type and content distribution of the video works can be more accurately counted, and whether the video author focuses on producing videos of the same type can be determined, and high verticality video authors who focus on producing videos of the same type can be accurately mined.

[0064] In a possible implementation, step S101 determines the first image feature of the plurality of video works according to the classification result of the cover image of the plurality of video works of the video author, and further includes the steps of:

[0065] Obtaining the cover image of the plurality of video works of the video author.

[0066] Determining the classification result of the cover image by using a pre-trained first classification model.

[0067] Determining the first image feature of the plurality of video works according to the classification result of the cover image.

[0068] In the embodiments of the present disclosure, the first classification model can use a 50-layer residual network (RestNet50) model. The RestNet50 classification model can be used to classify the cover image to obtain the corresponding classification label. In an example, the classification result can be a binary classification label of whether it belongs to the target type, for example, two labels respectively represent whether it belongs to the knowledge class cover or does not belong to the knowledge class cover, or whether it belongs to the color value class cover or does not belong to the color value class cover.

[0069] Generally, the cover image of a video work can reflect the video content and the video type to a great extent, and therefore, the classification result of the cover image is highly relevant to the type of the video work, that is, the classification label of the cover image as a feature of the video work can improve the accuracy of classifying the video work.

[0070] According to the scheme of the embodiments of the present disclosure, the cover image of the video work is identified and classified using the classification model, and the classification result is taken as the first image feature of the video work, which is helpful for classifying the video work.

[0071] In a possible implementation, the step S104 determines the second work feature of the video author according to the proportion of the second video works containing the target attribute in the plurality of video works, and further includes the steps of:

[0072] S1041, for the attribute information of at least one dimension, determine the number of second video works containing the target attribute. Wherein, the target attribute is one of the attribute information.

[0073] S1042, according to the number of second video works and the number of plurality of video works, determine the proportion of second video works.

[0074] S1043, according to the proportion corresponding to the attribute information of at least one dimension, determine the second work feature of the video author.

[0075] In the embodiments of the present disclosure, the target attribute can be one or more of the vertical class and the label, that is, the vertical class and the plurality of labels can be taken as different dimensions. The proportion of the second video works containing the target number in all video works is counted, and the statistical feature obtained therefrom is taken as the second work feature.

[0076] According to the scheme of the embodiments of the present disclosure, part of the attribute information of the video work is taken as the statistical object, the proportion of the video works containing the target attribute in all video works is determined, and the proportion is taken as the second work feature, which is helpful for determining the verticality of the video author to a certain type of video.

[0077] In a possible implementation, the step S1041 determines the number of second video works containing the target attribute for the attribute information of at least one dimension, including at least one of the following ways:

[0078] For the attribute information of the video category dimension, determine the number of second video works containing the target video category.

[0079] For the attribute information of the description label dimension, determine the number of second video works containing the target label.

[0080] For the attribute information of the cover image type dimension, according to the classification result of the cover image, the number of second video works of the target type of the cover image is determined.

[0081] In an example of the embodiments of the present disclosure, for the attribute information of the video category dimension, when the target attribute is the film and television vertical category, the second video works are all video works of the film and television vertical category published by the video author. In another example, as shown in Figure 2 For the attribute information of the description label dimension, when the target attribute is the beauty label, the second video works are all video works containing the beauty label published by the video author, so that the proportion of the video works containing the beauty label can be obtained. The classification result of the cover image can also be a statistical object, and for the attribute information of the cover image type dimension, the number of second video works containing the cover image of the target type can be counted, so that the proportion of the video works with the beauty cover image can be obtained.

[0082] According to the scheme of the embodiments of the present disclosure, the multiple attribute information of the video works is screened, and the number of second video works containing the target attribute is determined, so that the verticality of the video author is determined.

[0083] In a possible implementation, for the attribute information of the video category dimension, the number of second video works containing the target video category is determined, specifically including:

[0084] For the attribute information of the video category dimension, according to the number of video works contained by each video category, the sorting result of the video category is obtained.

[0085] For the target video category in the top N of the sorting result, the number of corresponding second video works is determined, wherein N is a positive integer.

[0086] In the embodiments of the present disclosure, all video categories involved in the video works are determined first, and then the number of video works of each video category is counted. In an example, as shown in Figure 2 The video category with the largest number (i.e., N is 1) can be used as the main vertical category of the video works, and the proportion of the video works of the main vertical category can reflect the verticality of the video author to a great extent. At the same time, the top 3 (i.e., N is 3) video categories in the sorting result of the video category can also be used as the statistical object, and the proportion of the second video works of the top 3 vertical categories can also reflect the verticality of the video works to a certain extent.

[0087] According to the scheme of the embodiments of the present disclosure, the vertical category to which the video works belong is finely counted, and the verticality of the video author is measured from the dimension of the vertical category.

[0088] In a possible implementation, the method for identifying the video author according to the embodiments of the present disclosure further includes the steps of:

[0089] Collect a plurality of key frame images from a plurality of video works.

[0090] Determine a second image feature according to the plurality of key frame images.

[0091] Determine a genre classification label of the plurality of video works according to the second image feature.

[0092] Determine a third work feature of a video author according to the genre classification label of the plurality of video works.

[0093] Determine an author type label of the video author according to the first work feature and the second work feature, comprising:

[0094] Determine the author type label of the video author according to the first work feature, the second work feature, and the third work feature.

[0095] In the embodiments of the present disclosure, a plurality of key frame images are obtained by cross-step sampling, and then the plurality of key frame images are input into a classification model pre-trained based on a temporal shift module (TSM), so as to obtain a genre classification label. The TSM model is a video classification model constructed by inserting a temporal shift module into a ResNet network. The model improves the video understanding ability in the time dimension by introducing context interaction in the temporal dimension in the feature map. By extracting key frames, the video content can be identified and classified, so as to determine the genre of the video content. The purpose of determining the genre is that the knowledge video is not a random guess or a fabrication of the cause of the unsolved mystery. The appearance value video is not a narrow shooting of people or a show of body shape, and is not a vulgar content involving sexual suggestion or nudity. Therefore, the video works with low genre content need to be identified, so as to determine the number of video works involving low genre content in all video works of the video author, as a third work feature, so as to reduce the recommendation of related content and the video author.

[0096] It should be noted that there may be some false positives and false negatives in identifying the high and low genres of the video content by the classification model. The videos with high play counts can be manually reviewed to ensure accuracy.

[0097] According to the scheme of the embodiments of the present disclosure, the content involving low genre in the video works can be identified, so as to more accurately determine the creation direction of the video author.

[0098] In a possible implementation, after determining the video author of a plurality of target types, the GCF (Global Chess Federation, global comparison framework) graph model can be used for diffusion to find more authors of the same type.

[0099] In a possible implementation, the above-mentioned video author identification method can be periodically performed to monitor the video author's posts and prevent the video author's video works from being deteriorated, and the deteriorated author can be quickly released through a blacklist.

[0100] Figure 3 A flowchart of a video resource recommendation method provided by an embodiment of the present disclosure is shown in FIG. 3. As shown in FIG. 3, the method comprises at least the following steps: Figure 3

[0101] S301, determining a target video author according to an author type label of the video author, wherein the author type label of the video author is obtained according to the video author identification method of any of the above-mentioned embodiments.

[0102] S302, screening third video works from a plurality of video works of the target video author to form a video resource pool.

[0103] S303, in response to a user request, recalling resources from the video resource pool to obtain a recall result.

[0104] S304, obtaining video resources to be recommended according to the recall result.

[0105] In the embodiment of the present disclosure, when mining target type videos for recommendation, taking value class videos as an example, a value class author with high straightness can be selected according to the author type label. Then, part of the newly published video works of the value class author are screened to obtain third video works, so as to form a high-quality video resource pool through the new video works of a plurality of value class authors with high straightness.

[0106] When a user watches a video on a video website, platform or software, various recall methods can be used to recall resources from the video resource pool to obtain personalized recall results matched with the user portrait of the user. The user portrait can be obtained according to the behavior log formed by user feedback. The recall channel (recall method) includes but is not limited to:

[0107] Cold start: recalling core user group favorite but less distributed long-tail resources.

[0108] Popular: recalling popular resources that the core user group likes.

[0109] CB (Content-Based Filtering, Content-Based Filtering): recalling related video resources based on user interest points.

[0110] ​ICF (Item-Item Collaborative Filtering): based on the satisfied clicks of the video resources, recall related video resources.

[0111] UCF (User-User Collaborative Filtering): recall the video resources published by similar authors of the attention author.

[0112] LookLike (Look-Alike): recall similar video vectors through UCF user vectors.

[0113] Attention author: recall the video resources published by the video authors that the user follows.

[0114] According to the scheme of the embodiment of the present disclosure, based on the text semantics, cover image, key frame and other multi-modal signals of the existing video works, the target type author is mined. After the video author is determined as a high straightness author of a certain type, the newly published video works of the author can be simply screened as the video of the type, without the need to accurately identify and classify each video work before determining the video type, reducing the amount of calculation, and also having the ability to discover fresh and high-quality videos.

[0115] In a possible implementation, step S304 obtains the video resources to be recommended according to the recall results, further comprising steps of:

[0116] S3041, sorting the recall results.

[0117] S3042, obtaining the video resources to be recommended according to the sorted results.

[0118] In the embodiment of the present disclosure, as shown in Figure 4 After the recall stage, the multiple recall results obtained by different recall channels can be sorted. The sorting can include a coarse sorting stage and a fine sorting stage. The coarse sorting stage can perform grouping and forwarding. Grouping and forwarding refers to the process of placing resources with good posteriori conversion effects in the front.

[0119] According to the scheme of the embodiment of the present disclosure, through sorting, the video resources that the user is most likely to be interested in can be preferentially recommended to the user.

[0120] In a possible implementation, S3041 sorts the recall results, further comprising:

[0121] Determine the estimated click rate, estimated completion rate and video duration of the multiple video resources in the recall results.

[0122] According to the estimated click rate, the estimated completion rate, and the video duration, an estimated score of the plurality of video resources is determined.

[0123] The plurality of video resources in the recall result are sorted in descending order of the estimated score.

[0124] In the embodiments of the present disclosure, the following calculation formula can be used as the sorting strategy in the fine sorting stage:

[0125]

[0126] Wherein, Q is the estimated score, ctr is the estimated click rate, fr is the estimated completion rate, duration is the physical duration of the video resource, CP, FP, and DP are corresponding fusion hyperparameters, which can also be understood as weight values.

[0127] According to the scheme of the embodiments of the present disclosure, the estimated score can be determined by the estimated click rate, the completion rate, and the video duration, and the video resource with a high estimated score is preferentially recommended, which can improve the user's viewing time.

[0128] In a possible implementation, the video resource to be recommended is obtained according to the sorting result, including:

[0129] The sorted result is re-sorted according to a business exploration strategy.

[0130] The video resource to be recommended is obtained according to the re-sorted result.

[0131] In the embodiments of the present disclosure, the business exploration strategy includes: ranking and bucketing according to the estimated score, combining the user's high interest and attention author card control to estimate the score, selecting exploration resources, and making exploration with in-and-out constraints. The business exploration strategy is used to preferentially recommend the determined video resource, also known as exploration strong insertion, such as the rearrangement stage (appearance value exploration strong insertion) shown in FIG. 8. Figure 4

[0132] Ranking and bucketing according to the estimated score: first, ranking according to the estimated score of each video resource, and then dividing the video resources ranked at the top in different vertical categories into a bucket. This bucket can be regarded as a pre-ranking set.

[0133] Combining the user's high interest: next, considering the user's interest factor, combining the historical behavior (such as consumption history), user characteristics, and interest preferences in the user portrait with the information of the video resource to further optimize the sorting. For example, if a user has shown strong interest in a certain field or theme in the past period of time, the video resource related to this field can be preferentially considered.

[0134] ​Pay attention to the author's card control estimated score: At the same time, the source and author information of the video resource need to be paid attention to to ensure that high-quality video resources are displayed first. This can be realized by card control estimated score, that is, only video resources with estimated score reaching a certain threshold are selected for display.

[0135] Select exploration resources: While maintaining the existing ranking, an exploration mechanism needs to be introduced to discover new high-quality video resources. A certain proportion of video resources from the lower end of the ranking can be selected for exploration, thereby increasing the discovery ability of the system.

[0136] When exploring, entry and exit constraints need to be considered. The entry constraint means that new video resources can only be introduced into the display list under certain conditions, such as when the estimated score of existing video resources is below a certain threshold. The exit constraint means that existing video resources need to be removed from the display list when certain conditions are met, such as when their estimated score exceeds a certain threshold.

[0137] Combine the strength of the exit signal: The strength of the exit signal refers to the feedback information obtained by the video during the display process, such as click rate, viewing time, etc. According to the strength of the exit signal, the quality of the video resource and the degree of user interest in the video can be judged. If the exit signal is weak, it means that the quality of the video resource is not high or the user's interest in the video is low, at which time the exit operation can be considered.

[0138] According to the strength of the exit signal and different exit constraint conditions, different exit strategies can be adopted. For example, according to the user's interest point, the interest point exit can be performed, that is, when the user's interest point changes, the video resource related to the interest point is removed from the display list. According to the user's session, the Session exit can be performed, that is, when the user's session ends, the video resource displayed during the session is removed from the display list. According to the time, the 24H exit can be performed, that is, when a video resource has not been browsed for more than 24 hours, it is removed from the display list. Permanent exit can also be performed according to other conditions, such as when the video resource violates relevant regulations or the quality is seriously substandard.

[0139] According to the scheme of the embodiment of the present disclosure, based on the business exploration strategy and the estimated score, the user's interest can be explored, which helps to improve the user's retention rate.

[0140] Figure 5 The structure diagram of the video author identification device provided by an embodiment of the present disclosure is shown in FIG. 1. Figure 5 As shown in the figure, the device comprises:

[0141] The first determination module 501 is configured to determine the first image feature of the plurality of video works according to the classification result of the cover image of the plurality of video works of the video author.

[0142] The first classification module 502 is configured to determine a first video work of a target type from the plurality of video works according to text features and first image features of the plurality of video works. The text features are obtained according to label information of the video works.

[0143] The second determination module 503 is configured to determine a first work feature of the video author according to a proportion of the first video work in the plurality of video works.

[0144] The third determination module 504 is configured to determine a second work feature of the video author according to a proportion of a second video work containing a target attribute in the plurality of video works, and

[0145] The second classification module 505 is configured to determine an author type label of the video author according to the first work feature and the second work feature.

[0146] In a possible implementation, the first determination module 501 is configured to:

[0147] Obtain cover images of a plurality of video works of the video author.

[0148] Determine a classification result of the cover images by using a pre-trained first classification model.

[0149] Determine first image features of the plurality of video works according to the classification result of the cover images.

[0150] In a possible implementation, the third determination module 504 includes:

[0151] The quantity determination sub-module is configured to determine a quantity of the second video works containing the target attribute for attribute information of at least one dimension. The target attribute is one of the attribute information.

[0152] The proportion determination sub-module is configured to determine the proportion of the second video works according to the quantity of the second video works and a quantity of the plurality of video works.

[0153] The feature determination sub-module is configured to determine the second work feature of the video author according to the proportion corresponding to the attribute information of at least one dimension.

[0154] In a possible implementation, the quantity determination sub-module is configured to:

[0155] Determine the quantity of the second video works containing a target video category for attribute information of a video category dimension.

[0156] Determine the quantity of the second video works containing a target label for attribute information of a description label dimension.

[0157] For the attribute information of the cover image type dimension, according to the classification result of the cover image, the number of second video works of the target type of the cover image is determined.

[0158] In a possible implementation, the number determination sub-module is configured to:

[0159] For the attribute information of the video category dimension, according to the number of video works contained in each video category, a ranking result of the video category is obtained.

[0160] For the target video category in the top N positions of the ranking result, the number of corresponding second video works is determined, where N is a positive integer.

[0161] In a possible implementation, the device further includes a fourth determination module configured to:

[0162] A plurality of key frame images are collected from a plurality of video works.

[0163] According to the plurality of key frame images, a second image feature is determined.

[0164] According to the second image feature, a style classification label of the plurality of video works is determined.

[0165] According to the style classification label of the plurality of video works, a third work feature of the video author is determined.

[0166] The second classification module is further configured to:

[0167] According to the first work feature, the second work feature, and the third work feature, an author type label of the video author is determined.

[0168] The specific functions and examples of the modules and sub-modules of the device of the embodiments of the present disclosure are described above in the related description of the corresponding steps in the method embodiments, and will not be described here.

[0169] Figure 6 The structural schematic diagram of the video resource recommendation device provided by an embodiment of the present disclosure is shown in FIG. 6. Figure 6 As shown in the figure, the device includes:

[0170] The author determination module 601 is configured to determine a target video author according to an author type label of the video author, where the author type label of the video author is obtained according to the device provided by any embodiment of the present disclosure.

[0171] The screening module 602 is configured to screen third video works from a plurality of video works of the target video author to form a video resource pool.

[0172] The recall module 603 is configured to perform resource recall from the video resource pool in response to a user request to obtain a recall result. And

[0173] The recommendation module 604 is configured to obtain the video resource to be recommended according to the recall result.

[0174] In a possible implementation, the recommendation module 604 includes:

[0175] The sorting sub-module is configured to sort the recall result.

[0176] The resource determination sub-module is configured to obtain the video resource to be recommended according to the sorted result.

[0177] In a possible implementation, the sorting sub-module is configured to:

[0178] Determine the estimated click rate, the estimated completion rate, and the video duration of the plurality of video resources in the recall result.

[0179] Determine the estimated score of the plurality of video resources according to the estimated click rate, the estimated completion rate, and the video duration.

[0180] Sort the plurality of video resources in the recall result in descending order of the estimated score.

[0181] In a possible implementation, the sorting sub-module is further configured to:

[0182] Resort the sorted result according to a business exploration strategy.

[0183] Obtain the video resource to be recommended according to the resorted result.

[0184] The specific functions and examples of the modules and sub-modules of the apparatuses in the embodiments of the present disclosure are described above in the corresponding steps of the method embodiments, which will not be described here.

[0185] In the technical solutions of the present disclosure, the acquisition, storage, and application of user personal information comply with relevant laws and regulations and do not violate public order and good customs.

[0186] According to the embodiments of the present disclosure, the present disclosure further provides an electronic device, a readable storage medium, and a computer program product.

[0187] Figure 7A schematic block diagram of an example electronic device 700 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit the implementations of the present disclosure described and / or claimed in this document.

[0188] As shown in Figure 7 The electronic device 700 includes a computing unit 701 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 702 or a computer program loaded into a random access memory (RAM) 703 from a storage unit 708. Various programs and data required for the operation of the electronic device 700 can also be stored in the RAM 703. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other through a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0189] Various components in the electronic device 700 are connected to the I / O interface 705, including an input unit 706, such as a keyboard, a mouse, and the like; an output unit 707, such as various types of displays, speakers, and the like; a storage unit 708, such as a magnetic disk, a magneto-optical disk, and the like; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit 709 allows the electronic device 700 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0190] The computing unit 701 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs various methods and processes described above, such as the identification method of a video author and the recommendation method of a video resource. For example, in some embodiments, the identification method of a video author and the recommendation method of a video resource can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded onto the RAM 703 and executed by the computing unit 701, one or more steps of the identification method of a video author and the recommendation method of a video resource described above can be performed. Alternatively, in other embodiments, the computing unit 701 can be configured to perform the identification method of a video author and the recommendation method of a video resource by any other suitable means, such as by means of firmware.

[0191] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0192] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces a means for implementing the functions / acts specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0193] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0194] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0195] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0196] The computer system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0197] It should be understood that the various forms of flow shown above can be re-ordered, added to, or have steps deleted, using the steps described above. For example, the steps described in the present disclosure can be performed in parallel, in series, or in a different order, as long as the desired results of the technology disclosed in the present disclosure can be achieved, which is not limited herein.

[0198] The specific implementation described above does not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for identifying video authors, comprising: Based on the classification results of the cover images of multiple video works by the video author, the first image features of the multiple video works are determined; Based on the text features of the plurality of video works and the first image features, a first video work of the target type is determined from the plurality of video works; wherein, the text features are obtained based on the tag information of the video works; The first work characteristics of the video author are determined based on the proportion of the first video work among the multiple video works; The characteristics of the second works of the video author are determined based on the proportion of second video works containing the target attribute among the multiple video works; Multiple keyframe images are captured from the aforementioned multiple video works; Based on the multiple keyframe images, determine the second image features; Based on the second image features, determine the style classification tags of the multiple video works; Based on the style classification tags of the multiple video works, determine the third work characteristics of the video author; and Based on the first work characteristics, the second work characteristics, and the third work characteristics, the author type tag of the video author is determined.

2. The method according to claim 1, wherein, Based on the classification results of the cover images of multiple video works by the video creator, the first image features of the multiple video works are determined, including: Retrieve cover images of multiple video works by a video creator; The classification result of the cover image is determined using a pre-trained first classification model; Based on the classification results of the cover images, the first image features of the multiple video works are determined.

3. The method according to claim 1, wherein, The characteristics of the second works of the video author are determined based on the proportion of second video works containing the target attribute among the multiple video works, including: For attribute information in at least one dimension, determine the number of second video works containing a target attribute; wherein the target attribute is one of the attributes in the attribute information; The proportion of the second video works is determined based on the number of the second video works and the number of the plurality of video works; The second work characteristics of the video author are determined based on the proportion of attribute information corresponding to at least one dimension.

4. The method according to claim 3, wherein, For attribute information in at least one dimension, determine the number of second video works containing the target attribute, including at least one of the following methods: Based on the attribute information of the video category dimension, determine the number of second video works that contain the target video category; Based on the attribute information describing the tag dimension, determine the number of second video works containing the target tag; Based on the attribute information of the cover image type dimension, and according to the classification results of the cover image, determine the number of second video works whose cover image is of the target type.

5. The method according to claim 4, wherein, Based on the attribute information of the video category dimension, determine the number of second video works that contain the target video category, including: Based on the attribute information of the video category dimension, the ranking results of the video categories are obtained according to the number of video works contained in each video category; For the target video categories that rank in the top N, determine the number of corresponding second video works, where N is a positive integer.

6. A method for recommending video resources, comprising: The target video author is determined based on the author type tag of the video author; wherein the author type tag of the video author is obtained by the identification method according to any one of claims 1 to 5; A third video work is selected from multiple video works by the target video author to form a video resource pool; In response to a user request, resources are retrieved from the video resource pool to obtain a retrieval result; and Based on the recall results, video resources to be recommended are obtained.

7. The method according to claim 6, wherein, Based on the recall results, the video resources to be recommended are obtained, including: The recall results are sorted. Based on the sorting results, the video resources to be recommended are obtained.

8. The method according to claim 7, wherein, The recall results are sorted, including: Determine the estimated click-through rate, estimated completion rate, and video duration of multiple video resources in the recall results; Based on the estimated click-through rate, the estimated completion rate, and the video duration, the estimated scores of the multiple video resources are determined. The video resources in the recall results are sorted from highest to lowest according to the estimated scores.

9. The method according to claim 7 or 8, wherein, Based on the sorting results, the video resources to be recommended are obtained, including: Based on the business exploration strategy, the sorting results are reordered; Based on the results of the reordering, the video resources to be recommended are obtained.

10. A device for identifying video authors, comprising: The first determining module is used to determine the first image features of the multiple video works based on the classification results of the cover images of the multiple video works of the video author; The first classification module is used to determine a first video work of a target type from the plurality of video works based on the text features and the first image features; wherein the text features are obtained based on the tag information of the video works; The second determining module is used to determine the first work characteristics of the video author based on the proportion of the first video work among the multiple video works; The third determining module is used to determine the characteristics of the second work of the video author based on the proportion of second video works containing the target attribute among the plurality of video works; and The second classification module is used to determine the author type label of the video author based on the features of the first work and the features of the second work; The fourth determining module is used for: Multiple keyframe images are captured from the aforementioned multiple video works; Based on the multiple keyframe images, determine the second image features; Based on the second image features, determine the style classification tags of the multiple video works; Based on the style classification tags of the multiple video works, the third work characteristics of the video author are determined; The second classification module is also used for: Based on the first work characteristics, the second work characteristics, and the third work characteristics, the author type tag of the video author is determined.

11. The apparatus according to claim 10, wherein, The first determining module is used for: Retrieve cover images of multiple video works by a video creator; The classification result of the cover image is determined using a pre-trained first classification model; Based on the classification results of the cover images, the first image features of the multiple video works are determined.

12. The apparatus according to claim 10, wherein, The third determining module includes: The quantity determination submodule is used to determine the number of second video works containing a target attribute based on attribute information of at least one dimension; wherein, the target attribute is one of the attributes of the attribute information; The percentage determination submodule is used to determine the percentage of the second video work based on the number of the second video work and the number of the plurality of video works; The feature determination submodule is used to determine the second work features of the video author based on the proportion of attribute information corresponding to the at least one dimension.

13. The apparatus according to claim 12, wherein, The quantity determination submodule is used for: Based on the attribute information of the video category dimension, determine the number of second video works that contain the target video category; Based on the attribute information describing the tag dimension, determine the number of second video works containing the target tag; Based on the attribute information of the cover image type dimension, and according to the classification results of the cover image, determine the number of second video works whose cover image is of the target type.

14. The apparatus according to claim 13, wherein, The quantity determination submodule is used for: Based on the attribute information of the video category dimension, the ranking results of the video categories are obtained according to the number of video works contained in each video category; For the target video categories that rank in the top N, determine the number of corresponding second video works, where N is a positive integer.

15. A video resource recommendation device, comprising: An author determination module is configured to determine a target video author based on an author type tag of the video author; wherein the author type tag of the video author is obtained by the apparatus according to any one of claims 10 to 14; The filtering module is used to filter out a third video work from multiple video works of the target video author to form a video resource pool; The recall module is used to respond to a user request by recalling resources from the video resource pool to obtain a recall result; and The recommendation module is used to obtain video resources to be recommended based on the recall results.

16. The apparatus according to claim 15, wherein, The recommendation module includes: The sorting submodule is used to sort the recall results; The resource determination submodule is used to obtain the video resources to be recommended based on the sorting results.

17. The apparatus according to claim 16, wherein, The sorting submodule is used for: Determine the estimated click-through rate, estimated completion rate, and video duration of multiple video resources in the recall results; Based on the estimated click-through rate, the estimated completion rate, and the video duration, the estimated scores of the multiple video resources are determined. The video resources in the recall results are sorted from highest to lowest according to the estimated scores.

18. The apparatus according to claim 16 or 17, wherein, The sorting submodule is also used for: Based on the business exploration strategy, the sorting results are reordered; Based on the results of the reordering, the video resources to be recommended are obtained.

19. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-9.

20. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-9.

21. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-9.

Images